A Dynamic Adaptive Boiler Efficiency Prediction Model Based on VMD-ILSSVM-LSTM

Accurate prediction of boiler efficiency is critical for the optimal operation of power plant combustion systems. This study presents a hybrid prediction model of boiler efficiency combining Variational Mode Decomposition (VMD), improved Least Squares Support Vector Machine (ILSSVM), and Long Short-Term Memory (LSTM) network. Initially, VMD is used to decompose and reconstruct the boiler efficiency sequence, effectively capturing both high- and low-frequency features. The LSTM network is then leveraged to model the low-frequency characteristics of the data, while the LSSVM is utilized for the accurate prediction of high-frequency components. In cases where the prediction error exceeds the predefined threshold, model update strategy is conducted by replacing, adding, or deleting LSSVM samples and updating corresponding model parameters. Finally, the low- and high-frequency components predicted by LSTM and improved LSSVM are linearly combined to form the final prediction result. The simulation results demonstrate that the proposed dynamic adaptive VMD-ILSSVM-LSTM model significantly improves prediction accuracy with the lowest MAE and RMSE values.