Power System State Estimation Research Based On Deep Support Vector Machine And Unscented Kalman Particle Filter

As wind energy,solar energy,nuclear power,and a variety of smart devices are connected to the power system,the structure and operation of the power system are very complicated.The correct operation of intelligent equipment requires the power system to provide real-time status data quickly,accurately and comprehensively.Power system optimization scheduling,safe operation and monitoring,and online control are also very dependent on state estimation data of power systems.Therefore,it is of great significance to study state estimation that adapts to modern power characteristics.The measurement data of the power system state estimation is mainly provided by Wide Area Measurement System(WAMS)and Supervisory Control And Data Acquisition(SCADA).There are problems in time and phase angle matching,and there may be missing or bad data.In this paper,the research on the processing of mixed measurement data and the estimation method of power system state are carried out.The core work and innovative contributions are summarized as follows:Firstly,the measurement characteristics of SCADA and WAMS are introduced.The functions and sampling characteristics of SCADA and WAMS are analyzed.The fusion measurement of the two measurement data is carried out.In addition,this paper proposes a method based on the depth and time-time memory network bus load prediction method to obtain the pseudo-measurement data of the node.The pseudomeasurement data is integrated into the measurement data,which can supplement the missing data and replace the bad data to improve the accuracy of the state estimation.This part provides complete and reliable measurement data for the static estimation and dynamic estimation of the following text.Secondly,the static estimation study is carried out by using the mixed measurement data.The weighted least squares and least squares support vector machine(LS-SVM)methods existing in static estimation are studied.In the LS-SVM model,the parameters c and σ are parameter optimization through K-group cross-validation and grid search.The above two methods not only limit the search scope,but also the parameter optimization time is very long.In view of the above problems,this paper uses the improved particle swarm optimization algorithm to optimize the parameters,and optimizes the deviation b in the model to improve the estimation accuracy.Since SVM has only one hidden layer,performance is too dependent on the choice of kernel function and the ability of model to be characterized.Based on the parameter optimization,this paper proposes a state estimation method based on principal component analysis-least square depth support vector machine(PCA-LS-DSVM),which effectively improves the estimation accuracy.Then,dynamic estimation studies were performed using the mixed measurement data.The extended Kalman filter and unscented Kalman filter for dynamic estimation are weak in processing nonlinear models and the correctness of particle filter method is greatly affected by prior probability function.In this paper,a dynamic estimation method based on square root form of unscented Kalman particle filter is proposed.This method can not only deal with the nonlinear model well,but also ensure the correctness of particle filtering by using the unscented Kalman filter of square root form as the prior probability.In addition,in the particle filtering process,the method adds particle splitting and uses Markov chain Monte Carlo method to maintain the diversity of particles,which effectively solves the problem of particle starvation.Compared with other methods,this method performs better in estimating accuracy and stability performance.Finally,the proposed method is programmed on the TensorFlow platform and Matlab platform,and the simulation results are verified in the IEEE-14 and IEEE-30 transmission systems,which proves that the proposed method is compared with other methods.The accuracy is estimated to be higher.