You are on page 1of 14

This article has been accepted for publication in a future issue of this journal, but has not been

fully edited. Content may change prior to final publication.


IEEE TRANSACTIONS ON VISUALIZATION AND COMPUTER GRAPHICS, MANUSCRIPT ID 1

A Novel Visualization Technique for Electric Power Grid Analytics


Pak Chung Wong, Kevin Schneider, Senior Member, IEEE , Patrick Mackey, Harlan Foote, George Chin Jr., Ross Guttromson, and Jim Thomas, Senior Member, IEEE Computer Society
Abstract The application of information visualization holds tremendous promise for the electric power industry, but its potential has so far not been sufficiently exploited by the visualization community. Prior work on visualizing electric power systems has been limited to depicting raw or processed information on top of a geographic layout. Little effort has been devoted to visualizing the physics of the power grids, which ultimately determines the condition and stability of the electricity infrastructure. Based on this assessment, we developed a novel visualization system prototype, GreenGrid, to explore the planning and monitoring of the North American Electricity Infrastructure. The paper discusses the rationale underlying the GreenGrid design, describes its implementation and performance details, and assesses its strengths and weaknesses against the current geographic-based power grid visualization. We also present a case study using GreenGrid to analyze the information collected moments before the last major electric blackout in the Western United States and Canada, and a usability study to evaluate the practical significance of our design in simulated real-life situations. Our result indicates that many of the disturbance characteristics can be readily identified with the proper form of visualization. Index TermsData and knowledge visualization, information visualization, visualization techniques and methodologies, applications.

E live in a society that is vitally dependent on a network infrastructure of natural, man-made, and human resources to functionfrom food to water supplies, from electric power to other fuel sources, and from communication and transportation to medical and emergency services. While these resources are seamlessly integrated into the fabric of our society, electric power has the highest network reachabilityand all the other network resources depend on it to operate. Losing electric power inevitably impairs the ability of the other resources to perform, which could cripple society if a widespread outage persisted for a prolonged period [10]. The vulnerability of our nations electric power systems is the focus of research for a group of scientists and engineers at the Pacific Northwest National Laboratory (PNNL). This paper reports an information visualization system prototype developed by the group, known as GreenGrid, that helps explore the planning and monitoring of the North American power grids. A power grid [11] is informally described as a network of electrical generators that supply power to end users via transmission and distribution systems. When there is a disruption of power from any single generator, or a small group of generators, the system responds to ensure that there is adequate generation to supply the electrical load plus system losses. A major challenge of GreenGrid is to help identify the

INTRODUCTION
vulnerabilities of the power grids at the first sign of trouble by detecting the onset of a network separation. Failing to quickly address the separation problem could result in a full-blown blackout in the power grids. The visualization of GreenGrid moves away from a traditional geographic information system (GIS) design approach and towards the understanding of the power grids physics in an exploration setting. The paper explains the rationale underlying the GreenGrid design, describes its implementation details, evaluates its strengths and weaknesses, and demonstrates its capability using real-life data collected moments before the last major blackout that occurred in the Western United States on August 19, 1996. The development of GreenGrid is an ongoing collaboration between the U.S. Department of Energy (DOE), the National Visualization and Analytics CenterTM (NVACTM) [20], the Electricity Infrastructure Operations Center (EIOC) [6] at PNNL, and partners from the electric power industry. EIOC is a fully capable grid control center that serves as a platform for grid research and development, testing, and training. Through EIOC and its industrial partners, we have access to real-time telemetered data from both the Western Electricity Coordinating Council (WECC) and its counterpart in the Eastern United States, the Eastern Interconnect [21].

RELATED WORK

P.C. Wong, K. Schneider, P. Mackey, H. Foote, G. Chin, R. Guttromson, and J. Thomas are with the Pacific Northwest National Laboratory, P.O. Box 999, Richland, WA 99352. E-mail: {pak.wong, kevin.schneider, patrick.mackey, harlan.foote, george.chin, ross.guttromson, jim.thomas}@ pnl.gov.
Manuscript received (insert date of submission if desired).

The development of GreenGrid involves a number of power grid visualization, graph drawing, and graph analytics topics. This section highlights some of the work previously presented in literature.

xxxx-xxxx/0x/$xx.00 200x IEEE

Digital Object Indentifier 10.1109/TVCG.2008.197

1077-2626/$25.00 2008 IEEE

This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication.
2 IEEE TRANSACTIONS ON JOURNAL NAME, MANUSCRIPT ID

2.1 Power Grid Visualization The application of visualization has long been underutilized by the electric power industry. AREVAs Energy Management System (EMS) [2] and PowerWorlds Simulator [22][23] represent two of the most widely used tools that provide visualization support within the industry. However, their main functionalities are mostly about depicting collected or processed information using color or icons on top of a geographic layout such as a geographic map. This presentation-oriented visualization approach, while popular, fails to take advantage of the analytical strengths naturally gained by the visualization itself. We will revisit this topic in Section 3. 2.2 Graph Layout and Visualization The analytical technique presented here is mainly a graphbased visualization. The graph drawing community has led the studies in many graph layout issues for decades. The two textbooks by Di Battista et al. [5] and Sugiyama [25] summarize most of the major graph drawing algorithms and their applications. The proceedings of the annual Graph Drawing Symposia [12], now in its fifteenth year, and the Journal of Graph Algorithms and Applications [16] provide a wealth of information on the cutting-edge technology. Visualizing graphs and hierarchies has been a major study topic within the data visualization community since its conception in the early 90s. The two textbooks by Card et al. [3] and Chen [4] cover much of the major research and applications surrounding graph and hierarchical visualization. The survey paper by Herman et al. [14] represents the most complete literature review up to 2000. The annual IEEE Symposium on Information Visualization [15] continues to produce new results on various topics of graph visualization. A major difference between the graph visualization and graph drawing communities is that the former almost always involves some sort of interaction, whereas the latter focuses heavily on algorithmic developments. The latest challenge, however, is to integrate the best of the two communities and form a new environment of graph analytics. 2.3 Very Large Graph Drawing The North American power grids contain well over 50,000 electrical buses. Drawing a graph of this size nicely on screen can be computationally expensive [5][25]. There is an ongoing community effort to speed up the drawing process by developing adaptive algorithms with complexity of O(n2) and below. Notable work in this area has recently been presented by Harel et al. [13] and Walshaw [26]. These two papers describe cutting-edge layout algorithms and provide resourceful ideas for further design improvements. Their references sections also present a wealth of information covering topics from graph partitioning to Laplacian Eigenvector computation. GreenGrids visualization design is based partly on Walshaws algorithm with multiple enhancements to accelerate the eigenvector computation.

2.4 Weighted Graph Drawing One approach to further improve the analytical capability of a graph layout is to include an extra data dimension along the graph link through the concept of a weighted graph. In other words, the length of a graph link (or the Euclidean distance between the two corresponding graph nodes) is approximately proportional to the weight that the link represents. While the concept of drawing a weighted graph is not covered in popular references such as [5][25], its design and implementation can be reasonably straightforward. Using the classic force-directed layout approach that relies on the notion of graph distance between pairs of nodes, one can adjust the attractive and repulsive forces based on the information represented by the graph links to draw the graph. Similar work has been reported recently by Kobourov et al. [7][8]. GreenGrid uses a weighted graph design to model the physics of a power grid. 2.5 Network-Based Cartogram Our work also shares a number of data and design challenges with the study of network-based cartograms. Both of them involve some sort of network graph alongside geographic data. For example, Ahmed and Miller [1] use the scaling approach to visualize the time-space information of a transportation network. While the fundamental visualization approach of [1] is very different from the one in GreenGrid, they both apply the distortion concept to present geographic-related information without relying on the use of icons or glyphs.

GREENGRID

GreenGrid is built upon a collection of graph analytics and mining tools developed independently under the Have Green framework [31][32]. The discussion focuses on one of the newly developed visualization techniques customized for power grid analytics. This section describes some of the development background, presents the design and implementation details of a multilevel weighted graph layout technique, and discusses the computational performance of the system.

From Semantic Graph to Power Grid Visualization The Have Green framework was initially established to analyze very large semantic graphs [18] with up to a million nodes. A semantic graph is a network of heterogeneous nodes and links annotated with a domain ontology. The ontology of a semantic graph is a description or specification of the concepts and relationships that exist within the semantic graph. A semantic graph is basically a social network [27] that has high degrees of clustering and small average path lengths relative to their number of nodes. Despite the fact that both a semantic graph and an electric power grid possess small-world network [19][28] properties, they also have a number of distinctive characteristics that pose different challenges to our design. While a power grid is large, the entire North American power grid system contains well over 50,000 buses (or

3.1

This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication.
AUTHOR ET AL.: TITLE 3

graph nodes). This is far smaller than a semantic graph that can easily reach a million nodes and beyond. That means the quest for interactive response time is easier for GreenGrid to accomplish. A power grid has a fixed number of nodes and relative coordinates. If we include geographic information in our visualization, it even has fixed coordinates. This implies that we can pre-compute some of the graph layout information for real-time applications. On the other hand, GreenGrid applies a weighted graph layout approach (see Section 3.3) to visualize dynamic information. So we still need to compute a portion of the layout problem in real time. Structurally, a power grid, which is physically connected by transmission lines, is always planar due to its 2dimensional spatial orientation. A semantic graph, on the other hand, may contain cross-links; because of this, creating numerous structures of interest challenges many graph algorithms. The technology transfer between Have Green and GreenGrid is indeed smooth and effective. This allows us to focus on developing new, specialized visualization techniques for electric power grids rather than solving yet another very large graph matrix as we have done for Have Green.

layouts at the corresponding levels. Because the layout of a coarse graph provides critical clues about the global structure of the next finer one, the entire multilevel layout can be completed in an order of magnitude faster than any conventional force-directed algorithm.

3.2 Multilevel Graph Visualization A power grid graph that is drawn based on geographic locations (of electrical buses and transmission lines between them) does not necessarily represent the most desirable visualization because it fails to show the electronic view of an electric power system. Ideally, we want a visualization that reflects the properties and characteristics of the power grid with or without the auxiliary features such as icons that lay on top of it. As we will discuss later, a graph with a user-definable layout scheme can deliver that idea. But first we have to figure out how to draw a good-sized graph in interactive time. We mentioned earlier that the North American power grids contain a network of well over 50,000 buses. While a graph with 50,000 nodes is not an especially large dataset by todays standards, it is still a significant hurdle for many real-time systems that run on desktop computers. To address the problem, GreenGrid uses a multilevel approach [32] extensively to meet the interactive time requirement. Instead of drawing an entire graph all at once like a force-directed algorithm does, a multilevel algorithm first takes a series of coarsening steps to create a hierarchy of increasingly coarse graphs. Links are progressively removed from the graphs at each level, and nodes tied to the removed links are merged with their neighbors. This coarsening step is repeated until a predefined graph size is reached. The second step is to generate an initial layout for the coarsest graph in the hierarchy. This initial layout serves as a seed to generate the layouts of the other graphs in the hierarchy in the final (uncoarsening) step. The uncoarsening step refines the above initial layout and then extends it progressively through the hierarchy all the way to the original graph. Nodes removed in the coarsening step are now brought back to refine the graph

3.3 Design & Implementation of a GreenGrid Graph Unfortunately, a plain force-directed graph layout is not nearly enough to show the strengths or vulnerabilities of the power grids. We want the capability to map the properties and characteristics of the grids to the shape of the graph visualization itself. One viable option is to generate a weighted graph based on the information collected at the buses (nodes) and transmission lines (links). The notion of using a weighted graph to analyze the transmission of electric power along the grid represents a significant innovation in applying visualization in the electric power industry. A GreenGrid graph layout is no longer just a graphic or geographic template that ties icons or glyphs together; it has a first-class visualization status that physically reflects the contents and structurally characterizes the behaviors of the underlying data for different reasoning purposes or analytical applications. The design approach of a GreenGrid weighted graph is simple but powerful. The notion of a weight is used to reinforce the fundamental concept of a force-directed layout model that uses the (attractive) spring force to maintain the local connections and the (repulsive) electric force to preserve the global structure of a graph. More specifically, nodes that carry high values push away nearby nodes, and nodes with lower values end up closer together. Likewise, links with high values are longer; those with low values are shorter. A node in a GreenGrid weighted graph can represent parameters such as voltage or load (power delivered by a power station or transformer), whereas a link can show impedance or power loss. Multiple parameters can be combined using different weight proportions. The implementation of a GreenGrid weighted graph is built on the same multilevel graph visualization described in Section 3.2 with a few extensions. The multilevel graph hierarchy must now record not just the coarsening information but also the weights of all graph nodes and links within the multilevel hierarchy. The weights for nodes and links are normalized before the coarsening process begins. At each coarsening level, a coarsened node carries the average of the weights of all the nodes it represents. For links, the process is a little more complicated. When a node is merged into another, the weight of the removed link is stored in a list in the new merged node. After the merging step is finished, the link in the new coarsened graph is assigned a new weight based on the average of all the link values that were stored in the lists of both its source and destination nodes. In other words, nodes are tuned before the links in all levels. Figures 1a and 1b show the node weights before and after a coarsening step. Figures 1c and 2d show the link weights before and after the same step. It should be mentioned that the shapes of the graphs (or lengths of the links) have not yet been adjusted in this figure.

This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication.
4 IEEE TRANSACTIONS ON JOURNAL NAME, MANUSCRIPT ID

The ability to change the source of the weighting values leads to the potential for hundreds of different plots being generated from a single power flow, or state estimation, calculation. Potential node weighting factors include: voltage magnitude, voltage angle, real power injection, reactive power injection, real load, reactive load, etc. Potential link weighting factors include: resistance, reactance, impedance, real power flow, reactive power flow, real losses, reactive losses, etc. In Figure 3a, all the nodes are weighted at the same perunit value of voltage and all the lines are weighted at the same per-unit values of impedance. In Figure 3b, the upper right node weighting has increased from 1.01 to 1.05 per unit. The result is a change in the way that the network is visualized.

Figure 1: Changes of node weights a) before and b) after a coarsening step. Changes of link weights c) before and d) after the same step.

Once we have a hierarchy of graphs that contain values for nodes and links, we can begin the uncoarsening process starting with the coarsest graph. A basic non-weighted force-directed layout is created on this smallest graph. The graph then goes through a spring-electric algorithm that rebalances the layout based on our node and link weighting. This works by changing both the spring and electrical forces to accommodate for the weights of each node and link. The uncoarsening step is repeated until the finest graph of the hierarchy is reached and generated. Figure 2 shows the impacts of the a) link weight, b) node weight, and c) node & link combined weighting.

3.4 Visualization Design Rationale There is a natural parallel between the nodes and links of a force-directed layout model and the electrical buses and transmission lines of a power system. For example, the electric repulsive force (of the force-directed model) that is applied to graph nodes could be proportional to the voltage angle at each bus. Similarly, the corresponding spring attractive force (of the force-directed model) that is applied to the graph links could be proportional to the impedance of the transmission lines.

Figure 3: a) Same per unit values of impedance and per unit value of voltage generate a symmetric circuit graph. b) Same per unit values of impedance but different per unit value of voltage (upper right one is higher) generate a distorted graph.

3.5 Computational Performance Because the graph nodes (buses) and links (transmission lines) of the power grids have fixed coordinates, a graph layout based on geographic locations can be pre-compiled for various uses afterwards. Similarly, because the power grids have a fixed set of nodes and links, even non-geographic layouts such as hierarchical or force-directed can be pre-computed and stored for different applications later on.

Figure 2: a) Link weighting onlythe link connecting the clusters is weighted at 1, all the other links are 0.1. The graph is stretched along the blue arrows. b) Node weighting onlythe center cluster node on the right is weighted 1; all the other nodes are 0.1. The graph is stretched along the cyan arrows. c) Node and link weightingthe center node on the right is weighted 1, the middle connecting link is weighted 1, and all the other nodes and links are weighted 0.1.

This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication.
AUTHOR ET AL.: TITLE 5

That leaves the drawing of a weighted graph that requires a user to set the weights of both the node and link parameters for a new or time-varying visualization. Fortunately, the combination of a small data size and a very effective multilevel graph drawing system (as described earlier) allows GreenGrid to generate all the power grid visualizations interactively with little or no delay. Table 1 shows the computation time (in wall clock seconds) required to lay out a weighted graph of the WECC 255kV+ system on a Dell Precision 670 desktop computer with an Intel Xeon 2.8 GHz CPU and 4GB of memory. The WECC 255KV+ grid contains 1,868 nodes and 4,109 links (as compared to 14,090 nodes and 16,246 links in the entire WECC grid). The results in Table 1 represent only a subset of options to weight a GreenGrid graph.

network. The Lower Monumental Dam and McNary Dam are directly connected via a 500kV transmission line. The network between Lower Monumental and Four Corners is a much longer path that is composed primarily of 230kV and 345kV transmission lines, resulting in a weaker coupling. The combination of geographic distance and network strength leads to generator couplings that are not always intuitive when viewed in a purely geographic representation. If the coupling between two generators, or groups of generators, becomes too weak, or the system oscillations become too strong, it is possible for the generators to lose synchronism. The proper visualization of system information can help identify which areas of the system are weakly coupled.

Table 1: Computation time required to draw a weighted graph layout of the WECC system. The 225KV+ grid contains 1,868 nodes and 4,109 links. Link Weight Factor Impedance Impedance Impedance Resistance Resistance Resistance Reactance Reactance Reactance Node Weight Factor Base Voltage Voltage Level Phase Angle Base Voltage Voltage Level Phase Angle Base Voltage Voltage Level Phase Angle Time (Wall clock seconds) 0.234 0.25 0.25 0.234 0.297 0.875 1.421 0.953 0.953

CASE STUDY

In order to show the utility of GreenGrid in visualizing power system information, it is necessary to examine a significant power system event that is well understood. Due to their economic as well as social impact, no power system events have been studied in more detail than widearea blackouts. The last major wide-area blackout in the WECC system occurred on August 10, 1996 [17][24].

4.2 The August 10, 1996 Disturbance The case examined in this paper is the WECC Disturbance of August 10, 1996 that occurred at 15:46 Pacific Advanced Standard Time (PAST) [29]. The system conditions prior to the disturbance were characterized by a high system load due to warm weather that resulted in large transfers of power from Alberta and British Columbia into the Pacific Northwest and from the Pacific Northwest into California. In addition, three critical 500kV transmission lines were out of service. These three transmission lines had been supplying reactive power support to the Portland, Oregon area. Additional equipment outages at the Keeler substation reduced the amount of reactive power that the Keeler Static VAR Compensation (SVC) [11] could supply, further reducing the available reactive power support in the area. At 15:42:57 PAST, the heavily loaded Keeler-Allston 500kV line sagged too low and flashed to ground; protective devices subsequently removed the line from service. This was the initiating event that caused large system oscillations and the eventual separation of the system into four synchronous islands. Although Remedial Action Scheme (RAS) systems existed that were designed to successfully island the WECC in precisely such circumstances, these systems had not been activated; as a result, the system islanded uncontrollably resulting in more islands. The four islands are shown in Table 2.
Table 2: Electric power information of the four islands. Island Alberta Northern (BC, OR, WA, MT, WY, ID) No Cal (North of Los Angeles to OR border) Southern (So Cal, NV, AZ, NM, El Paso, Baja) Interrupted Load (MW) 968 2,099 Lost Generation (MW) 146 5,689 Min Freq. (Hz) 59.00 60.00 Max Freq. (Hz) 60.40 60.40

4.1 Some Background Under steady state conditions, all of the WECC system generators operate synchronously. As a result of synchronous operation, any perturbation to a generator will affect every other generator in the system. The extent to which the other generators are affected depends on how strongly coupled the generators are. For example, if the settings are changed on a generator at Lower Monumental Dam in Washington State, there could be an appreciable effect at McNary Dam in Oregon. In contrast, generators located in the Four Corners may not see an appreciable effect. While the geographic distance between McNary and Four Corners does account for some of the differences in response, the key difference is the interconnecting electrical

11,602

7,937

58.54

60.70

15,820

13,497

58.80

61.30

This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication.
6 IEEE TRANSACTIONS ON JOURNAL NAME, MANUSCRIPT ID

In general, each island comprised generators that were strongly coupled to one another. Because there had initially been large power transfers from the north to the south, the resulting four electrical islands either had insufficient or excessive generation with respect to the electrical load. In islands with insufficient generation, the system frequency began to drop and load was shed. In islands with excessive generation, the frequency increased and actuated protective devices. The imbalance of load and generation after the system separation resulted in 7.5 million customers experiencing an interruption of service, ranging in duration from several minutes to several hours.

4.3 Disturbance Analysis and Visualization As was mentioned in the previous section, there are currently system separation schemes in operation that are designed to prevent uncontrolled separation of a system. One of the significant limitations of the current schemes is that the points at which they separate the system are fixed. This type of fixed-point separation scheme can be seen on the California Oregon Intertie (COI). The separation schemes that exist in the WECC are examples of RAS. In these systems, the status of critical components are monitored and if certain threshold values are exceeded, then actions are automatically taken. For example, if a pre-fault flow of greater than 2,500MW from north to south exists at two given substations, and there is a subsequent loss of a transmission line between the two, then a separation signal is generated. The theoretical basis

for such a signal is that it is better to control the separation of a system than to allow it to occur uncontrollably. As a result, these systems are generally installed in regions where the interconnections are relatively weaki.e., the loss of 1 or 2 lines will cause a significant instability that could lead to uncontrolled islanding. These RAS systems require a significant amount of system analysis prior to installation, and once installed, they cannot be simply moved to monitor another transmission line. The result is a system that accomplishes the task for which it was designed but lacks flexibility. Each time there is a significant change to the systeme.g., the installation of a new transmission lineit is necessary to confirm that the natural system separation points have not changed. Historically, this has not been a significant issue, but new operating paradigms are changing the way the electricity infrastructure is being used. One capability of GreenGrid is quickly identifying where the system would be mostly likely to separate if an uncontrolled islanding event were to occur. To demonstrate this capability, the key points of separation for the August 1996 blackout are first plotted with a geographic visualization and then transposed to an impedance visualization. In Figure 4, the left side shows the WECC system, 230kV and higher, geographically represented. The right side shows the impedance representation of the same physical system. In the impedance representation, the nodes are weighted using the voltage phase angle, and the links are

Alberta

Northern Northern California

Southern
Figure 4: The left side shows the geographic layout of WECC, 230kV or higher. The right side shows a GreenGrid layout with additional weights applied to both nodes (using voltage phase angle) and links (using impedance). See Table 2 for individual areas covered in the four islands (i.e., Alberta, Northern, Northern California, and Southern).

This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication.
AUTHOR ET AL.: TITLE 7

weighted using the transmission line impedance values. The impedance representation clearly shows that the system nodes are clustered into various groups. These groups show the electrical buses and their associated generators, which have a strong synchronous coupling. The weak coupling between strongly coupled areas indicates the lines where an unstable system is most likely to separate. This is seen in Figure 4 where the points of separations of the blackout are highlighted. The points of separations are indicated by the brushing links between the two visualizations in the figure.

DESIGN STRENGTHS AND WEAKNESSES

This section discusses the strengths and weaknesses of the GreenGrid design as compared to the traditional GIS-type visualization widely adopted by the industry.

5.1 Visualizing Geography versus Physics The visualization designers within the electric power industry tend to place considerable attention on presenting the geographic information and fail to see the importance of integrating the metadata of the graph with the graph layout itself. As a result, many overlook the opportunity to capitalize on the power of data visualization. Instead of visualizing the geography of a power grid, GreenGrid uses a weighted graph layout approach to model the physics (or electronics) underlying the power grid. A constrained graph layout, such as one that follows geographical coordinates, can sometimes create visual obstacles for users to detect the hidden structures behind a perfectly drawn and very accurate geographic visualization. For example, Figure 5a visualizes the base

voltage of the WECC buses using a geographic layout. The combined Northern (cyan) and Southern (blue) California power grid represents no more than one-tenth of the entire WECCs size. In reality, California represents a much larger share of the power system, which is accurately depicted in the GreenGrid layout in Figure 5B. GreenGrid enables operators, engineers, and planners to effortlessly obtain information that previously had only been accessible with significant analysis. For example, the entire WECC grid looks like a planar mesh that connects all the Western areas in Figure 5a. By weighting the nodes with the voltage phase angle and the links with the transmission line impedances, the well-known WECC power flow paths can be instantly seen. In general, power flows from Canada into the United States, and there is a ring path from Washington, to California, to the Four Corners. This ring feature is clearly shown in the GreenGrid layout in Figure 5b. Perhaps the most significant result from analysis performed with GreenGrid is the ease with which natural system separation points can be identified, as discussed in the case study. Also, if the proper weighting values are used, then the relative stress on these separation points can be examined.

5.2 Visualization Expressiveness and Scalability Unlike a GreenGrid design that embeds the information in the graph, a geographic layout has to rely on the use of icons and glyphs to present the information. Icons and glyphs are graphical representations that play a critical role in the success of information visualization. However, they also bring a number of inherent weaknessesincluding a lack of expressiveness and scalabilityto our application.

Figure 5: The base voltage of WECC buses is mapped to a heat-temperature colormap in a) a traditional geographic layout and b) a GreenGrid layout. The Northern and Southern California portion of the WECC are marked in cyan and blue. The ring path of WECC is shown in purple.

This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication.
8 IEEE TRANSACTIONS ON JOURNAL NAME, MANUSCRIPT ID

Icons and glyphs are particularly good at indicating local spatial details of various spots within visualization. However, they are relatively weak in showing global structures that cover a wide area because they require the users to mentally decipher the integrated meaning of individual icons and glyphs. GreenGrid incorporates the metadata into the shape of the graph and lets the stretched or compressed graph links bind the sparsely located information together. A second weakness comes from the fact that the size of icon is generally not scalable. This is particularly true when we have to visualize information in a densely populated area, where we find a large selection of information but a small amount of display space. However, we find relief when we apply the icon concepts on top of a GreenGrid layout, which doesnt have the same geographic restrictions of a GIS visualization. For example, we visualize the phase angle information of WECC, which broadly reflects the electricity flow direction from the sources (white) to the sinks (black) by following the color of the heat temperature colormap in Figure 6. While we can see the flow direction in both Figures 6a and 6b, we can only see the information of individual cities in dense areas such as Northern and Southern California in Figure 5b. Although we merely use color pixels in this example, the more semantic-rich icon design will only exacerbate the problem.

or combined layout approaches best serve. Furthermore, because GreenGrid is designed to promote specific visualization capabilities through features such as semantic encoding, multilevel graph visualization, and forcedirected layout, we wish to evaluate how well GreenGrid supports its intended benefits and use. The usability study was designed and directed by a usability specialist that was not a member of the GreenGrid software development team.

6.1 Participants Although we have presented GreenGrid as a visual tool for electric power grid analysis, GreenGrid may be considered a general visual graph analysis tool that is applicable to many different network problems. The ability to integrate semantic properties into the layout and shape of a graph should prove useful and beneficial to the larger scientific and engineering community. To evaluate this, we conducted usability testing with two groups of participants from various departments at PNNL. Nine participants were electrical engineers that specialized in electrical power systems. As expected, these engineers were familiar with electric grid and geographical concepts and representations. Nine other participants were junior computational science researchers and developers with no power systems expertise. These participants were generally familiar with geographic layouts as found in conventional maps, atlases, and globes that people encounter in their daily lives. All participants had no or very limited prior knowledge of and experience with the GreenGrid software. 6.2 Apparatus Evaluations were conducted at EIOC [6] using a Windows desktop computer on which the GreenGrid software had been installed. Participants interacted with the GreenGrid software primarily through the use of the mouse. As shown back in Figure 5, the GreenGrid software is capable of displaying network data through either a geographic or GreenGrid layout. These two layouts represented the two conditions of the usability experiment. As part of the same software system, the geographic and GreenGrid layout tools have similar interactive features. Both layout tools provide common functions for selecting nodes and edges, bringing up and viewing node and edge meta-data, and panning across and zooming into and out of the graph. As such, the two layout tools may be seen as having similar levels of interactivity, which better enables us to compare the core layout paradigms as opposed to the effects of auxiliary features. 6.3 Procedure Participants took part in individual sessions that lasted approximately thirty minutes. Each usability session involved a single participant and a single evaluator. At the beginning of each session, the evaluator described to the participant the general use of the GreenGrid software and the objectives and unique features and capabilities of the GreenGrid layout. The evaluator also presented the geographic layout in contrast to the GreenGrid layout. Following the software introduction, the evaluator then asked the participant to work through seven different

Figure 6: The phase angle field is mapped to a heat-temperature colormap. Both a) and b) clearly show the flow direction, but only b) shows the information of individual cities and their neighborhoods.

USABILITY EVALUATION

We conducted a usability evaluation of GreenGrid with participants from PNNL. For the evaluation, we derived seven network analysis problems that participants were to solve by analyzing electric power grid data using a traditional geographic layout and a GreenGrid layout. The nodes in both layouts were color-coded to represent voltage level or phase angle values. The electric power grid data displayed in the layouts were strictly generated through simulation. The objective of the usability study is to assess and compare the usefulness and usability of the geographic and GreenGrid layout approaches. We expect that both layouts will have particular advantages and drawbacks and wish to explore those conditions and situations that the individual

This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication.
AUTHOR ET AL.: TITLE 9

network analysis problems, one problem at a time, using each of the layouts. In introducing a problem, the evaluator would describe the analysis problem to the participant and, in some cases, provide simple visual examples explaining particular visualization or interface concepts. Participants were asked to complete the task twice using each of the layout tools on different networks of comparable size and complexity. Half of the total participants were asked to solve the set of problems using the geographic layout first, while the other half were asked to use to GreenGrid layout first. Each participant would alternate the layout that was applied first across the different problems. Given that each participant applied both layout tools, the usability experiment was of a repeated measures design with the layout tool serving as the independent variable. After presenting a problem and fielding any questions the participant might have, the evaluator would direct the
Table 3: Usability study problem descriptions. Problem Problem Description Number Identify the three transmission lines with the highest impedance 1 and three transmission lines with the lowest impedance as indicated by the lengths of graph edges Given two regions of a power grid network or graph, identify 2 which region consists of electrical buses or nodes with the higher phase-angle values as indicated by longer edges 3 4 5 6 Identify the five asynchronous islands or clusters of an electric network or graph Given a power grid network or graph with five outlined synchronous islands or subgraphs, rank islands or subgraphs by generator or node population from largest to smallest From a set of four select cities, determine whether each city is an energy source, conduit, or sink based on the colors of surrounding generator nodes From a set of four select states, determine whether each state is an energy source, conduit, or sink based on the colors of contained generator nodes From a set of four select states, rank them by generator or node population from largest to smallest

participant to start solving the problem using a particular layout tool. The evaluator would then track and document the time required for the participant to complete the task. The evaluator would then direct the participant to solve the same problem using the alternative layout tool on a different but comparable dataset or graph, and again track and document the task completion time. After the timings, the evaluator would ask the participant to rate his or her satisfaction with each layout tool towards the solving of the problem using a 5-point Likert scale (1 strongly disagree, 2 mildly disagree, 3 neither agree nor disagree, 4 mildly agree, 5 strongly agree). Finally, the evaluator would calculate and document the accuracy of the participants answers for each layout. Accuracy was determined by calculating the percentage of targets or patterns correctly identified or ranked (see Table 3 for information on how accuracy was computed for each problem).
How Accuracy Is Measured % of transmission lines or links correctly identified % of generator or node pairs correctly identified % of asynchronous islands or clusters correctly identified % of islands or subgraphs correctly ranked by population % of cities correctly identified as energy source, conduit, or sink % of states correctly identified as energy source, conduit, or sink % of states correctly ranked by population

Salient Feature Tested Physics embedded in graph (GreenGrid) Physics embedded in graph (GreenGrid) Visibility of global structures (GreenGrid) Visibility of all nodes (GreenGrid) Geographic locations (GreenGrid and Geographic) Geographic boundaries (Geographic) Visibility of all nodes (GreenGrid) and geographic boundaries (Geographic)

Hypothesis GreenGrid will achieve higher performance than Geographic GreenGrid will achieve higher performance than Geographic GreenGrid will achieve higher performance than Geographic GreenGrid will achieve higher performance than Geographic GreenGrid and Geographic will achieve comparable performance Geographic will achieve higher performance than GreenGrid Combined GreenGrid and Geographic will achieve higher performance than GreenGrid or Geographic alone

6.4 Analysis Problems Participants were to complete the seven network analysis problems listed in Table 3. The problems were described and worded in a way that both electrical engineers (EEs) and information technologists (ITs) would understand. For example, the evaluator would describe the power grid or network as consisting of generators or nodes and transmission lines or edges. For the most part, EEs and ITs had little problem mapping power grid concepts to general components of a graph. The set of analysis problems was designed to test the performance of specific features and capabilities of both layout tools. Performance was to be evaluated in terms of participants being able to solve certain tasks more rapidly, with greater accuracy, and higher user satisfaction. For problems 1-4, we hypothesized that GreenGrid layouts encoding of semantics into the graph and higher visibility of both global structures and individual nodes would result in higher user performance over the geographic layout for specific tasks. For problem, 5, we hypothesized that both

layouts capabilities to identify geographic locations would lead to comparable user performance for certain tasks. For Problem 6, we hypothesized that the geographic layouts ability to show geographic boundaries would result in higher user performance over the GreenGrid layout for specific tasks. Finally, for Problem 7, we hypothesized that the combined use of the GreenGrid and geographic layouts would result in higher performance than the individual layout tools alone for tasks that require both full node visibility and geographic boundaries. To better illustrate the analysis problems, Figure 7 presents sample pairs of GreenGrid and geographic layout visualizations associated with the problems of the usability study. Only a subset of the visualizations applied in the usability study is shown. Both the geographic and GreenGrid layouts contained nodes that were similarly color-coded to represent voltage level or phase angle values. A colormap legend was displayed on the screen as participants worked through each problem. Figure 7g and 7h are annotated for the readers only. The annotations are not part of the test questions.

This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication.
10 IEEE TRANSACTIONS ON JOURNAL NAME, MANUSCRIPT ID

g f

Figure 7: Sample set of GreenGrid and geographic layout visualizations applied used in usability evaluation. Graphs show simulated electric power grid data in the form of a-b.) close-up GreenGrid and geographic layouts with edges weighted by impedance values for Problem 1, c-d.) global GreenGrid and geographic layouts with nodes weighted by phase angle values and color-coded target regions for Problem 2 (participant is assigned two random regions), e-f.) global GreenGrid and geographic layouts with nodes weighted by voltage level values for Problem 3, g-h.) global GreenGrid and geographic layouts with nodes weighted by voltage level values and synchronous islands pre-circled for Problem 4, and i-j.) global GreenGrid and geographic layouts with nodes weighted by phase angle values and cities pre-annotated for Problems 5-7 (participant is assigned four random cities for Problem 5 and four random states for Problems 6 and 7).

6.5 Experimental Results and Discussion To determine whether the layout tools have different effects on different classes of participants, we separated our analysis of experimental results across three populations: electrical engineers (EE), information technologists (IT), and the combined set of all participants (All). The performance measures we collected and examined in our usability study were task completion time, solution accuracy, and user satisfaction. To analyze the effect of the layout tool on task completion times, we applied a t-test for correlated samples for the two conditions (geographic and GreenGrid layouts) of problems 1-6 and a one-way Analysis of Variance (ANOVA) for correlated samples for the three conditions (geographic, GreenGrid, and combined geographicGreenGrid layouts) of Problem 7. As for accuracy and user satisfaction measures, participant scores were limited to a small, fixed number of categories or results (e.g., 1 of 5 patterns correctly identified, 1-5 satisfaction rating). For these measures, non-parametric tests were appropriate for analyzing effects. Thus, for problems 1-6 and their two conditions, the Wilcoxon Signed-Rank Test [30] was applied to analyze accuracy and user satisfaction results. For the three conditions of Problem 7, the Friedman Test [9] was applied to analyze accuracy and user satisfaction results. With all statistical tests, we sought a standard 0.05 level of significance. Experimental results for each problem and measure are presented in Table 4 showing the mean scores, standard deviations, and p-values associated with participant classes, layout tools, and performance measures. 6.5.1 Semantic Encoding From our observations, we found that the encoding of

physical or semantic properties onto the topology of the network allowed participants to explicitly view the impedances on transmission lines (Problem 1) and the effects on electrical buses with low or high phase angles (Problem 2) in the GreenGrid layout. In contrast, these quantities were not explicit or observable in the geographic layout, thus requiring participants to compute and manage information and numbers in their heads. As a result, the participants were able to identify the critical transmission lines and high phase angle regions in the graph more rapidly and with higher user satisfaction (with the exception of EEs for Problem 2) using the GreenGrid layout over the geographic layout.

6.5.2 Global Structures Earlier in this paper, we described how the GreenGrid layout allows users to view and identify more global structures and patterns, while traditional layout approaches tend to focus more on local spatial details. Global structures emerge in the GreenGrid layout because physics properties and semantics are incorporated into the shape of the graph. For this problem, participants were to look for synchronous islands that arose in an electric power grid based on physical properties such as the positions of electrical generators and the flow of electricity. The synchronous islands appeared in the GreenGrid layout as dense node areas separated by long links. The nodes within a synchronous island might span different value ranges as denoted by node colors. As participants applied the geographic layout towards solving Problem 3, we observed that islands in the geographic visualization did not visually emerge because the nodes were fixed to geographic locations. Rather, participants would identify islands by selecting sets of nodes that were close in both color and distance, which was a suboptimal approach

This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication.
AUTHOR ET AL.: TITLE 11

because synchronous islands may span more than a single value (or color) range. Overall, the GreenGrid layout was generally applicable and useful for solving Problem 3, while the geographic layout had very limited utility, as

illustrated by the faster task completion times and higher user satisfaction ratings of the GreenGrid layout over the geographic layout.

Table 4: Experimental results for comparison of GreenGrid and geographic layouts in regards to task completion time, solution accuracy, and user satisfaction. Nine electrical engineers (EE) and nine information technologists (IT) participated in usability study. Task Completion Time (s) Solution Accuracy (%) Satisfaction Rating (1-5) ProbPartiLayout lem cipant p-value Test p-value Test p-value Test Wilcoxon SignedRank Test Geographic GreenGrid Geographic GreenGrid Geographic GreenGrid Geographic GreenGrid Geographic GreenGrid Geographic GreenGrid Geographic GreenGrid Geographic GreenGrid Geographic GreenGrid Geographic GreenGrid Geographic GreenGrid Geographic GreenGrid Geographic GreenGrid Geographic GreenGrid Geographic GreenGrid Geographic GreenGrid Geographic GreenGrid Geographic GreenGrid IT Geographic Combined GreenGrid 7 EE Geographic Combined GreenGrid All Geographic Combined 35.11 9.56 47.67 10.44 41.39 9.11 22.78 8.78 29.22 8.94 26.00 26.56 44.33 15.33 44.89 20.94 44.61 18.67 19.56 16.56 22.78 17.61 21.17 16.00 16.22 18.33 17.44 17.17 16.83 23.89 19.44 23.00 23.89 23.44 21.67 21.78 22.11 51.44 26.11 22.89 41.22 23.94 22.50 46.33 8.02 4.16 21.61 3.81 17.08 6.45 15.52 5.04 21.88 5.62 18.70 12.05 13.31 6.06 17.00 10.91 14.82 9.72 11.76 9.79 17.77 9.52 14.71 4.74 2.39 8.32 5.29 6.68 4.03 10.83 7.13 9.41 15.37 9.85 11.85 8.71 19.80 11.36 11.52 11.78 10.07 9.27 16.66 <0.0001 <0.01 1-way ANOVA, correlated samples 7.04 <0.0001 100 100 100 100 100 100 t-test, correlated samples 100 100 100 100 100 100 t-test, correlated samples 9 100 22 100 16 67 t-test, correlated samples 33 71 31 69 32 100 t-test, correlated samples 100 100 100 100 100 100 t-test, correlated samples 100 100 100 100 100 33 33 67 56 25 89 44 29 78 0 0 0 0 0 0 0 0 0 0 0 0 20 0 23 0 22 26 14 30 15 28 14 0 0 0 0 0 0 0 0 0 0 0 0 13 13 25 37 13 22 29 13 26 <0.001 <0.01 <0.05 Friedman Test 1.56 4.22 2.33 4.44 1.94 Wilcoxon SignedRank Test 4.00 2.00 3.56 2.78 3.78 2.39 4.67 Wilcoxon SignedRank Test 1.44 4.67 1.67 4.67 1.56 Wilcoxon SignedRank Test 3.67 1.56 3.56 2.00 3.61 1.78 4.44 Wilcoxon SignedRank Test 4.11 4.33 3.22 4.39 3.67 Wilcoxon SignedRank Test 3.33 4.00 3.89 4.00 3.61 4.00 2.56 2.89 4.11 2.78 2.33 4.33 2.67 2.61 4.22 0.53 0.67 1.00 0.62 0.87 1.00 1.12 1.01 0.83 1.00 1.04 0.50 0.73 0.50 0.71 0.49 0.70 0.71 0.88 1.01 0.87 0.85 0.88 0.73 0.60 0.71 0.83 0.70 0.84 1.00 1.00 0.78 0.50 0.92 0.77 0.73 0.78 0.60 1.20 0.87 0.50 0.97 0.85 0.55 <0.0001 <0.01 <0.01 Friedman Test Wilcoxon SignedRank Test Wilcoxon SignedRank Test Wilcoxon SignedRank Test Wilcoxon SignedRank Test Wilcoxon SignedRank Test Wilcoxon SignedRank Test t-test, correlated samples IT 1 EE All IT 2 EE All IT 3 EE All IT 4 EE All IT 5 EE All IT 6 EE All GreenGrid 11.33 3.43 <0.0001 <0.01 <0.0001 <0.05 <0.05 <0.01 <0.001 <0.01 <0.0001 >0.05 >0.05 >0.05 >0.05 >0.05 >0.05 >0.05 >0.05 >0.05 100 0 >0.05 >0.05 >0.05 >0.05 >0.05 >0.05 <0.01 <0.01 <0.001 <0.05 <0.05 <0.001 >0.05 >0.05 >0.05 >0.05 >0.05 >0.05 4.67 0.50 <0.01 <0.01 <0.01 <0.05 >0.05 <0.01 <0.01 <0.01 <0.001 <0.05 <0.05 <0.001 >0.05 >0.05 <0.05 >0.05 >0.05 >0.05

This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication.
12 IEEE TRANSACTIONS ON JOURNAL NAME, MANUSCRIPT ID

6.5.3 Node Visibility The visualizations for Problem 4 showed two separate pairs of synchronous islands that were similar in population sizes. From our observations and results, we found that participants were better and more efficiently able to judge and rank the sizes of synchronous islands using the GreenGrid layout because they were able to view all the nodes of each island, which led to better accuracy and higher user satisfaction. We further expect that the ability to view the full resolution of nodes will become increasingly critical with larger datasets and graph sizes, where the potential for data points to be hidden or obscured in a visualization becomes much more likely. As for speed, the mean task completion times appeared to improve with the GreenGrid layout, but the differences were not found to be significant for any participant class. 6.5.4 Geographic Locations As we hypothesized, we found the performance of the GreenGrid and geographic layouts to be similar in characterizing cities as sources, conduits, and sinks. Differences between the GreenGrid and geographic layouts with respect to task completion time, accuracy, and user satisfaction were generally not significant. The lone exception was that user satisfaction among all participants was found to be significantly higher for the GreenGrid layout over the geographic layout. Because the GreenGrid layout exposes all nodes while the geographic layout does not, the values (or colors) of nodes surrounding city locations may become more obvious and prominent. From our observations, we did find that some participants were better able to discern the values or conditions of nodes surrounding a city more easily using the GreenGrid layout. These participants found that being able to view and discern the spatial relationships surrounding a city was more critical for solving Problem 5 than knowing the citys precise geographic location. 6.5.5 Geographic Boundaries Our analysis results failed to confirm our hypothesis that the GreenGrid layout would incur lower performance than the geographic layout for Problem 6. To the contrary, we found that differences between the GreenGrid and geographic layouts with respect to task completion time, accuracy, and user satisfaction were not significant as participants were able to apply other strategies to compensate for the GreenGrid layouts inability to identify geographical areas or state boundaries. In our observations, we found participants focusing on the cities residing in the target states. From a city, the participant would then extrapolate out to an area that might be reasonably associated with the size of the target state. Although this approach was imprecise, it did yield 100% accuracy for Problem 6 because the data trends across the states were gradual, and thus, the states of generators near a city generally corresponded to those across the rest of the state. We would expect the accuracy of this analysis approach using the GreenGrid layout to be much lower if the data in the graph were much more varying and dynamic. The general strategy that participants applied

was not particularly time-consuming and participants were generally satisfied with the results, which explains why the differences in task completion time and user satisfaction were not significant between the GreenGrid and geographic layouts for Problem 6.

6.5.6 Combining Layouts From our analysis, we were surprised by the time participants required to complete the task using the combined layouts compared to that of the individual layouts. We had expected participants to struggle solving Problem 7 using the individual layouts because critical details and information were missing from each of the visualizations. We further had expected that participants would compensate for the shortcomings of the individual layouts by spending more time analyzing the visualizations and mentally extrapolating from the data shown. Instead, we found that participants often quickly produced rough, instinctual estimates from the individual visualizations with minimal introspection. As participants worked with the combined layouts, however, we found that they took more time and effort to conceptually grasp the contents of both visualizations and examine the mappings across nodes of the two layouts. Thus, participants tended to conduct a deeper, more thorough analysis. As a result, participants generally expended considerably more time with the combined layouts to develop a more accurate solution. We surmise that if the overall problem had required more accuracy in the solution, such as to estimate the exact number of generators per state, participants would have had to spend much more time examining and analyzing the individual layouts to reach reasonable estimates. In this case, we would expect the time spent to complete the problem would be more comparable between the combined and individual layouts. With respect to task completion times and user satisfaction, the combined layouts performed significantly better than the GreenGrid and geographic layouts alone as hypothesized. 6.5.7 EEs vs. ITs One of the questions we sought to examine in the usability study was to determine whether user class had any effect on layout tool performance. Would EEs achieve better or worse performance than ITs using either of the layout tools to solve the network problems? As we presented throughout the previous sections, we found that all differences in performance with regard to task completion time, accuracy, and user satisfaction were not significant between EEs and ITs using either layout tool. This indicates that both the GreenGrid and geographic layout tools support general graph concepts and analyses as opposed to just those associated with electric power grid analyses. This result should not be surprising given that the two layout tools draw from and were developed upon the general Have Green graph analytics architecture. 6.5.8 Summary of Evaluation Results From the usability study, we found that the GreenGrid layout generally offers equal or improved performance

This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication.
AUTHOR ET AL.: TITLE 13

over the geographic layout in terms of accuracy, time to completion, and user satisfaction for most problems in head-to-head comparisons. The GreenGrid layout was particularly promising with regard to accuracy as demonstrated in Problems 1, 2, 5, and 6, where participants achieved 100% accuracy. An expected weakness of the GreenGrid layout in its inability to represent and convey boundaries and areas was also confirmed through Problem 6. Problem 7 represents a scenario in which both the GreenGrid and geographic layouts would benefit the analysis of the problem. We found performance improvements in both accuracy and user satisfaction for the combined layouts compared to the individual tools, but reduced performance in terms of completion time due to the participants greater thoroughness and effort using the combined layouts. Overall, the evaluation results confirm that the GreenGrid layout meets most of our intended design goals, benefits, and uses, as well as exhibits expected weaknesses.

addition to the results shown in this paper, the visualization technology presented potentially offers enormous benefits to the power industry.

ACKNOWLEDGMENT
This work has been supported in part by the National Visualization and Analytics CenterTM (NVACTM) located at the Pacific Northwest National Laboratory in Richland, WA and the U.S. Department of Energy Office of Electricity Delivery and Energy Reliability (DOE-OE). The Pacific Northwest National Laboratory is managed for the U.S. Department of Energy by Battelle Memorial Institute under Contract DE-AC05-76RL01830.

REFERENCES
[1] N. Ahmed and H.J. Miller, Time-Space Transformations of Geographic Space for Exploring, Analyzing, and Visualizing Transportation Systems, Journal of Transport Geography, vol. 15, pp. 2-17, 2007. AREVA T&D Energy Management Systems, http://www.arevatd.com. S.K. Card, J.D. Mackinlay, and B. Shneiderman, Readings in Information Visualization, Using Vision to Think, Morgan Kaufmann, 1999. C. Chen, Information Visualization beyond the Horizon, second edition, Springer, 2004. G. Di Battista, P. Eades, R. Tamassia, and I.G. Tollis, Graph Drawing: Algorithms for the Visualization of Graphs, Prentice Hall, 1999. Electricity Infrastructure Operations Center (EIOC), http://eioc.pnl.gov. C. Erten, S.G. Kobourov, V. Le, and A. Navabi, Simultaneous Graph Drawings: Layout Algorithms and Visualization Schemes, Journal of Graph Algorithms and Applications, vol. 9, no. 1, pp. 165-182, 2005. D. Forrester, S.G. Kobourov, A. Navabi, K. Wampler, and G.V. Yee, Graphael: A System for Generalized Force-Directed Layouts, Graph Drawing 2004, LNCS 3393, pp. 454-464, 2004. M. Friedman, The Use of Ranks to Avoid the Assumption of Normality Implicit in the Analysis of Variance, Journal of the American Statistical Association, vol. 32, no. 200, pp. 675-701, 1937. P.H. Gilbert, Implications of Power Blackouts for the Nations Cybersecurity and Critical Infrastructure Protection: The Electrical Grid, Critical Interdependencies, Vulnerabilities and Readiness, Testimony of the first session of 108th US Congress, Sept 4, 2003. J.D. Glover and M.S. Sarma, Power System Analysis and Design, 3rd edition, Brooks/Cole, 2002. GD2008, 16th International Symposium on Graph Drawing 2008, http://www.ics.forth.gr/gd2008/. D. Harel and Y. Koren, A Fast Multi-Scale Algorithm for Drawing Large Graphs, Journal of Graph Algorithms and Applications, vol. 6, no. 3, pp. 179-202, 2003. I. Herman, G. Melancon, and M.S. Marshall, Graph Visualization and Navigation in Information Visualization: A Survey, IEEE Transactions on Visualization and Computer Graphics, vol. 6, no. 1, pp. 24-43, IEEE CS Press, 2000.

ONGOING AND FUTURE WORK

[2] [3]

Up to this point, GreenGrid has only been used to visualize the outputs of steady state simulations. We envision two further visualization steps for GreenGrid. The first is to use time-varying dynamic simulations for the input. This will allow for more complex visualizations that will be relevant to operations and planning engineers. The second step is to integrate GreenGrid into a control center Energy Management System. By using near-real-time data as an input, the state of the system can be viewed and the tool becomes relevant to control room operators and operations engineers as well as to planning engineers who can use it for post-disturbance analysis. In addition to the two additional visualization steps, there is an analytic step. The structure of the impedance visualization shown in Figure 4 contains a great deal of information about the physical system as well as questions about the system, such as: is there a structure in the impedance visualization that yields a more reliable system? In addition to the impedance visualization previously discussed, there are well over a hundred other options to weight the nodes and links of the power grids. Each of them will give a unique visualization that reflects different aspects of the system. We plan to continue the investigation and report the results when they are available.

[4] [5]

[6] [7]

[8]

[9]

[10]

CONCLUSION

[11] [12] [13]

This paper has introduced the concept of using weighted force-directed visualizations for power system applications. It has been shown that there is tremendous value to leveraging the existing visualization knowledge base to a field that has traditionally not expended significant resources in the area of visualization. The GreenGrid tool has been used to identify several key characteristics of the WECC system. These observations have been confirmed via comparison with the postdisturbance report of the August 10, 1996 blackout. In

[14]

This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication.
14 IEEE TRANSACTIONS ON JOURNAL NAME, MANUSCRIPT ID

[15] IEEE InfoVis 2008, IEEE Conference on Information Visualization 2008, http://conferences.conference.org/infovis/infovis2008/. [16] Journal of Graph Algorithms and Applications, http://jgaa.info. [17] D.N. Kosterev, C.W. Taylor, and W.A. Mittelstadt, Model Validation for the August 10, 1996 WSCC System Outage, IEEE Transactions on Power Systems, vol. 14, issue 3, pp. 967-979, Aug 1999. [18] T. Kolda, D. Brown, J. Corones, T. Critchlow, T. Eliassi-Rad, L. Getoor, B. Hendrickson, V. Kumar, D. Lambert, C. Matarazzo, K. McCurley, M. Merrill, N. Samatova, D. Speck, R. Srikant, J. Thomas, M. Wertheimer, and P.C. Wong, Data Sciences Technology for Homeland Security Information Management and Knowledge Discovery, Report of the DHS Workshop on Data Sciences, Jointly released by Sandia National Laboratories and Lawrence Livermore National Laboratory, Alexandria, VA, Sep. 22-23, 2004. [19] S. Milgram, The Small World Problem, Psychology Today, vol. 2, pp. 60-67, 1967. [20] National Visualization and Analytics Center, http://nvac.pnl.gov. [21] North American Electric Reliability Corporation (NERC), http://www.nerc.com/regional/. [22] T.J. Overbye and J.D. Weber, New Methods for Visualization of Electric Power System Information, Proc. IEEE Symposium on Information Visualization 2000, pp. 131-136, 2000. [23] T.J. Overbye and J.D. Weber, Visualization of Power System Data, Proc. of the 33rd Hawaii International Conference on System Sciences, IEEE Computer Society, 2000. [24] L. Pereira, Cascade to Black, IEEE Power & Energy Magazine, vol. 2, no. 3, pp. 54-57, May/June 2004. [25] K. Sugiyama, Graph Drawing and Applications, World Scientific Publishing, 2002. [26] C. Walshaw, A Multilevel Algorithm for Forced-Directed GraphDrawing, Journal of Graph Algorithms and Applications, vol. 7, no. 3, pp. 253-285, 2003. [27] S. Wasserman and K. Faust, Social Network Analysis-Methods and Applications, Cambridge University Press, 1999. [28] D.J. Watts and S.H. Strogatz, Collective Dynamics of SmallWorld Networks, Nature, pp. 440-442, Macmillan, June 1998. [29] Western Electricity Coordinating Council, Western System Coordinating Council Disturbance Report for the Power Outage that Occurred on the Western Interconnection August 10th, 1996 15:48 PAST, WSCC Operations Committee, Oct 18, 1996. [30] F. Wilcoxon, Individual Comparisons by Ranking Methods, Biometrics, vol. 1, pp. 80-83. [31] P.C. Wong, G. Chin Jr., H. Foote, P. Mackey, and J. Thomas, Have GreenA Visual Analytics Framework for Large Semantic Graphs, Proc. IEEE Symposium on Visual Analytics Science and Technology (VAST) 2006, pp. 67-74, Oct. 2006. [32] P.C. Wong, H. Foote, P. Mackey, G. Chin Jr., H. Sofia, and J. Thomas, A Dynamic Multiscale Magnifying Tool for Exploring Large Sparse Graphs, Information Visualization, vol. 7, no. 2, pp. 105-117, Summer 2008. Pak Chung Wong received the PhD degree in computer science from the University of New Hampshire in 1997. He is a chief scientist and project manager at the Pacific Northwest National Laboratory in Richland, Washington. He serves on the executive committee at IEEE VisWeek 2008 and the steering committee at IEEE Symposium on Visual Analytics Science and Technology (VAST) 2008. His current research interests include visual analytics, visualization, bioinformatics, and homeland security. Kevin Schneider received his B.S. degree in Physics and his M.S.

and Ph.D. degrees in Electrical Engineering from the University of Washington. His main areas of research are power system operations and visual analytics. He is currently a research engineer at the Pacific Northwest National Laboratory in Richland, Washington. He is also an Adjunct Faculty member at the Washington State University Tri-Cities campus and a licensed Professional Engineer in Washington State. Dr. Schneider is a Senior Member of the IEEE. Patrick Mackey received the BS degree in computer science from Washington State University in 2004 and is working on his MS degree in Computer Science at the same university. He is a research scientist at the Pacific Northwest National Laboratory, where he has worked on multiple visual analytics projects. His research interests include visualization, scientific computation, and computer graphics. Harlan Foote received the BS degree in physics from Washington State University in 1966. He is a senior research scientist at the Pacific Northwest National Laboratory, where he has worked on a wide variety of remote sensing projects. His interests include hyperspectral image classification, multiscale image processing for stereo matching, and multisensor data fusion. He is currently working on problems of feature extraction from 3D millimeter wave holographic images. George Chin received the PhD degree in computer science from Virginia Tech. He is a chief scientist at the Pacific Northwest National Laboratory in Richland, Washington. His main area of expertise is in human-computer interaction. At PNNL, he has conducted extensive user and work studies with intelligence analysts and scientists from various domains. Other current research interests include visual analytics, computer-supported collaborative work, social networks, scientific workflow, and scientific problem-solving environments. Ross Guttromson received his BSEE and MSEE degrees from Washington State University, and his Executive MBA degree from the Michael G. Foster School of Business at the University of Washington. Ross is a Senior Research Engineer and Manager of PNNLs Electricity Infrastructure Operations Center at the Pacific Northwest National Laboratory, Richland, WA. He is a licensed Professional Engineer in the state of Washington and a Senior Member of the IEEE. Jim Thomas received the MS degree in computer science from Washington State University. He is a PNNL lab fellow and chief scientist for Information Technologies at the Pacific Northwest National Laboratory. He is the director of the U.S. Department of Homeland Security National Visualization and Analytics Center. He specializes in the research, design, and implementation of innovative information and scientific visualization, multimedia, analytics, and human computer interaction technology. He led the direction of and co-edited, with Kris Cook, the recent book, Illuminating the Path: The Research and Development Agenda for Visual Analytics.

You might also like