A Predict Method of Measuring Equipment Operation Performance Based on Improved Local Weighted Partial Least Squares

Abstract

In view of the fact that the operation data of metering equipment in the power system has strong nonlinearity and immediacy, traditional methods cannot effectively model, predict and analyze. This paper proposes an improved local weighted partial least squares algorithm (K-MLWPLS) based on K-means clustering. The method first uses the K-means clustering algorithm to divide the training set into k sub-training sets; then uses the locally weighted partial least squares algorithm combined with the Two-scale similarity measure to model the sub-training sets, and uses grid search and Cross-validation adjusts the model parameters to obtain the optimized k sub-models; then for the test set samples, the sub-models weighted based on the centroid neighbour’s radius are integrated to calculate the final predicted value corresponding to the test set samples. The algorithm is applied to the prediction analysis of the operation data of the metering equipment. The experimental results show that the K-MLWPLS algorithm significantly improves the prediction accuracy of the model compared with the traditional modeling algorithm.