Wavelet Transform and Support Vector Machine Jointly Identify the Characteristics of Electricity Theft in Photovoltaic Systems of Dedicated Transformer Users

Abstract

In order to solve the problem that the existing methods of electricity theft detection in dedicated user photovoltaic systems are difficult to capture subtle anomalies in non-stationary electricity consumption data, this paper introduces a method combining wavelet transform and support vector machine (WT-SVM). The Daubechies wavelet basis function is used to perform multi-scale decomposition of photovoltaic electricity consumption data, extract time-frequency features, and capture transient anomalies in electricity theft behavior. The extracted features are input into the SVM classification model, and the model is trained through the RBF kernel function. Grid search and cross-validation are used to optimize hyperparameters to improve the generalization ability of the model.The results show that under the same photovoltaic power theft detection dataset and test environment, the WT-SVM in this paper extracts time-frequency features through multi-scale wavelet decomposition and combines RBF (Radial Basis Function) and SVM classification, achieving an F1 score of 94.5%, a low latency of 35ms and a noise resistance of 91.2%, and outperforms the comparison model (Time-Freq Transformer: 62.4MB; MobileNetst: 5.7MB) with a lightweight of 2.1MB. The method in this paper has a good recognition effect on electricity theft behaviors such as current bypass, inverter tampering, and data injection, verifies the effectiveness of the fusion of wavelet time-frequency analysis and machine learning, and provides a high-precision and high-practicality solution for electricity theft detection in photovoltaic systems.