Design of electric power theft detection method based on convolutional neural network and Kalman filter

To address the key issues of low accuracy and high false alarm rate in electricity theft detection in smart grids, this paper proposes an innovative detection method based on the deep integration of convolutional neural network and Kalman filter. Firstly, according to the data structure of the power system, the learning model based on neural network is constructed, and the data processing, feature extraction, and anomaly detection are carried out on the convolution layer and the pooling layer. Secondly, design a Kalman filter, linearize the nonlinear measurement function output by convolutional neural network using the Jacobian matrix, and achieve accurate state estimation through the iterative square root of the state covariance matrix. Thirdly, establish a joint decision-making mechanism of convolutional neural network feature space and Kalman state space to achieve dynamic detection and prediction update of electricity theft behavior, breaking through the single limitations of existing methods in feature representation or noise processing, and achieving the organic combination of feature learning ability and state estimation ability. Compared with the traditional detection methods, the accuracy of the proposed method was 98.9% and the false-positive rate was 0.71%, this improves the detection accuracy of electric theft and the robustness of the system, providing reliable support for the safe operation of the power grid.