In order to quickly identify the location of the leakage point and the leak aperture in the coal mine, a model was proposed for identifying and locating the leak aperture by using the pressure and flow signals generated when the water supply pipeline leaked. Modal energy entropy and genetic algorithm combined with envelope entropy were used to optimize the parameters of variational mode decomposition (VMD), and then VMD was used to denoise the pressure signal. Convolutional neural network(CNN) was used to extract the deep feature sequence of pressure and flow signal, and the long short-term memory network(LSTM) was used to extract the time sequence of deep feature sequence to identify and locate the leak aperture. The experimental results show that compared with Kalman filter, mean value filter and low-pass filter, the variational modal decomposition with optimized parameters has higher root-mean-square error (RMSE), mean absolute error(MAE), signal-to-noise ratio(SNR) and normalized cross correlation(NCC), which indicates that it can effectively reduce noise components and retain effective signals. Compared with LSTM, the MAE, mean absolute percentage error (MAPE) and RMSE of CNN-LSTM in leak location decrease by 65.97%, 61.22% and 59.11%. In the identification of leak aperture, MAE decreases by 12.04%, MAPE decreases by 22.45%, and RMSE decreases by 3.29%, which proves that CNN-LSTM can make full use of the spatial and temporal characteristics of pipeline pressure and flow signals to identify the leak location and aperture, and its detection effect is more accurate and stable than LSTM.