Prediction of SO₂ Emission Concentrations at FGD Outlets in Thermal Power Plants on GRU-XGBoost

As environmental regulations become increasingly stringent, desulfurization systems in thermal power plants play a crucial role in reducing SO<inf xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</inf> emission mass concentrations. Traditional methods for predicting SO<inf xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</inf> concentrations often face limitations in accuracy and adaptability when dealing with complex non-linear dynamic characteristics. To address this, we propose a hybrid model that combines the GRU (Gated Recurrent Unit) neural network and XGBoost to predict the mass concentration of SO<inf xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</inf> emissions at the outlet. This method dynamically adjusts the weights of the combined model using an adaptive allocation approach, significantly improving prediction accuracy. Experimental results show that, compared to a single LSTM model, GRU model, XGBoost model, and a GRU-XGBoost model with mean weighting, the proposed GRU-XGBoost model with adaptive allocation achieves predictions that closely match the actual SO<inf xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</inf> emission mass concentrations at the FGD outlet. This enhanced accuracy is vital for conserving resources and improving operational efficiency in power plants.