Research On Fast Monte Carlo Simulation Methods And Risk Grade Asssessment For Composite Power System Reliability

Composite power system reliability evaluation can effectively identify system risk and weak points,thus can provide important reference for the decision-making of power system planning and operation,but its computational complexity is high.Monte Carlo Simulation(MCS)is a powerful tool to evaluate composite power system reliability: sequential MCS can flexibly model the chronological correlation of system state,but it suffers from slow convergence;non-sequential MCS can conduct probabilistic modeling and sampling on spatial correlation,but this sampling undermines simulation efficiency.Besides,in industrial applications,composite power system risk grade assessment can provide early warnings about system risk grade for planners and operators,but the requirement for evaluating system reliability repetitively in several load scenarios makes it time-consuming.Thus,research on fast sequential and non-sequential MCS based on chronological and spatial correlation modeling,and on using these fast MCS in composite power system risk grading is of great academic and practical value.This dissertation studies in depth fast MCS based on chronological and spatial correlation modeling,and the use of these fast MCS in assessing composite power system risk grade.The main contributions are as follows:With respect to the low efficiency problem of sequential MCS(especially when evaluating system reliability under several load scenarios),this dissertation proposes a fast sequential MCS(FSMCS)based on the simplified chronological correlation model of system component state.Based on the conditional kernel density estimation for the properties of system component state(consisting of all component states),the proposed method establishes the simplified chronological correlation model of system component state and uses it for fast system reliability evaluation under different load scenarios.The simplified model is utilized to obtain the chronological sequence of system component state,then the sequence is combined with system load chronological model to form system state chronological sequence,so the chronological correlation of system state is modeled to make system reliability indices and their probability distributions accurately estimated;as the load curtailment of each system state can be easily obtained by using the properties of system component state,the optimal load curtailment calculations can be avoided to enhance simulation efficiency.Case studies of RBTS,IEEE-RTS79 and Modified IEEE-RTS79 with wind farms verify the effectiveness of FSMCS.With respect to the low efficiency problem of non-sequential MCS in incorporating spatially correlated continuous variables(e.g.bus loads and the wind speeds of wind farms),this dissertation proposes a fast non-sequential MCS based on normal probabilistic space transformation and its distribution parameter cross entropy optimization.Parametric models require assumptions on probability distribution type,thus leading to poor modeling accuracy for continuous variables which do not satisfy the assumption.Thus,this dissertation uses data-driven Gaussian mixture model(GMM)to model the joint PDF of correlated continuous variables,and uses EM method to estimate the parameters of GMM.In power system reliability assessment,Nataf transformation has been utilized for the direct sampling of correlated continuous variables.Unlike previous studies,this paper uses Nataf transformation for the cross entropy importance sampling of correlated continuous variables to enhance simulation efficiency: the GMM modeled correlated continuous variables are converted to correlated standard normal variables via Nataf transformation,then cross entropy parameter optimization and importance sampling are conducted on the normal variables;the obtained samples of the normal variables are converted back to the space of GMM to obtain the samples of the original correlated continuous variables,so that cross entropy importance sampling is indirectly realized on the original correlated continuous variables.Case studies of RBTS and IEEE-RTS79 verify that the proposed method is accurate in modeling and efficient in sampling correlated continuous variables.With respect to the problem that the former fast non-sequential MCS is only applicable to normal copula correlated continuous variables,this dissertation proposes a fast non-sequential MCS based on direct cross entropy optimization for the parameters of GMM.The proposed method can conduct probabilistic modeling and cross entropy importance sampling on continuous variables with any correlation type under a unified framework of GMM and EM algorithm: GMM is utilized to model the original joint PDF and the Importance Sampling Probability Density Function(IS-PDF)of correlated continuous variables,and EM algorithm is utilized to estimate the parameters of the original joint PDF,and conduct the cross entropy parameter optimization of IS-PDF.Compared with the traditional unimodal IS-PDF utilized by continuous variables,the GMM modeled IS-PDF used here can more flexibly capture the complex multimodal shape which the optimal IS-PDF may have,thus enhancing the sampling efficiency of correlated continuous variables.Case studies of IEEE-RTS79 and Modified IEEE-RTS79 with wind farms verify that the proposed method is accurate in modeling correlated continuous variables and efficient in sampling them.With respect to the problem that composite power system reliability indices can quantify system risk but cannot reflect system risk grade,this dissertation proposes a novel method to assess composite power system risk grade based on the relationship between system risk and load,and uses the proposed fast MCS to improve the efficiency of estimating system reliability indices in risk grade assessment.The proposed risk grade assessing method defines system risk grades based on the changing process of the system performance under N-1 contingencies,with the increase of system load.Under system reference operational mode,the risk grade assessing method uses fast MCS to calculate system severity indices in several load scenarios to establish system risk grading standard.With the system risk grading standard established under system reference operational mode used as a benchmark,system severity index under any other operational mode is converted to the system risk grade under system reference operational mode.After such “risk conversion”,system risk grades in different operational modes can reflect the impact of system operational mode change on system risk grade,thus can provide useful reference information for system operational planning.Case studies of Modified RBTS and Modified IEEE-RTS79 verify the effectiveness of the proposed method.