Status Forecast and Fault Classification of Smart Meters Using LightGBM Algorithm Improved by Random Forest

Abstract

The objectives are to explore the effect of a random forest algorithm on the state prediction and fault classification of smart meters, so that the smart meters can run more stably. Based on the principle of the random forest algorithm and Light Gradient Boosting Machine (LightGBM) algorithm, its theoretical basis and application are deeply analyzed and improved. An improved fault classification and state prediction model of smart meters is designed based on a random forest-improved LightGBM algorithm. The built model algorithm is evaluated by utilizing public data sets. The results show that, by preprocessing the fault data set of smart meters, 8 fault feature types including fault type, working time, and fault month are obtained. When the improved LightGBM algorithm is trained based on random forest, the average accuracy of the algorithm is 67.65%, the average recall rate is 64.11%, and the average F1 value is 65.73%. Meanwhile, the difference between the algorithm and the random forest algorithm and the Correlation-based Feature Selection (CFS) algorithm is studied. Therefore, the prediction accuracy and fault classification of the constructed model algorithm for smart meters are higher than those of the other two algorithms. It indicates that the algorithm has a good application effect and high practical application value and can provide a scientific and useful reference for the follow-up research of smart meters.