Research on Prediction of Turbine Mechanical Performance Degradation Based on Attention LSTM

With the continuous development of manufacturing industry, performance degradation prediction is of great significance to improve the performance reliability of equipment. In practical engineering, the source of equipment performance data is complex and time-dependent, and different performance data have different effects on equipment performance degradation prediction, which leads to the limitation of traditional prediction methods. In this paper, a Long-Short Term memory (LSTM) neural network model based on attention mechanism is proposed (Attention-LSTM). This model can effectively predict the long-term performance time series, automatically learn the weight of each performance data, and describe the impact of different performance indicators on the prediction of equipment performance degradation. Taking the “CTC three unit” turbomachinery provided by a company in Sichuan as the research object, and the results show that the Attention-LSTM model can more accurately predict the future performance decline trend of the equipment than other algorithms.