【摘要】 针对燃煤烟气中二氧化硫（SO₂）浓度预测精度不足的问题，提出了一种融合光声光谱技术与Stacking集成学习的预测模型。首先，搭建光声光谱实验平台采集SO₂光谱数据，并对比了SG滤波、多元散射校正（MSC）、标准正态变换（SNV）和中值滤波四种预处理方法，依据信噪比（SNR）选取最优方法以提升数据质量。随后，将预处理后的数据按7∶3划分为训练集与测试集，并利用核主成分分析（KPCA）进行特征降维以消除冗余信息。最后，构建Stacking集成学习模型进行回归预测。实验结果表明：Stacking模型在测试集上的预测性能最优，决定系数（R²）达到0.8408,均方根误差（RMSE）低至0.0939,显著优于对比的单一机器学习模型。因此，验证了光声光谱结合Stacking模型在SO₂浓度预测中的有效性，为燃煤烟气污染的精准检测提供了关键技术支撑。更多还原

【Abstract】 To address the issue of low prediction accuracy of sulfur dioxide（SO₂） concentration in coal combustion flue gas, a model integrating photoacoustic spectroscopy（PAS） with Stacking ensemble learning is proposed. First, an experimental platform was built to collect SO₂ spectral data, and four preprocessing methods（SG filtering, MSC, SNV, and median filtering） were compared. The optimal method was selected based on signal-to-noise ratio（SNR） to improve data quality. The preprocessed data was split into training and testing sets, with kernel principal component analysis（KPCA） applied for dimensionality reduction. Finally, a Stacking ensemble model was used for regression prediction. Experimental results show that the Stacking model achieved the best performance, with an R² of 0.8408 and RMSE of 0.0939, outperforming individual machine learning models. This study confirms the effectiveness of combining PAS and Stacking for SO₂ concentration prediction, offering key technical support for accurate flue gas pollution detection.更多还原