Looseness or overload of bolts on the head cover of the hydropower unit is one of the key factors leading to equipment failure, which seriously affects the safe operation and economic benefits of the unit. A combined model for bolt fault prewarning based on XGBoost and Bidirectional Long Short-Term Memory (BiLSTM) is proposed in this paper. Firstly, the obtained bolt axial force time series data were preprocessed, including missed value filling and abnormal value deletion. Secondly, the role of BiLSTM is to capture the dependencies in the bolt time series and extract high-dimensional data features. Then, the BiLSTM features were fused with the original data and input into the XGBoost model for nonlinear regression prediction. The results show that the RMSE of the combined model is 5.3215 and R2 is 0.9063, which is improved significantly compared with that of other models (RMSEs of XGBoost model, BiLSTM model and CNN-BiLSTM are 9.4341, 10.4616 and 7.1566 respectively). It can accurately diagnose the loosening and overload state of bolts and provide theoretical support for the intelligent operation and maintenance of hydropower units.