You are on page 1of 6

Decision Support for Hydropower Plant Upgrading

Integrating Multi Attribute Decision Making and Knowledge Based Systems


G. Tangen

I f . Staa

Norwegian Electric Power Research lnstitute Energy Systems Division 7034 Trondheim, Norway E-mail: grethe.tangen@efi.sintef.no

Abstract: There is a@ to improve the procedures for ranking reconditioning projects n n hydropower plants to optimize the allmacion of resources, and a concept for decision support based on value focused thinking and multi attribute decision making is developed. This paper suggests an integration of multi attribute decisnonmaking and knowledge based support to realize the concept. The two disciplines have different qualities and they can contribute at different stages of the decision process.
Keywords: Hydropower plant reconditioning, decision support, multi attribute decision making, knowledge based systems.
1. INTRODUCTION
This paper proposes an integration of multi attribute decision making (MADM) and knowledge based support (KBS), to develop a system for supporting decision makers during evaluationand ranking of reconditioning projects in hydro power plants. Despite similar point of departure, MADM and KBS have developed in different directions. They are characterized by different properties and can complement each other. Initially, some general aspects related to decision making and decision processes are discussed, and a proposed concept for supporting the evaluation and ranking of projects is outlined. The paper discusses the characteristic properties of hydropower plant reconditioning projects. Shortcomings related to the procedures for project selection are referred and practical counteractions are presented. Based on our experiences f r o m applying the concept for ranking reconditioning projects [l], an integration of MADM and KBS is suggested to realize the concept. Both disciplines are briefly discussed. We pinpoint their characteristic features to demonstrate the potential benefits of a joint approach and indicate how the two approaches may be used to develop a practical decision making support tool for reconditioning planning.
2. DECISION MAKING THEORIES APPLIED TO

is denoted satisficiirlg and optimizing in 121. Human decision makers tend ito settle for solutions which are good enough, a n d they do not look for the optimal solution. This is satisficing, and it is characteristic for the present routines for reconditioning projects ranking. Optimizing means to find the optimum of a11 objective function. By changing the problem solving strategy from pure satisficing roward optimizing, the fulfilment of objectives become more ceneal and the quality of the decisions is improved. Hence, the quality of clecisions depends considerably on the decision process and to improve the quality of the decisions one must improve the quality of the decision process. We have studied the decision process in general texms and used a schematic process description for the development of a decision support concept [31. The structureis similar to the one presented by [41. We will nor go into the details of the decision process, but it should be noted that even simple models contributes to a deeper understanding of the decision process as such. It helps identifying where major improvements can be achieved and what kind of achievements one can expect at each stage. This is important to suggest effective changes in the decision process.

RECONDITIONING PROJECTS RANKING


M e n proposing a concept for decision support, the goal is to improve the qualify of the decisions made by the decisionmakers. The qudity of decisions is closely related to the fulfilment of objectives. It depends on whether all subjective and objective requirements are taken into account and to which extent they are met by the suggested solution. The efforts of the decision maker to fulfill the objectives can be related to two strategies of problem solving which

[51 presents a way of thinking which can serve as a basis far a f o " d approach for developing a decision making support tool. In "Value Focused Thinking" the general principle is to focus on what really matters when decision problems are evaluated. Through systematic identification of objectives and attributes a value model is established, which represents everythmg the decision maker considers to be important about the problem. The alternurives and the relationships between the altematives and the value model are identified. This, makes it is possible to analyse the consequences of ,various decision alternatives in a systemafk manner . [SI suggests practical ways fo identify and structure the components of the decision problem.

0-7803-31 15-X/96$5.00 0 1996 IEEE.

145

Value focused thmking provides insight about the decision problem and it helps the decision maker to identify and formulate the essence of the problem. Together with a method for making trade-offs among alternatives, it is a good basis for developing a decision makmg support tool. Based on the theory briefly introduced in this section, a concept for supporting evaluation and ranking of reconditioning projects is proposed [l]. The concept represents a tailored framework which facilitates the implementation of the suggested means. The ultimate goal is to develop a system which assists decision makers in exploiting their own capability and expertise to perform evaluations and to make rational decisions. The decisions should be in accordance with a strategy, explicitly formulated as a set of objectives selected by the decision maker. The concept is divided in five parts, and the three first of them are closely related to value focused thinking. In the schematic description, a l l iterations within each stage and interactions between the stages are omitted for the sake of simplicity.

iv) Classification of alternatives and analysis: Before performing a thorough analysis, the alternatives are classified. Based on some simple economic and safety criteria, several alternatives can be ruled out. "he remaining altematives are analysed to provide a complete project description. v) Comparing and ranking reconditioning projects: The weighted objectives and the attributes now constitute the relationships between the reconditioning projects (alternatives) and the strategy for decision making, see figure 1. Several competing criteria must be considered and a multi attribute trade-off method will be used to perform the ranking of the altematives. Sensitivityanalyses are used to evaluate t h e results, and in case the decision maker is not satisfied, he/she can return to any previous stage in the decision process. The decisions and their grounds must be documented.
3. HYDROPOWER PLANT RECONDITIONING
The Norwegian Energy Corporation defines recondirioning as bringing a component back to its original state to elongate its life time. Reconditioning includes major repairs, carried out once in 10-20 years. They may be initiated for several reasons such as economy, techcal state, governmental injunctions, safety requirements and accidents. 'Ihe purpose of maintenance is to keep the component in operation without improving it standard. Even though reconltioning and maintenance are related, the difference in purpose affects the size of the projects and their costs. Large power companies carry out several reconditioningprojects each year, and the high costs related h a t even slight improvements in the to the projects indicate t allocation of resources can contribute to annual savings.
An improved strategy for reconditioning planning in Nmegjan hqdropower plants is needed. Today, hardly any new plants are built, and additional upgrading is required as the average age of the plants increases. At the same time, the introduction of the competitive power market reduces the available funds for reconditioning.

i) Identification and structuring of objectives: Objectivesand subobjectives are identified and formulated. They are organized in a structure, compared and analysed to establish a set of weights which expresses the chosen strategy for resource allocation.

ii) Identification of measurable attributes: ?he attributes whch measure to what extent the objectives are fulfilled are identified and registered. In cases where natural attributes are not available, substitute attributes must be constructed or indirect measures must be identified. The resulting saucture of objectives and attributes is denoted the value model.
iii) Identification of alternatives:
All relevant alternatives are identified and documented. In some cases the decision maker may want to construct new alternatives, for instance by combining already identified altemat ives.

The present practice for hydropower plant reconditioning in the Norwegian Energy Corporation has been studied, and shortcomings which complicate the evaluation and prioritization of reconditioning projects have been identified [3].The conclusions can be summarized as follows:

* Project disparity:
Altributes

The projects are extremely diverse because they are closely connected to power plant components which are different in nature. Besides, the information needed to evaluate the projects include both quantitative and qualitative variables. The disparity makes the projects hard to compare.

0
146

* Multiple participants:
Anernatives

Figure 1: Relating altematives and strategy

A large number of persons are involved in the decision process. The identification and description of projects involve specialists from many disciplines, and the

suggested projects are evaluated at various levels in the organization. This put demands on the communication across organizational levels and the project documentation.

* Lack of strategy:
An overall reconditioning strategy is not formally expressed

and the cause-effect relationships between the reconditioning projects and the reconditioning strategy are indistinct and difficult to assess. Despite an imperfect infomution basis, decisions regarding millions of NOK are made every year. Major improvements can be achieved in the description, evaluation and ranking of reconditioning projects, and considerablesavings may be a result.
4. MEANS FOR WPROVEMENT

provide project rankings and thereby contribute to the development of long term reconditioning plans. The results will be compared with the results from a traditional planning process, which is carried out in parallel. We are now identifying objextives and attributes, and the plant is currently investigated to uncover all reconditioning needs. The reconditioning projects will be described in a new form which includes the informationrequired for the tradeoff analysis. However, our experience shows that existing tools for supporting MADM analysis provide limited assistance in the process of ansalysing the decision problem and esitablishingthe decisl.onmodel. There is a need for support systems which assist decision makers not only in making the trade-offs, but also in the preparations and conclusions of the analysis.
6. AN EXTENDED APPROACH FOR DECISION

Based on the study of the present practice, several means are suggested to overcome the most serious shortcomings.

SUPPORT
We suggest an extended approach for decision support to realize the concept outlined in section 2. The main idea is to combine multi atmibute decision making (MADM) and knowledge based support (KBS)to exploit their unique qualities at different stages of the decision process. The punpose of a decision support system is to enable decision makers to make rational and wellconsidered decisions. To contribute effectively, the system should assist decision makers in controlling the decision process and to the largest extent possible make use of human abilities and make up for the cognitive deficiencies. M.ulti attribute decision making and knowledge based supportrepresent two schools for decision support with the same a i m and historically the similar point of departure. Still, they have developed in different directions and until recently there has been limited interaction between the two disciplines. A joht approach can benefit from both directions. The direct cause for this approach is the id&& need for a decision support system for evaluating and ranking reconditioningprojects in hydro power plants. An integrated approach can offer an implementationof the proposed concept which is more effective than implementations based on MADM or KBS alone. MADM artd KBS are narrrow approaches, selected due to our focus on reconditioning planning. [6], 171 and [8] discuss integration of deciision support systems and decision analysis in a wider perspective.
6.1 Terminology. Multi attribute decision making (MADM) is a subarea of mlulti criteria decision making (MCDM). MCDM has developed from operational research and management science, and represents theories for assisting human decision makers in hmdling multidimensional situations. It includes a number of approaches and techniques for analysing decision situations, generating altematives and balancing conflicting objectives.Problems addressed cover aspects such as ec'onomy, technology, management and i m at aiding einvironmental issues. MCDM methods a

* Standard project description:


All relevant features of a projects must be included explicitly in a standard project description. It leads to a well-considered and more complete information basis. The projects become easier to compare and the communication can be definite. Moreover, standard descriptions depend less on the person who wrote them, and the company can use them for documentation and accumulation of experience.

* Standard procedures:
Standard procedures for evaluating and selecting reconditioning projects is also suggested, because a methodology for making trade-offs among the project altematives will increase the degree of awareness in the process of analysing and selecting projects. Standardized procedures will duect the decision maker in making decisions which are consistent with the available information basis and, unless the reconditioning strategy is changed, the decisions are consistent from one year to another. Just like the introduction of standard project descriptions, the standard procedures will improve documentarionand contribute to putting the communication in more concrete terms.

* Refiiement of the information basis:


Decision makers should increase their awareness regarding the utilization of both quantitative and qualitative information. Technical and economic analysis should be used to derive information which is not readily available and the level of detail in the information basis should be carefully chosen to restrict the information base without excluding relevant aspects of the task.

5. APPLICATION
In a current research project, multi attribute decision making methods are used to evaluate and prioritize reconditioning projects. Related to the upgrading of MAr hydropower plant the proposed concept is followed to

147

decision makers in selecting optimal alternatives by making use of computation, judgements, compromises and inspiration. [4]presents a detailed description of the MCDM field.

KBS
Descnplwe

t -i
I 1

Normative

In its widest extent, MCDM include multiple attributes, multiple objectives and multiple goals. Thus, MCDM is a vast area with several subareas. Controlled by the nature of our problem, we have focused on methods related to multiple attributes. The purpose is to make trade-offs among the attributes to identify the alternative which to the largest possible extent fulfils the objectives. Multi attribute decision making (MADM) is based on the assumption of a rational decision maker and it requires that human preferences can be extracted and modelled. Multi objective decision makmg (MODM) is another main direction within MCDM, which is based on vector optimization. Methods such as linear multi objective programming and goal programming are used to analyse problems where every combination of attributes are possible decision alternatives. It requires that the decision problem is expressed as an objective function with constraints. In [9] MODM is used for power system planning.
Expert systems applyknowledge in a way which resembles human problem solving. They include a knowledge base with information about a restricted domain and inference methods to carry out reasoning functions. Such systems focus on problems usually solved by human problem solvers. The purpose may be to replace the human problem solver, or more often to assist the problem solver in the process of making a decision. The expert systems emerged from artificial intelligence (An, and as a result of a major scientific progress concerning knowledge representation and reasoning techniques.

Analytic and synthetic


Dedaratieand procedural

Dedarative knowledge Numenc algorithms


Expliclt preferences
I

mwge
Symbd processing

lmpliat preferences
L

Process onenled

Interactiveprobtem solving
I

Autonomous problem solving


Non-tractable
I

No mathemahcalvallaation

Mathematical validation

Figure 2: Characteristics of KBS and MADM Their area of application is also different as KBS are used both for synthesis, such as design and configuration, and analysis, such as diagnosis. MADM is a method for analysing decision problems. When it comes to modelling there is a difference in the strategy for establishing a model. In the process of knowledge acquisition for developing KBS, the experts declarative and procedural knowledge about the domain is extracted-The expert has obtained hisher knowledge from years of experience, and it includes concepts, description, relationships and procedures for problem solving. Procedures for eliciting knowledge can assist a system analyst in creating a representative model which capture the experts knowledge about the domain and about problem solving. The model is oriented towards the problem solvingprocess, and the solution is a result of the process.
In MADM the model is oriented towards the solution of the decision problem. There are no efforts on capturing the human decision making process. To establish a model one focuses on identifying objectives, attributes and aitematives, and they are organised in a structured model defined by the selected trade-off method. Unlike KBS, the implicit preferences of the decision maker is extracted and represented explicitly, for instance as weights and utility functions. However, the approach is not rooted in knowledge about human behaviour, because human decision making is definitely not similar to opthizing. Consequently, in the MADM model the distance between the decision makers knowledge and the information incorporated in the model is bigger than in KBS. The model aims at performing numerical optimization, often related to economic aspects. This does not mean that knowledge about the procedures are irrelevant. On the contrary, procedural knowledge is necessary to gain insight but it is not included explicitly in the model.

We use the term knowledge based system (KBS), because expert system tend to be restricted to a particular class of computer systems. By KBS we mean a general approach for acquiring, representing and exploiting human knowledge, based on symbolic programming and automated reasoning. KBS have focused on human machme interaction, which is reflected in the design of interfaces, in the facilities related to modelling of the experts knowledge and in the mechanisms for explaining the line of reasoning.

6 . 2 Differences bemeen MADM and KBS.


There are several major differences between MADM and KBS, and a summary of the most characteristic ones are presented in figure 2. Many of MADMs characteristics are typical for mathematical algorithms.
A typical KBS model is descriptive, while the MADM model is normative. In systems based on KBS the experts knowledge is elicited and represented to be used in a similar manner as the expert would do in a real situation. MADM models establish a norm for decision malang which make the result optimal with respect to the objectives. This is reflected in the information captured i n the models and how it is represented.

148

As one can see, there is an overlap in information in the two approaches. It can be worthwhile to study acquisition procedures in both areas. For KBS there exists computerized acquisition tools which emphasize humanmachine interaction. In MADM the models to a large extent must be established before the computer tools are applied. However, there is a development toward including modelling facilities in the MADM support tools. There is great flexibility when it comes to representing information in KBS. Symbolic programming, rules and frame structures offer many possibilities for model development, compared with the traditional computer software used for developing MADM algorithms. Despite this difference in flexibility, there has been more focus on explicit modelling of preferences, uncertainty and attitude toward risk in MADM than in KBS. There is a fundamental difference between KBS and MADM in the strategy for problem solving. In MADM optimization is central, because the goal is to find the altemative which fulfils all objectives to the largest possible extent. Consequently, the result is decided by scores on various attributes. In KBS the intention is to find a solution which is satisfactory, and there is no intention of finding the optimal solution. This is in accordance with human decision making (satisficing). If the optimal solution should be found, all possible consequences must be evaluated whch will cause combinatorial explosion.

Byrelating the discussion about differences between KBS and MADM to the proposed concept, it becomes clear that the two approaches can contribute at various stages in the concept. An interactive tool for evaluating and giving priority to reconditioning projects can be created by extending an approach based on MADM with KBS. Such a tool can provide appropriate support at all stages of the &!&ion process. Figure 3 illustrates a possible structure of an extended decision support system. The numbers i) to v) refer to the stages of concept outlined in section 2.

rT
I I I I

P l

6
I

I
I

In KJ3S the course of the problem solving is not predefined and it is influenced by events and result achieved during the problem solving. It may also be directed by the decision maker through interaction. After reaching a conclusion, the structured inference mechanism makes it possible to generate explanations to show the line of reasoning. In MADM the problem is formulated in a model and a solution is calculated without any interaction. In case the decision maker is not satisfied, the model can be changed and the computation repeated. The results are based on an algorithm, and is therefore not directly tractable. However, they can be evaluated by using sensitivity analysis.
A strength of MADM is the mathematical validjty of the results. 'Ihe algorithms are restricted by mathematical axioms and the soundness of the final results can be proved. In KBS the validity must be tested. There are various techniques for thls task, but it might be difficult to measure the performance of a support system.

Figure 3: Smicture of decision support system

With focus on the decision process, KBS can be used to develop a framework for decision making which incorporates the ranlcing of altemativesinto a total system. Ely making use of symbol processing, information structures and interaction with the decision maker, the model needed to make the trade-offs can be established. This part of the process will be closely related to knowledge about human behaviour and human-machme interaction. Value focused thinking will be a basic idea in the establishment of the model, together with the requirements related to the specific method for MADM. hMDM is the best approach for trade-off analysis. Neither traditional rule processing or pattem matching can offer more appropriate solutions for handling this task. They are bloth methods based on descriptive information, while our g,oal is to create a new, normative procedure for handling projects and to support decision makers to reach better dlecisions. The nmlative approach is based on knowledge about how project ranking is conducted today, together with guidelines based on how the procedures ought to be conducted to achieve optimal allocation of resources.

6 3 Integrating KBS and MADM The goal is to develop a system which can support decision makers in allocating resources to reconditioningprojects by compensating the human decision makers difficulties with organising and evaluating many aspects at the same time. A concept based on practical means for meetkg the shortcomings, value focused thinking and methods for making trade-offs among altematives has been proposed, and we want to develop an approach which can realize this concept.

149

KBS will also strengthen the analysis following the tradeoffs and the sensitivity analysis of the result. What-if analysis can be organised to evaluate the consequences of changed conditions. To some extent this is included already in some MADM computer tools, but the wide experience with human-machine interfaces in KBS can significantly improve their flexibility. Moreover, KBS information structures can invite the decision maker to interpret the results. Complex decisions based on human preferences will be coloured by the decision maker. Therefore, interpretations should be included, enabling other decision makers to understand and gain experience in the future. The results and their interpretation can be documented and reports can be generated. To summarize, the suggested approach is a comprehensive solution, closelyrelated to the decision process. In addition to introducing a support system for making trade-offs in a complex decision task, it enables the decision maker to make more efficient use of the trade-off analysis by supportingproblem analysis and model establishment. The system also assists the decision maker in generating proper documentation of both projects and results. [2] H. Simon: The Science of the Artificial The MIT Press, 1981. 131 G. Tangen; B. A. Foss: Decision Support for Planning Rehabilitation in Hydropower Production System XIII European Annual Conference on Human Decision Making, 1994, Finland, p.80-90. [4] M. Zeleny: Multiple Criteria Decision Making McGraw-Hill Book Company, 1982. [5] R. L. Keeney: Value Focused Thinking Harvard University Press, 1992. [6] D. L. Nazareth: Integrating MCDM and DSS W O R , 1993, V01.31, No.1 p.1-15. [7] E. U. Weber; 0. Coskunoglu: Descriptive and Prescriptive Models for Decision Making: Implications for the Development of Decision Aids IEEE Transactions on systems, Man and Cybemetics, 1990, p. 3 10-317. [8] C. C. White: A Survey on Integration of Decision Analysis for Decision Support IEEE Transactions On Systems, Man and Cybemetics, 1990, p. 358-364. [9] F. C. Schweppe; H. M. Merill: Multiple Objective Trade Off Analysis in Power System Planning 9th Power Systems Computation Conference (PSCC), 1987, p.14-21.

7. CONCLUSIONS
This paper has briefly described the present practice and shortcomings in hydropower plant reconditioning planning. A proposed concept for a decision suppport system aiming at more efficient project ranking, is referred.
To realize the concept an integration of knowledge based systems (Kl3S) and multi attribute decision making (MADM) is suggested to provide support in the decision model establishment as well as in the value trade-off. A joint approach will provide appropriate support at all stages in the decision process, because it utilizes the unique qualities of both KBS and MADM.

The work reported in the paper is part of a current research project aiming at making hydropower reconditioning planning more efficient. During 1995 MADM methods will be used to evaluate and rank reconditioning projects at hlAr hydropower plant, to assist the development of long term reconditioning plans. This paper is directed towards the further plans for the project which include specification of a computer system.
8. ACKNOWLEDGEMENTS

The research is funded by the Research Council of Norway, n association of the Norwegian Energy Corporation and a Norwegian hydropower producers (EFFEN).
9. REFERENCES

[I] G. Tangen; B. A. Foss: Decision Support for Planning Upgrading of a Hydropower Plant to be presented at 1995 IEEE Intemational Conference on Systems, Man and Cybemetics.

150

You might also like