STATE ADEQUACY EVALUATION USING GENERALIZED REGRESSION NEURAL NETWORK FOR NON-SEQUENTIAL MONTE CARLO SIMULATION BASED COMPOSITE POWER SYSTEM RELIABILITY ANALYSIS

This paper presents a new approach for state adequacy evaluation of sampled system state in composite power system reliability analysis. Generalized regression neural network (GRNN) is used in conjunction with non-sequential Monte Carlo simulation (MCS) to evaluate the loss of probability and the power indices. GRNN approach predicts the test functions for all the sampled states after sufficient training patterns are obtained in the initial MCS sampling with dc load flow based load curtailment minimization model. This model predicts the test functions for both success and failure states. The sampled system states are used to evaluate annualized system and load point indices. The indices evaluated are loss of load probability, loss of load expectation, expected demand not served and expected energy not supplied. The results obtained in this approach are compared with the conventional non-sequential MCS which uses load curtailment minimization model for state adequacy evaluation. An error analysis for different reliability levels is also carried out to check applicability of GRNN approach for calculating the test functions in reliability optimization, where several reliability levels are analyzed. The application of the proposed GRNN approach is illustrated through case studies carried out using RBTS and IEEE-RTS test systems and annualized indices are presented. It is found that the proposed approach estimates indices nearer to the conventional non-sequential MCS.