Parameter Prediction of Marine Seawater Cooling System Based on EMD-BO-LIESN Combined Model

To enhance the short-term prediction accuracy of parameters in marine seawater cooling systems, this paper addresses the nonlinearity and non-stationarity characteristics of seawater cooling system state parameters. This paper selects the Leaky Integrator Echo State Network (LIESN), known for its superior performance in nonlinear and small-sample predictions, for parameter forecasting. This study utilizes Empirical Mode Decomposition (EMD) to stabilize data series, addressing the issue of loss of original information during the stabilization process of time series prediction. Meanwhile, the Bayesian Optimization (BO) algorithm is utilized to select the optimal hyperparameters for the LIESN model. And a sliding window mechanism is designed to enable adaptive prediction of time series. Ultimately, the EMD-BO-LIESN combined model was constructed. This model is applied to train and test 300 sets of condenser freshwater outlet temperature data, and the prediction accuracy is compared with that of the single-structure LIESN model, EMD-ARIMA combined model and EMD-SVR combined model. Experimental results show that the EMD-BO-ESN combined model outperforms the single-structure LIESN model, EMD-BO-SVR model and EMD-BO-ARIMA model, with a reduction in RMSE to 0.0951, a decrease in MAPE to 0.2055%, a reduction in MAE to 0.0836, and an increase in R² to 0.9843. These results indicate that the EMD-BO-LIESN combined model can be effectively used for short-term forecasting in marine seawater cooling systems, offering high accuracy and reliability.