Electricity user behavior data is complex and diverse, resulting in significant variability and uncertainty in user behavior data, which increases the difficulty of monitoring electricity user behavior and leads to low monitoring rates. This paper utilizes the singular value equivalent matrix to obtain a non-Hermitian matrix and performs standardization processing on the aforementioned matrix. Considering the ARMA equation system for time series stationarity, the proposed numerical solution is used to calculate the expression, thereby extending RMT from a purely Gaussian environment to a non-Gaussian environment. An ETD-SAC electricity theft detection model framework is constructed to determine whether users are engaging in electricity theft during the detection period. Through user electricity consumption behavior detection, it was found that the electricity load trend of electricity theft users fluctuated between [8.54, 38.54] kWh after July 15, 2023. One of the suspected users detected bypassed the meter for electricity theft, with the meter current ranging from -0.1 to 0.4 A, while the actual incoming current was 0.6 to 1 A, constituting electricity theft behavior. Using the same method for electricity theft behavior analysis, CZ Factory was found to have engaged in electricity theft on October 1, 2023, requiring the recovery of 1,354 units of electricity and 1,126.528 yuan in electricity fees. The anti-electricity theft application model based on ARMA achieved good results.
