Research On Robust Forecasting-Aided State Estimation For Power System

Power system state estimation is an important technical foundation for power system control,dispatch and safety assessment.It is also an indispensable core component of the energy management system.Therefore,work in power system state estimation is particularly important.The following issues need to be addressed in power system state estimation:Firstly,the sudden change in the operation mode of the power system may have an adverse impact on the power grid.First,the sudden change in the operation mode of the power system may cause abnormal changes in the frequency,voltage and other parameters of the power system,leading to the instability of the power system.Secondly,sudden shutdown or failure can cause damage or even destruction of the generator set.Finally,an emergency may trigger automatic protection of the power system,resulting in widespread blackouts.Secondly,the existence of abnormal data in power system will affect the performance of state estimation.The existing robust methods cannot suppress,detect and distinguish all abnormal data.Abnormal data may interfere with the stability and reliability of the power system.If abnormal data is mistaken for normal data,it may lead to unsafe operation or wrong decision-making of the power system.Thirdly,improving accuracy and robustness is an important research topic in power system state estimation.However,due to the complex layout of the grid and the lax management operation,bad data are generated in the measurement system,which increases the uncertainty of the state estimation and poses a great challenge to the robustness of the state estimation.This paper proposes the following solutions to the above three problems of state estimation:Firstly,the operation mode of power system may be affected by a variety of factors,which result in a diversity of power system data sets in terms of characteristics.This article uses convolutional neural networks(CNN)to train and extract features between these datasets,effectively distinguishing different operating modes,in order to achieve more accurate detection and provide better real-time monitoring and intelligent management for power systems.Secondly,this article applies support vector machines to various types of system data classification,which can effectively distinguish the actual state of the system.To address the issue of different penalty factors and kernel functions affecting classification accuracy,different optimization methods were used to optimize the penalty factors and kernel functions,thereby improving classification accuracy.Further combining t-SNE to achieve the purpose of visual classification.Thirdly,aiming at a variety of power system M estimation methods,a stable and reliable power system network model is adopted,and a reasonable comprehensive performance index is established to evaluate the performance of different state estimators considering the influence of the change of Gaussian model and bad data on the estimation results.In this paper,the basic principle of power system state estimation and WLS,LAV and Huber state estimation methods are introduced.A GM estimator combined with improved PS and Huber convex fraction function is proposed.The GM estimator is processed by iterative reweighted least squares algorithm.Numerical comparison of power reference system and DGs integration demonstrates the efficiency and robustness of the proposed method.In this paper,the above methods are verified by MATLAB simulation,and the simulation results on different bus test systems prove the effectiveness and accuracy of the proposed method.