Advanced Prediction of Major Equipment Performance Based on Deep Learning

Abstract Based on the deep learning theory, this paper proposes a CNN-LSTM neural network prediction model with attention mechanism by using the actual workload data set of key components in major equipment. Firstly, the process of data preprocessing and time series transformation into supervised learning is introduced. Including the comparison of various filtering methods, data standardization, supervised learning conversion and batch division, etc., can speed up model training and facilitate convergence. The processed data is transmitted into the model. Firstly, CNN is used to extract the high-dimensional abstract representation from the preprocessed data, and attention mechanism is added to give each channel different attention scores. Then, LSTM is used to process the time series data and output the final prediction results. In this study, 200 data points in the future are predicted, and the final predicted results are RMSE of 33.2626 and MAPE of 15.37677, which can well fit the future load change trend. At the same time, the multi-condition identification module is added to the code, which can better adapt to the changing load of mechanical equipment in the actual service process. Finally, this paper also designed an ablation experiment, which proved the effectiveness of the overall structure of the prediction model designed in this study and the necessity of the three parts of CNN, ECA-Net and LSTM.