Cross Entropy Based Composite System Reliability Evaluation By M-H Sampling And Computational Burden Prediction

Reliability evaluation can be used to analyze the risk level and find the weak parts of a power system by using probability theory,and gradually becomes an effective auxiliary tool for power system planning and operation,and plays an important role in ensuring the safe and reliable operation of power systems.When simulating and evaluating large-scale complex power system,Monte Carlo simulation has absolute superiority because its computational efficiency cannot be affected by the size of system,so it is widely used in the reliability evaluation of complex power system.However,the Monte Carlo simulation method faces the problem of slow convergence when dealing with highly-reliable power systems.It is a significant subject to improve the computational efficiency of the Monte Carlo simulation method.Cross entropy method can effectively accelerate the reliability evaluation of power systems.The classic cross entropy uses an iterative updateing algorithm based on importance sampling to update the parameters of importance sampling function.But for highly-reliable systems,the efficiency in generating the concerned samples for parameter updating may still be low.In addition,the convergence criterion of this iterative algorithm is usually determined subjectively,which may incur insufficient or excessive parameter updating and subsequent high computational burden of the entire simulation,i.e.pre-sampling stage and main-sampling stage.To address the above two problems,a CE method integrated with Metropolis-Hastings sampling(M-H sampling)and computational burden prediction is proposed in this paper,the main contents of this paper are as follows:(1)In the cross entropy optimization,the M-H sampling combined with subset simulation is adopted as an effective alternative to the importance sampling method.This method uses a fixed ratio of failure samples or a subset of near failure samples as the subset,and a triangular distribution is adopted as the proposed distribution of M-H sampling to obtain system state samples belonging to that subset,which can effectively improve the capturing efficiency of the desired samples in each iteration.The IEEE-RTS79 system with reduced load and IEEE-RTS79 system with correlated wind farms are used to analyze the results and demonstrate the performance of the proposed cross entropy method based on Metropolis-Hastings sampling(M-H sampling)and subset simulation method.Compared with the classic cross entropy method,our proposed method significantly increases the proportion of failure system states in the pre-sampling stage,speeds up the efficiency of parameter updating,and reduces the number of iterations,which can significantly accelerate the convergence speed of the cross entropy method.(2)A noval quantitative convergence criterion—minimum computational burden predication is proposed.The computational burden for the entire simulation is successively predicted and compared after each iteration and the optimal CE parameters achieving the minimum computational burden can be determined,the power flow calculation time and the optimal load shedding time are used as approximations of the computational burden.Fusion of M-H sampling and computational burden prediction can significantly improve the efficiency of cross-entropy parameter updating and minimize the computational burden.Reliability simulation tests are performed using three test systems: IEEE-RTS79 with reduced load,IEEE-RTS96,and IEEE-RTS96 with correlated wind farms.Compared with the classic cross entropy method and cross entropy method based on M-H sampling,the proposed method can significantly reduce the number of iterations required for cross-entropy iterative optimization,and ensure that the computational burden of the entire simulation process is minimum when using the optimal parameters with minimum computational burden predication,and the reliability evaluation speed can be efficiently accelerated.