Detection of power theft in sensitive stations based on generalized robust distance metric and multi-classification support vector machine

As a phenomenon of illegal occupation of power resources, power stealing has seriously damaged the fairness and normal order of the power market. In sensitive areas, that is, areas with large electricity consumption, complex lines and frequent power stealing, the detection rate of power stealing is low. Therefore, this paper studies the detection method of power stealing in sensitive areas based on generalized robust distance measurement and multi-classification support vector machine. Based on the non-technical linear loss characteristics of sensitive area, a mathematical model of electricity stealing behavior scene is constructed. The model selects current, voltage, power factor and electric quantity as key detection indexes, and analyzes them by collecting electricity consumption data. In order to weaken the influence of noise data, the generalized robust distance of electricity stealing behavior data is calculated by using the generalized robust probability distribution distance measurement criterion. The samples whose distance exceeds the set threshold are regarded as potential stealing behavior samples, and input into the multi-classification support vector machine constructed by one-to-many method. Through training and optimization, the support vector machine outputs the detection results of electricity stealing behavior according to the decision function, and realizes the accurate detection of electricity stealing behavior. The experimental results show that this method can accurately detect different types of electricity stealing behaviors such as unauthorized wiring and opening seals in sensitive areas, and the detection rate is higher than 98%.