Air cooler hot air temperature prediction based on working condition parameters and improved LSTM

In view of the low level of intelligence in the water supply systems of traditional pumped storage power station technology and the coupling problem between the temperature of the water supply objects and many parameter variables, a generator air cooler hot air temperature prediction model based on operating condition parameters and improved long short-term memory（LSTM） neural network was established. Firstly, the original data was cleaned to eliminate redundant data. Secondly, the Random Forest（RF） feature dimension reduction method was used to select several high-dimensional measurement point variables involved in the target prediction to verify the correlation between the proposed working parameters and the predicting temperature objects. Finally, these working parameters were keyed into the PSO-LSTM neural network for prediction. The proposed improved LSTM methods based on working condition parameters were compared with the least squares method, BP neural network, and the original LSTM method. The results show that the proposed model can effectively predict the hot air temperature of the generator air cooler. Compared with other models, the prediction error can be reduced by 50%, and it has better prediction stability.