1204

IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 15, NO. 4, NOVEMBER 2000

Optimal Power Flow Solutions Under Variable Load Conditions

Katia C. Almeida and Roberto Salgado
AbstractThis work presents a methodology to calculate a sequence of Optimal Power Flow (OPF) solutions under variable load conditions. The aim is to obtain a set of optimal operating points in the neighborhood of the bounds of the region defined by the load flow equations and a set of operational limits. For this, an algorithm based on the Continuation method and on a PrimalDual Interior Point optimization method is proposed. Such algorithm consists of two main steps: the predictor step, which uses a linear approximation of the KarushKuhnTucker (KKT) conditions to estimate a new operating point for an increment in the system load; and the corrector step, which calculates the optimum corresponding to the new load level via a nonlinear PrimalDual Interior Point method. Indices for critical buses and inequality constraints are a byproduct of the methodology. In addition, sensitivity analysis is performed to calculate the amount of reactive compensation which allows for a pre-specified increase in the system load. Results for realistic test systems are presented. Index TermsContinuation method, nonlinear interior point method, optimal power flow.

I. INTRODUCTION WO FACTORS have added more complexity to the problem of determining adequate solutions for the electric power system steady state operation problem: a progressive increase of the load and the deregulation of the electric energy systems. Two research areas, therefore, are of great interest for power system analysis: the study of voltage collapse and OPF solutions. In particular, there is a need for methodologies which are able to simultaneously analyze these two aspects and to indicate the behavior of power systems, being operated with the help of OPF algorithms, near the maximum loadability limit. Different methodologies were proposed to calculate the maximum loadability limit of power systems. The approach presented in [1] proposes the determination of this limit through the computation of the steady state multiple solutions. In reference [2], sensitivity relationships between the power system variables are used to calculate the critical load. The Singular Value Decomposition of the conventional NewtonRaphson Jacobian matrix was also applied [3]. The parameterization of the steady state power system equations was also used to formulate the problem of maximum loadability [4], [5]. These two last works applied the continuation method to track the load flow solution for an increasing system demand.
Manuscript received April 23, 1999; revised December 20, 1999. This work was supported by the Conselho Nacional de Desenvolvimento Cientfico e Tecnolgico (CNPq), Brazil. The authors are with GSP/LABSPOT, Departimento de Engenharia Eltrica, Universidade Federal de Santa Catarina, CP 476 - 88040-900 - Florianpolis SC, Brazil (e-mail: katia@labspot.ufsc.br; salgado@labspot.ufsc.br). Publisher Item Identifier S 0885-8950(00)10349-9.

On the other hand, OPF algorithms have been proposed since the middle of the sixties and are the basic tool to assess the economic aspects of the power system operation. Some of these algorithms apply parametric optimization techniques, providing methodologies based on different versions of the Continuation method [6][12]. Some of these methodologies are based on the Newton OPF method [13]. The results have shown that the adequate combination of the Continuation methods with the optimization algorithms can provide numerical tools of high potential for power system studies. It allows the development of robust methods for the solution of the OPF problem and the differentiation of the various critical conditions where most algorithms fail to find the solution [14]. Recently, the performance of Interior Point (IP) algorithms in solving linear programming problems has led to many applications of these algorithms to the nonlinear OPF problem [15][18]. The efficiency of finding the optimal solution and the effective way of handling the inequality constraints have been claimed as its main features. Some of these works proposed the use of an OPF algorithm to compute the point of maximum loadability of the power systems via nonlinear versions of Interior Points methods [17], [18]. The use of optimization algorithms for the study of heavily loaded systems allows the representation of all the operational limits and, depending on the OPF formulation, the adoption of an criterion to be optimized [14], [17], [18]. The not well known steady state behavior of power systems being optimally operated under heavy load can be, in this way, better analyzed. This work proposes a methodology to analyze the behavior of the solution of the power system equations close to the point of maximum loadability. This methodology combines the Continuation method with a nonlinear version of the Interior Point algorithm. The first provides a sequence of estimates for the solution of the KarushKuhnTucker (KKT) conditions from a base case to the point of maximum loadability. Each solution of this sequence is determined through the OPF Interior Point algorithm. This combination allows the optimal tracking of the load growth, even in the neighborhood of the feasibility limit, where the Newtons solver is bound to diverge due to the ill-conditioning of the Jacobian of the KKT conditions [14]. Good indicators for critical buses and operational limits can be straightforwardly obtained. In addition, sensitivity studies provide means to calculate the increase in var injections which allow for a specific increase on the total demand. This work is organized as follows. In Section II the maximum loadability problem is discussed. Section III presents the proposed algorithm and discusses the predictor and corrector steps. Section IV presents some additional features of the approach.

08858950/00$10.00 2000 IEEE

ALMEIDA AND SALGADO: OPTIMAL POWER FLOW SOLUTIONS UNDER VARIABLE LOAD CONDITIONS

1205

Numerical results are presented in Section V. Finally, in Section VI, some conclusions regarding the numerical results are drawn. II. THE MAXIMUM LOADABILITY PROBLEM The resolution of the maximum loadability problem yields the maximum real and reactive power demand that a power system can withstand, while operating at a stable point (i.e., one which does not change considerably for small increments on the systems parameters such as load or operational limits), that respects a set of pre-defined operational limits. A steady state formulation of this problem can be made in terms of the load flow system of equations. The parameterization of the bus loads yields a modified set of power balance equations, in which the load increase direction is explicitly represented: (1)

which causes an additional difficulty in the tracking of the solution of up to the maximum value of . Thus the analysis of the OPF behavior near the maximum loadability limit must be done with algorithms which can diminish the problem of ill conditioning observed near such limit. This is the main motivation of the present work. III. THE PROPOSED APPROACH The application of the Interior Point algorithms to solve consists basically of: a) converting the inequality problem constraints in equality constraints, through nonnegative slack variables; and b) adding a logarithmic barrier function to the objective function, to preserve the nonnegativity condition of the slack variables. The modified parameterized optimization is: problem, Min (6)

where is the load parameter is the set of power flow equations and is the pre-specified load increase direction. In subject to this case, the calculation of the maximum loadability of power systems consists of solving (1) to find the complex bus voltages corresponding to the maximum value of . For systems being optimally operated, the maximum loadability problem can be stated as the problem of finding the max: imum value of for which problem Min subject to (3) (4) has feasible solutions. , the vector of decision variables, , is composed In of the active power generations, bus voltage magnitudes and angles, transformer tap settings and phase shifter angles. The , can represent the power generation objective function, cost, the transmission losses, the voltage deviation from a pre-specified voltage level or any combination of these three , which indices. The set of inequality constraints, comprises the upper and lower limits of the decision variables and functional inequalities such as the limits on the generated reactive power and line flows, can also be dependent on the bus loads: (5) where represents a pre-specified load increase direction. can be tracked for increasing until The solution the maximum loadability limit is reached. The difficulties to solve this nonlinear optimization problem are well known, and presently most of the algorithms which were successful in its resolution are based on the solution of the its pure or modified KKT conditions by linear approximations (Newton method). However, it can be shown that near the feasibility limit the is ill-conditioned [14] Jacobian of the KKT conditions of (2) subject to (7) (8) where is the logarithmic barrier parameter is the vector of slack variables and is the number of inequality constraints. The interior point OPF model (6)(8) is, a parameterized and . The proposed model with two distinct parameters methodology consists of varying each of these parameters at a time: in the predictor step, is increased so that a new load level is considered; in the corrector step, is decreased so that, at the end of the correctors iterations, the original OPF problem is solved. Since we are interested in analyzing the behavior of the OPF solutions for increasing , the continuation process will be explained below only in terms of this last parameter, while the optimal solution will be tracked for varying . Nevertheless, Interior Point methods can also be interpreted as a special class of parametric optimization methods [19]. For an interval of variation of where is a regular point [20], the first order optimality conditions for problem (6)(8) are (9)

(10) (11) (12) where and are the Lagrange multipliers associated with the equality and inequality constraints, respectively, and is a diagonal matrix composed of the elements of .

1206

IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 15, NO. 4, NOVEMBER 2000

As the parameter varies, a new OPF problem and a new set of optimality constraints are defined. Let and be a point which satisfies the optimality conditions at a given . The system of (9)(12), which connects the OPF problems defined for every value of , is the so called homotopy and , which characterizes the function, defined for solution trajectories of the OPF problem for variations in . To track the OPF solutions, it is assumed that this homotopy function is continuous and differentiable. This supposition is valid under a set of conditions which cannot be assured for the OPF problem for all values of [14], [20]. In particular, such regularity conditions are not valid at the maximum load limit (i.e., a feasibility limit) of the OPF problem, which leads to a poor convergence of OPF algorithms in the neighborhood of such critical condition. The approach presented here is intended to improve the convergence in such critical conditions. . Such system is Let the system (9)(12) be denoted as a generalization of the one composed of the parameterized load flow equations which is used to track the load flow solutions until the maximum loadability. The approach proposed here is , for and , on the based on the tracking of and intervals of variation of where the solution candidates are regular points. The approach is based on a predictorcorrector scheme [20] where the limits on and are respected. A. The Predictor Step Suppose that the tracking of the OPF solutions starts at an is given to , optimum, defined for . If an increment, , can be a prediction of the solution for obtained by taking a step of appropriate size in the direction of . Taking the linear approximation of on the tangent of , yields:

Since must be solved for and , suitable and , should be calculated for the primal and step sizes, dual variables, respectively, as (17) and (18) Thus, the predicted point is given by (19) and (20) where is used to assure a predicted point which is interior to the set of inequalities (4). The predicted point is an approximate solution. The exact , with and solution is found by solving the set , taking as initial solution. Since the number of equations is smaller than the number of variables, the value is specified equal to the of a component of vector predicted value. The choice of the component to be specified is based on the solution of (16). The component associated with the largest increment is set to the predicted value. The predicted point satisfies the inequality constraints of , but not the equality constraints. In other words, is which is interior to the set of inequality constraints of not far from the solution of the first order optimality conditions, since it was obtained from a linear approximation. Thus, the predicted point is a good initialization for the PrimalDual Interior Point algorithm which is used to obtain the new . optimum B. The Corrector Step with one additional constraint In the corrector step, specifying the th component of vector , is solved through a PrimalDual Interior Point algorithm. Thus, the optimality conare increased by one equation, that is, ditions for (21)

(13) satisfies the first order optimality Therefore, since is such that conditions, the tangent vector (14) System (14) has more variables than equations. To solve it, we impose a nonzero value to one of these variables: (15) System (14) therefore takes a new form:

The use of the additional equation in to parameterize the optimality conditions improves the convergence features as well as the numerical conditioning of the iterative process. , the set of nonlinear (21) is iterStarting at atively solved by Newtons method. Supposing that the conare linear on , at each iteration, the following straints of linear system must be solved: (22)

(16) where is a line vector with all components equal to zero except the th component, which is equal to 1.

(23)

ALMEIDA AND SALGADO: OPTIMAL POWER FLOW SOLUTIONS UNDER VARIABLE LOAD CONDITIONS

1207

whereas, the critical Lagrange multiplier (among those related to the inequality limits), is such that (30) The solution of (29), for the voltage magnitudes only, and of (30), indicates, respectively, the critical bus voltage magnitude and the critical inequality limit at that level of the system load. B. Sensitivity AnalysisReactive Power Injections (24) (25) (26) where, (27) (31) (28) where is a diagonal matrix composed of the elements of , and is the identity matrix. The nonlinear PrimalDual version of the Interior Point aland gorithm solves the set of (22)(26) iteratively, with , for decreasing values of the barrier parameter, . Thus, the solution of is obtained [15]. when Fig. 1 depicts the predictor–corrector algorithm. In this figure, is the tangent vector obtained from the linear apat the current solution , is proximation of is the corrected point. the predicted point and IV. ADDITIONAL STUDIES The parameterized optimization model, being a generalization of the parameterized load flow equations used for voltage collapse studies, provides similar additional information regarding the behavior of the system near the collapse point. The main differences are: i) all the conclusions are drawn with respect to an optimal operation point and ii) the influence of the operational limits are being considered. This last point has some important consequences on the nature of the index of the maximum variation on the voltages and on some sensitivities that can be calculated with the parameterized model. When limits are considered, the optimal solution trajectories vary continuously with only in those intervals where no new limit becomes active [10]; a break-point appears upon the activation of a new inequality constraint. As a consequence, indices based on the tangent vector and also some sensitivities which are a byproduct of the approach, are valid only for small intervals of variation of where no new limit is reached. A. Information Contained in the Tangent Vector Using the tangent vector, , it is possible to obtain the decision variable or Lagrange multiplier with the largest increment for a . The critical decision variable, is such that given (29) (32) and and bility limit. Also, are obtained from at the original feasiThe optimization model can provide approximate values of the increment on the reactive power injections necessary to allow a given increment on the maximum load. The reactive power limits are part of the functional inequalities (8). Therefore, the increment in corresponding to an increment on the maximum inequality can be obtained from

Fig. 1. PredictorCorrector algorithm.

(33) Substituting the expressions of and from (9) and (10), respectively, into (33), after some simplifications, we have (34) can be obtained by solving the folThe value of lowing system of equations (35) , are obtained directly from in which, all terms, but (9)(12). Therefore, using (32), (34) and (35), we can solve (31) which will allow a pre-specified and find the increment on increment on the system load, . V. TEST RESULTS The results presented here are divided into three categories: i) study the OPF behavior near the loadability limit; ii) analysis of the efficiency of the proposed methodology and iii) analysis of the critical bus indices and the sensitivity of the maximum load with respect to reactive power injections. A 6-bus example [21] and equivalents of two Brazilian systems were used: the first, of the Mato Grosso do Sul State system, with 127 buses and

1208

IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 15, NO. 4, NOVEMBER 2000

Fig. 2. Load flow variables for increasing load.

Fig. 4.

Voltage mag.176 bus system.

Fig. 3. OPF variables for increasing load.

Fig. 5.

Active and reactive gen.176 bus system.

total load equal to 1168 MW and 303 Mvar, in an application to minimize the total transmission losses; and the second, with 176 buses and total load equal to 2528 MW and 477 Mvar, of the Brazilian Southern and Southeastern regional system, in an application to minimize the total generation cost. A. Behavior of OPF Solutions Near the Maximum Loadability Limit The 6-bus example was used to compare the load flow results with the OPF results in an application to minimize generation , the load flow solution was set equal to cost. For this, at the OPF solution. For an increasing , the voltage magnitudes and active power generation obtained from (1), are depicted in Fig. 2. Fig. 3 shows the optimal trajectories of the same variables, as obtained by the OPF algorithm. The first noticeable difference is the maximum obtained in each case. For the load is allowable, thus indicating flow problem, a much larger the influence of the operational limits on the maximum loadability. The second difference is on the behavior of the variables. Whereas, in the load flow, all load bus voltages drop as increases, in the OPF, most of the voltages increase with the load, thus indicating the effort of the system to transmit more power. Also, it can be observed that one generator needs to decrease its real power output due to the limits imposed by the OPF. Figs. 4 and 5 depict the optimal trajectories of some voltage magnitudes and active and reactive power generations for the 176-bus system, near the loadability limit. In Fig. 4, the fullline and dashed-line curves correspond to voltage magnitudes at some load buses, and some generation buses of the system, respectively. In Fig. 5, the full-line curves represent some active

Fig. 6. Voltage magnitudes127 bus system.

power generations whereas the dashed-line curves represent the reactive power generations. It can be seen that due to the OPF constraints, there can exist, for an increase in the system load, a sharp increment or decrement of the load voltages, while some of the generator voltages increase and others remain almost constant. In addition, abrupt changes on the voltage magnitudes can be observed due to changes in the active constraint set. An interesting behavior can also be observed for the real and reactive power generations: although the system real and reactive loads are increasing, some of these quantities are decreasing. Figs. 6 and 7 depict some optimal trajectories for the 127 bus system. In Fig. 6, the full-line curves correspond to generation buses whereas the dotted-line curves correspond to some load buses. Near the loadability limit, the voltage magnitudes of the load buses and of some generation buses are sharply increasing so as to increase the transfer capability, whereas, due

ALMEIDA AND SALGADO: OPTIMAL POWER FLOW SOLUTIONS UNDER VARIABLE LOAD CONDITIONS

1209

TABLE III CRITICAL VOLTAGE INDICES

TABLE IV CRITICAL LAGRANGE MULTIPLIER INDICES

Fig. 7. Transformer tap ratios127 bus system. TABLE I CONDITION NUMBERS

TABLE II AVERAGE NUMBER OF IP ITERATIONS

to some reactive limits, two generation bus voltage magnitudes are decreasing. Also, in Fig. 7 one can notice that most of the transformer tap ratios are decreasing. The existence of operational limits and of a minimization criterion, and the fact that, no generation voltage magnitudes were fixed accounted for the behavior presented in Figs. 6 and 7. The results presented show that power systems operated according to some economic criteria near the maximum loadability limit can present quite a different behavior from the one observed in a steady state operation considering only feasibility requirements. B. Efficiency of the Proposed Approach The algorithm was implemented using MATLAB, Version 4.2c. The following results emphasize the need of a special solution methodology for OPF algorithms to be used in heavily loaded systems. The results show the efficiency of the predictorcorrector scheme described in Section III. Table I compares the condition number of the augmented which corresponds to the linmatrix of (16) to that of earization of the KKT equations for the primaldual interior point method. The condition numbers (CN) obtained at the initial loading condition were the same while, at the maximum load the augmented matrices were better conditioned. Table II shows the average number of corrector steps during the tracking of an increasing . The convergence of the iterative process is affected by the choice of the initial value of the

barrier parameter, by the selection of the load increment and by the specification of the continuation, variable. The first has direct influence on the optimization process of the corrector step. In the performed tests, an initial barrier parameter equal to 0.2 was found to yield good convergence. The choice of the load increment is reflected in the number of correction steps. A small value of this parameter results in a large number of correction and comsteps. However, this value can be controlled by puted during the prediction step. The initial increment on the load, at every predictor step, was set to 2% of the total load. The specification of the continuation variable is based on the estimates obtained in the predictor step. Since these estimates are computed taking into account the inequality constraints, this has a minor effect on the convergence of the iterative process of the corrector stage. The number of corrector steps tends to increase with the size of the predictor step. However, contrarily to what occurs in a previous implementation [10], here, due to the parameterization, and to the fact that, near the maximum loadability limit, the Lagrange mulipliers tend to infinity, there will usually be very small increments on the system load near this limit. Therefore, the number of corrector steps tends to decrease. The maximum loadability of the test systems was also determined by the approach proposed in [17], [18]. The results obtained for the 176-bus system were similar to the ones obtained by the continuation OPF. Differences were noticed with respect to the active power generation, which were attributed to the use of the performance index. For the 127-bus system, however, the analysis of the results revealed some differences regarding the critical loadability, reactive power and voltage magnitudes. The continuation OPF indicated a maximum load 8% higher than the one obtained by the other approach. Also, due to the optimization criterion, the bus voltage magnitudes and, accordingly, the amount of injected reactive power were considerably higher. At first sight, these differences could be attributed to the performance index and the methodology used. Additional studies are envisaged in order better analyze these results. C. Sensitivity Analysis Tables III and IV indicate the values of the critical voltages and Lagrange multiplier indices [(29) and (30)], whereas Figs. 8 and 9 depict the optimal trajectories (the dashed lines correspond to the most critical values). It can be noticed that the indices correctly indicate the most critical values. Due to

1210

IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 15, NO. 4, NOVEMBER 2000

VI. CONCLUSION The paper presented a parameterized OPF algorithm which is able to track the system load variation for a specific range of the load parameter, . This algorithm is based on the continuation method an on a primaldual interior point method. The adopted parameterization allows the resolution of the OPF problem for critical loading conditions, providing some insight onhe behavior of power systems being optimally operated near a feasibility limit. The approach provides critical variables and operational indices and can also be used in sensitivity studies to obtain reactive support which allows for a pre-specified load increase. The results indicated that the behavior of power systems near the maximum loadability limit can be considerably different from the one reported by numerous studies based on the load flow equations. With the introduction of market competition in the generation and transmission systems, power dispatch will be made following some cost or performance criteria. This paper shows that costs and performance concerns can considerably modify the steady state behavior of power systems.

Fig. 8. Critical Lagrange multipliers176 bus sys.

REFERENCES
[1] N. Yorino, S. Harade, and H. Sasaki, Use of multiple load flow solutions to approximate a closest loadability limit, in Proc. of the Bulk Power System Voltage Phenomena, Fairfax, VA, Aug. 1991. [2] N. Flatabo, R. Ognedal, and T. Carlsen, Voltage stability condition in power transmission system calculated by sensitivity methods, IEEE Trans. on Power Systems, vol. 5, no. 4, pp. 12861293, Nov. 1990. [3] A. Tiranuchit, L. M. Ewerbring, R. A. Duryea, R. J. Thomas, and F. T. Luk, Toward a computationally feasible on-line voltage stability index, IEEE Trans. on Power Systems, vol. 3, no. 2, pp. 669675, May 1988. [4] K. Iba, H. Sasaki, M. Egawa, and T. Watanabe, Calculation of critical loading condition with nose curve using homotopy continuation method, IEEE Trans. on Power Systems, vol. 6, no. 2, pp. 548593, May 1991. [5] V. Ajjarapu and C. Christy, The continuation power flow: A tool for steady-state voltage stability analysis, IEEE Trans. on Power Systems, vol. 7, no. 1, pp. 416423, Feb. 1992. [6] A. Fahmideh-Vojdani and F. D. Galiana, The continuation method and its application in system planning and operation, in Proc. of CIGR, Florence, Italy, 1983, paper 102-04. [7] G. Blanchon, J. C. Dodu, and A. Merlin, Developing a new tool for real time control in order to coordinate the regulation of reactive power and voltage schedule in large scale EHV power systems, in Proc. of CIGR, Florence, Italy, 1983, paper 209-01. [8] M. Innorta and P. Maranino, Advance dispatch procedures for the centralized control of real power, in Proc. of the Power Industry Computer Application Conference, Montreal, Canada, 1987, pp. 407413. [9] M. Huneault and F. D. Galiana, An investigation of the solution to the optimal power flow problem incorporating continuation methods, IEEE Trans. on Power Systems, vol. 5, no. 1, pp. 103110, Feb. 1990. [10] K. C. Almeida, F. D. Galiana, and S. Soares, A general parametric optimal power flow, in Proc. of the Power Industry Computer Application Conference, Phoenix, May 1993, pp. 6673. [11] V. Ajjarapu and N. Jain, Optimal continuation power flow, Electrical Power Systems Research, vol. 35, pp. 1724, 1995. [12] F. G. M. Lima, S. Soares, A. Santos Jr., K. C. Almeida, and F. D. Galiana, Optimal power flow based on a nonmlinear parametric approach, in The 13th Power Systems Computation Conference, Norway, 1999. [13] D. I. Sun, B. Ashley, B. Brewer, A. Hughes, and W. F. Tinney, Optimal power flow by Newton method, IEEE Trans. on Power Apparatus and Systems, vol. 103, no. 10, pp. 28642875, Oct. 1984. [14] K. C. Almeida and F. D. Galiana, Critical cases in the optimal power flow, IEEE Trans. on Power Systems, vol. 11, no. 3, pp. 5091518, August 1996.

Fig. 9. Critical voltage magnitudes176 bus sys.

TABLE V SENSITIVITY ANALYSISREACTIVE GENERATION LIMITS

the inequality constraints, this indication is valid only in the neighborhood of the point where they were obtained. Some studies were made so as to calculate the increase on the reactive generation limits necessary to permit a pre-specified increment on the maximum load delivered by each test system. The results are summarized in Table V. From this table, it can be noticed that the increments obtained from (31), for the gen, indeed allow an increment on the total erators at buses and ) which is close to the predicted value, load ( . It is important to emphasize that sensitivity relationships computed from the OPF are first order information and therefore are valid only in the neighborhood of the OPF solution. The only way of computing the exact value of the variables due to modification in the power system is through a nonlinear OPF.

ALMEIDA AND SALGADO: OPTIMAL POWER FLOW SOLUTIONS UNDER VARIABLE LOAD CONDITIONS

1211

[15] S. Granville, Optimal reactive dispatch through interior points methods, IEEE Trans. on Power Systems, vol. 9, no. 1, pp. 136146, Feb. 1994. [16] Y. Wu, A. Debs, and R. Marsten, Direct nonlinear interior primaldual predictor–corrector point method for optimal power flow, in Proc. of the Power Industry Computer Application Conference, Phoenix, USA, May 1993, pp. 138145. [17] G. D. Irisarri, X. Wang, J. Tong, and S. Mokhatari, Maximum loadability of power systems using interior point nonlinear optimization method, IEEE Trans. on Power Systems, vol. 12, no. 1, pp. 162172, Feb. 1997. [18] L. V. Barbosa, K. C. Almeida, and R. Salgado, The interior point approach in loadability limit studies, in Proc. of the Symposium of Specialists on Operation and Planning and Expansion of Power Systems, Salvador, Brazil, May 1998. [19] B. Kummer, Parameterizations of Kojimas system and relations to penalty and barrier functions, Mathematical Programming, vol. 76, pp. 579592, 1997. [20] J. Guddat, F. Guerra-Vazquez, and H. Th. Jongen, Parametric Optimization: Singularities, Pathfollowing and Jumps: John Wiley & Sons, 1990.

[21] M. A. Pai, Computer Techiniques in Power System Analysis: McGraw Hill, 1980.

Katia C. Almeida received her Ph.D. in electrical engineering in 1995 from McGill University and is currently a Lecturer in the Electrical Engineering Department of the Federal University of Santa Catarina, (UFSC), Brazil. Her research interests are in power system optimization and deregulation.

Roberto Salgado received his Ph.D. in electrical engineering in 1989 from the University of Manchester. Since 1978 he has been a Lecturer in the Electrical Engineering Department of the Federal University of Santa Catarina (UFSC). His main research interests are in voltage stability analysis and optimal power flow.