Reliability Evaluation Of Bulk Power System Based On Dimension Reduction Based Disaggregation Model And Cross Entropy

With the continuous development of social economy,the power system presents a large-scale,high-interconnect tendency.Reliability assessment can be used to analyze the power supply capability and weakness of power system,and has gradually become an effective tool for planning and operation.There are plenty of uncertainties during the operation of power system,such as wind speed and bus loads.How to accurately model the multivariate dependence has become an important concern in power system reliability evaluation.On the other hand,the power grid has the characteristics of large scale,strong randomness.It is a high-dimensional nonlinear complex system,and its computational complexity is very high.How to maintain the reliability evaluation efficiency of large power grids on the premise of ensuring the calculation accuracy has become an urgent problem to be solved.In response to the above two issues,the thesis carried out the following targeted research work:(1)The thesis innovatively proposed a non-parametric probabilistic dimension reduction based disaggregation model for wind power output and node load changes.In the modeling of high-dimensional correlation probability models,the existing literature not only ignores aggregation constraint in high-dimensional joint probability density estimation,but also faces the problem of the curse of dimensionality when the dimension is too high.This paper is based on the non-parametric probabilistic disaggregation technology.The core idea is to realize the probabilistic decomposition of aggregated variable(such as system load)to non-aggregated variables(such as bus load)based on lumped constraints.It innovatively proposes a hierarchical dimensionality reduction method that combines high-dimensional conditional probability.The density estimation problem is novelly transformed into multiple low-dimensional conditional probability density estimation problems,which effectively alleviates the problem of dimensional disasters.In the sampling of the high-dimensional correlation probability model,a three-stage conditional sampling method is further proposed for the dimensionality reduction model.Thus,from modeling to sampling,a complete non-parametric probabilistic dimension reduction based disaggregation model is formed,and this model is used in the reliability assessment of large power grids with high-dimensional random variables related to large power grids,to improve the IEEE-RTS79 and Taking the IEEE-RTS96 system as an example,the results show that the model is correct and effective.This model is not only applicable to the reliability evaluation of large power grids with high-dimensional correlation random variables,but also applicable to other areas of grid probabilistic analysis applications with high-dimensional correlation modeling.(2)The thesis carried out a cross-entropy important sampling study of the non-parametric probabilistic dimension reduction based disaggregation model to effectively improve the reliability evaluation efficiency of large power grids with high-dimensional correlation random variables.Because the sparsity of failure events has an important impact on Monte Carlo simulation,changing the probability distribution of random variables based on important sampling techniques to highlight the occurrence probability of failure time can significantly improve Monte Carlo simulation efficiency.The Monte Carlo simulation method based on cross entropy is an efficient and important sampling method proposed in recent years,but the current research focuses on the cross entropy optimization of the probability density distribution parameters of discrete random variables and one-dimensional continuous random variables.The optimization of the probability density distribution parameters of variables needs to be further explored.Although the non-parametric probabilistic dimensionality reduction set model can accurately model the correlation between high-dimensional random variables,for high reliability systems,the reliability evaluation of large power grids based on the non-parametric probabilistic dimensionality reduction set model has low sampling efficiency..Therefore,in this paper,cross-entropy optimization theory and non-parametric probabilistic dimensionality reduction set model are organically integrated,and cross-entropy parameter optimization is performed on the probability density distribution parameters of lumped variables.,Indirectly realizing the important sampling of high-dimensional random variables.The accuracy and efficiency of this method have been verified in reliability test systems such as improved IEEE-RTS79 and IEEE-RTS96 with wind farms.