Research on Abnormal Discrimination of Three-Phase Four-Wire Connection Based on Improved Machine Learning Models

Due to the large number of connections and the complex variety of connections in three-phase electricity meters, along with some users engaging in electricity theft, three-phase electricity meters are prone to miswiring, which adversely affects the accurate measurement of electricity consumption and the calculation of electricity bills. Traditional manual inspection methods, constrained by human and material resources, cannot meet the diagnostic needs for miswiring. Therefore, this paper focuses on the miswiring issues of three-phase meters, improving three machine learning algorithms: BP Neural Network, Support Vector Machine (SVM), and Extreme Learning Machine (ELM), to establish a model for miswiring in three-phase four-wire systems, helping to enhance the accuracy of miswiring identification. The main work includes first redefining the categories of miswiring due to the variety of three-phase four-wire miswiring types. Then, using the Random Forest algorithm for feature analysis to select the most relevant features. Next, the Grey Wolf Optimization algorithm is employed to optimize the parameters of the machine learning models, and the improved model is used for identifying miswiring in three-phase four-wire systems. Finally, instance validation shows that the Grey Wolf Optimized-SVM (GWO-SVM) algorithm achieves the highest accuracy, reaching 88.8%.