An elevator trapping fault warning method based on self-attention-LSTM

Due to the complexity of elevator structures and the numerous components involved, prolonged operation often leads to entrapment faults. However, current handling methods predominantly rely on post-incident rescue efforts and lack proactive preventive measures. To achieve autonomous and accurate predictions of elevator entrapment faults, we propose a predictive model integrating a Long Short-Term Memory Neural Network (LSTM), a Self-Attention Mechanism, and the Adaptive Moment Estimation Algorithm (Adam). This model leverages extensive IoT and historical data from elevators. Initially, we conduct thorough data analysis, followed by data cleaning and normalization, ensuring a robust foundation through weig hted screening. Subsequently, we construct an LSTM-based model, enhanced by a self-attention mechanism layer to improve performance. The model is then optimized using the Adam algorithm. Ultimately, this approach enables precise early warnings for elevator entrapment faults. Experimental results indicate that, with a learning rate of 0.001, the model achieves optimal training results, reducing the model loss rate to 0.125%, and achieving better parameter fitting. The warning model demonstrates a considerable accuracy rate in predicting elevator entrapment scenarios.