Abstract: To predict the thermal performance of the compressor and combined cycle unit, and ensure their efficient operation, a compressor and unit performance prediction model based on principal component analysis (PCA), particle swarm optimization (PSO) algorithm, and long short-term memory (LSTM) neural network was proposed. Firstly, PCA was applied to reduce the input parameters from 13 dimensions to 7 dimensions, preserving key information while reducing computational complexity and model training time. Then, combined with LSTM, the time series data were modeled to capture the dynamic trends and nonlinear relationships of the data. To further optimize model performance, PSO algorithm was introduced to automatically tune LSTM hyperparameters, thereby improving prediction accuracy and model stability. Finally, the alarm threshold was calculated according to predicting error distribution. Results show that the PCA-PSO-LSTM based model exhibits high accuracy in predicting compressor outlet temperature, unit corrected output and corrected heat consumption. Compared with traditional methods, the root mean square error (RMSE) and mean absolute error (MAE) of the proposed model are significantly reduced. RMSE is reduced over 30% and MAE remains below 0.8. In addition, the calculated alarm threshold can provide theoretical basis and technical support for abnormal warning of compressor operating status.