基于特征提取与MLSSVR的电力系统动态频率预测OA

In order to solve the problems such as too long computation or training time,low precision and poor model generalization performance when calculating the maximum information coefficient and traditional support vector machine are used to predict the dynamic frequency of power system,the mesh partitioning method of calculating the maximum information coefficient is improved,and the inequality constraint conditions and model loss function of the original SVR model are modified.A frequency response curve prediction model based on statistical information coefficient and multi-output least squares support vector machine(MLSSVR)is proposed to realize feature extraction,global output and chain prediction of dynamic frequency of power system,so as to improve the accuracy,accuracy and speed of the original machine learning model.Grey wolf optimization algorithm was used to optimize kernel function width and penalty factor in MLSSVR algorithm to improve the comprehensive performance of the model.Simulation experiments on 68 nodes of IEEE16 show that compared with SVR algorithm and other machine learning models,the accuracy,precision and training prediction speed of the proposed model are improved,and the model performance is stronger.