Early warning of steam turbine operation state based on LSTM

Aiming at the problem that traditional steam turbines lack effective early warning methods and are often in passive maintenance,a steam turbine state early warning method based on long short-term memory network（LSTM）is proposed. The proposed method includes three modules: data preprocessing module, health evaluation module, and abnormal warning module. Firstly, the source data is preprocessed to remove outliers and burr data.Secend, an optimized health index is obtained based on autoencoding neural network, cosine theorem and 3σ theorem. Finally, a steam turbine abnormality early warning model is established based on LSTM, and the different Results of deep LSTM network model and recurrent neural network（RNN） prediction. The final results show that the Mean Absolute Percentage Error（MAPE） of the prediction results of the best prediction model of LSTM does not exceed 4.31%, which is higher than the accuracy of the best prediction model of traditional RNN. Therefore, the proposed method has better detection accuracy in the early warning of turbine anomalies.