A Cross-Entropy-Based Three-Stage Sequential Importance Sampling for Composite Power System Short-Term Reliability Evaluation

摘要

Regarding short-term reliability of composite power system, probability of critical event resulting in system failure within a short lead time is extremely low, which renders classical sequential Monte Carlo simulation method inefficient. In this paper, a cross-entropy-based three-stage sequential importance sampling (TSSIS) method is proposed to solve the low efficiency problem resulted from the low rate of component state transition during a fixed lead time. First, by assuming the system state transition process conforms to continuous time Markov chain, an analytical solution to optimal distorted component state transition rate to be used for sequential importance sampling is found by means of cross-entropy method. Second, TSSIS for a fixed lead time is constructed as follows: 1) acceleration of producing system state transitions; 2) enhanced learning to give optimal distorted transition rate; 3) compensation to the cost function. Case studies based on a reinforced Roy Billinton reliability test system and RTS-79 are carried out respectively for illustration of parameter settings of TSSIS as well as efficiency gain in comparison with the classical sequential Monte Carlo simulation method. The results demonstrate that given rational setting of parameters, TSSIS is of relatively high efficiency for sequential short-term reliability evaluation of composite power system.