Accurately predicting the fault of rod-pumped wells is of great significance for oilfield production. Aiming at the fault situation of rod-pumped wells in a block of Xinjiang oilfield, the production data of 500 wells were collected, and the 5 main factors causing the fault of pumping wells were identified, such as scale, wax formation, rod corrosion, rod fatigue and rod partial wear. Based on long short-term memory networks (LSTM), the intelligent early-warning model of oil well failure was constructed. By selecting 14 characteristic parameters that affect oil well faults for wavelet denoising, the model was trained and tested with the help of adaptive moment estimation algorithm. The findings suggest that the prediction accuracy of the model is 96.81%, which can provide more accurate early warning for rod-pumped well faults.