State Estimation of Distribution Network based on Historical Data Mining and APSO-BP algorithm

For the safety and stability of the distribution network, the distribution management system (DMS) requires high precision state estimation results as the basic information. If there are errors in line parameters, the precision of estimation results of Weighted Least Squares (WLS) state estimation method will be lower because the WLS model relies on line parameters. This paper presents a state estimation model based on historical data mining via multivariate linear regression. BP neural network is used to directly obtain the mapping between input and output, which would avoid the dependency on line parameters. Therefore, the proposed model would not be affected by the errors of parameters. At first, Recursive Least Squares (RLS) is used to establish a historical data model to calculate pseudo measurement. Then the pseudo measurement is taken as part of the inputs to BP neural network which is improved by Adaptive Particle Swarm Optimization algorithm (APSO-BP). Finally, the proposed method is simulated based on IEEE 33 nodes system. The results are evaluated through maximum absolute error and mean absolute error, which show that the proposed method has high precision and performances well in avoiding the influence of line parameters errors.