Due to the high economic value of power cable, cable theft often occurs. In order to reduce the occurrence of cable theft in the cable trench of urban distribution system, online monitoring algorithm based on triaxial acceleration sensor is designed to distinguish whether the cable is damaged or not. Firstly, an experimental platform is built in the cable trench to collect the acceleration data of the sensor, and the wavelet coefficient correlation method is used for denoising. Then the angle between acceleration and gravity direction is calculated. Based on the DB4 wavelet basis, the data is decomposed into four layers of wavelet to obtain the angle-based wavelet energy spectrum. At the same time, the synthetic acceleration, skewness, standard deviation, entropy and other time-frequency characteristic criteria are calculated After classification by BP neural network algorithm, many tests show that the average accuracy of the algorithm is 96.25%, which provides a new idea for cable anti-theft.