Research On Detection And Correction Of Power Load Abnormal Data Based On Data Mining

With the establishment of intelligent grid automatic information platform,the scale,type and structure of load data in power system have changed greatly.In the actual operation of power grid,the quality of power load data has a decisive impact on the accuracy of load forecasting and the stability of power grid operation.In order to ensure the accuracy and integrity of load data,it is necessary to detect and correct abnormal data in power load data.However,the existing detection and correction method of abnormal load data is prone to miss and mistake detection of abnormal values,and reduce the accuracy of correcting abnormal data.That seriously affects the quality of load data in power system,and lead to predicted load variation law has lost its guiding significance to the power production and dispatch distribution,and even affects the safe and stable operation of power grid.For the problem of low detection efficiency and correction accuracy of abnormal load data in power system,a method of abnormal load data detection and correction based on data mining is proposed.The specific research work is as follows:1.Power load anomaly data detection methods based on Possible Fuzzy C-Means(PFCM)algorithm have some shortcomings,such as difficult to select initialization parameters and easy to fall into local optimum,which easily lead to poor validity of load curve clustering results and low efficiency of abnormal load data detection.In view of the above shortcomings,the optimized Particle Swarm Optimization(PSO)algorithm with dynamic weight adjustment and the redefined clustering validity function is adopted to optimize the initial center and number of PFCM algorithm respectively in this dissertation,and a power load abnormal data detection method based on the improved PFCM clustering algorithm is proposed.The results show that the improved abnormal data detection method not only improves the validity of load curve clustering,but also effectively reduces the error rate of abnormal load data detection.2.Aiming at the problem that the accuracy of data correction results is not high due to the unreasonable selection of network structure parameters when the traditional RBF neural network is used to correct abnormal data,a method of abnormal load data correction of power system based on RBF neural network optimized by Genetic Algorithms(GA)is proposed in this dissertation.Firstly,after initializing RBF neuralnetwork,the base function width,center,connection weight and the number of hidden layer neurons of the network are optimized jointly by genetic algorithm,which has strong global optimization ability.Then,the optimal population individuals found by multiple iterations are decoded to obtain the corresponding parameters of RBF neural network.Finally,the RBF neural network with parameter optimization is used to correct the abnormal load data of power system.The simulation results show that the proposed method can correct abnormal load data more accurately.3.Aiming at the problem of low accuracy of abnormal data correction results caused by unreasonable selection of training samples for RBF neural network,PSO-PFCM clustering results is used to train GA-RBF network to achieve more accurate correction of abnormal load data.The modified method makes full use of the clustering results,which can better reflect the overall characteristics of load curve to train GA-RBF neural network.The simulation results show that the abnormal value correction method of GA-RBF neural network based on clustering results training effectively improves the correction accuracy of abnormal load data.