CAE-LSTM Based Abnormal Battery Temperature Detection Method in Mine Power Supply

The abnormal battery temperature detection is important for maintaining the normal operation of mine power supply. The existing detection method uses a fixed threshold value based on experience, which will alarm if the temperature exceeds the threshold. However, if the threshold is set too low, some normal data will be identified as alarms. If the threshold is set too high, the abnormal temperatures cannot be fully detected. In order to solve this problem, this paper proposes a detection method based on Convolutional AutoEncoder (CAE) and Long-term and Short-term Memory Neural Network (LSTM). Firstly, the CAE-LSTM model is established, and the normal battery temperature data of mine power supply is used to train the model. Then, all of the reconstruction losses of normal battery temperature are calculated, and the maximum reconstruction loss is taken as the threshold. Finally, the input data is reconstructed with the trained model. If the model reconstruction loss exceeds the threshold, the input data contains abnormal data. Experiments showed that this method for battery temperature anomaly detection achieves 98% accuracy rate, which is more suitable for battery temperature anomaly detection of mine power supply than the traditional fixed threshold value detection method.