准确的CO₂排放计量与动态预测对于推动我国“双碳”目标（碳达峰与碳中和）的实现至关重要。该文结合排放因子法、CO₂-CEMS法和机器学习技术，提出了一种基于多源数据融合的CO₂排放计量与预测方法，旨在为燃煤电厂提供一个高效且精确的碳排放监测工具。研究中通过对比不同方法的碳排放量计算结果，评估了不同机器学习算法的性能，并基于多源数据开发了一个动态预测模型。在国能集团有限公司河北某电厂1#和3#机组上的实验结果表明，排放因子法和CO₂-CEMS法的CO₂排放量计算相对偏差分别为1.63%和-1.27%，两种方法具有较好的互证性。对多种机器学习模型（如XGBoost、LightGBM、AdaBoost）的性能进行了对比，在决定系数（R²）和平均绝对百分比误差（MAPE）这两个常规评价指标之外，提出一个新的筛选标准——将已训练模型应用至其他机组得到的与实测值的均值偏差x_c，将其用于机器学习算法的泛化能力评估及进一步筛选，发现AdaBoost算法在预测精度和稳定性方面表现最佳，并具有更好的泛化能力与鲁棒性。采用优选的AdaBoost算法进行CO₂排放的动态预测，在训练集和测试集上的决定系数均大于0.99，MAPE低于2%，表明该算法具有很高的预测准确性和稳定性，并且具有极强的泛化能力和鲁棒性。该文提出的多源数据融合方法不仅有效克服了传统方法在动态变化中的局限性，还能根据实时数据进行小时级的CO₂排放量精准预测。

Accurate CO₂ emission measurement and dynamic forecasting are crucial for achieving China’s “dual carbon” goals (carbon peak and carbon neutrality). This study integrates the emission factor method, CO₂-CEMS and machine learning technologies to propose a CO₂ emission measurement and forecasting method based on multi-source data fusion, for the purpose of providing an efficient and accurate carbon emission monitoring tool for coal-fired power plants. In the investigation, by comparing the CO₂ emission calculation results obtained by different methods, the performances of different machine learning algorithms are evaluated, and a dynamic forecasting model based on multi-source data is developed. Experimental results in Units 1 and 3 of a power plant of Guoneng Group Co., Ltd. in Hebei, China show that the relative deviations of CO₂ emission calculations between the emission factor method and CO₂-CEMS for the two units are 1.63% and -1.27%, respectively, meaning that the two methods are of good cross-validation. By comparing the performance of various machine learning models (such as XGBoost, LightGBM, and AdaBoost), beyond the two conventional evaluation metrics, namely the determination coefficient (R²) and the mean absolute percentage error (MAPE), a new selection criterion, namely the mean deviation (x_c), is proposed by applying trained models to other units. Then, x_c is used to assess the generalization capability of machine learning algorithms for further model screening. The results reveal that AdaBoost exhibits superior performance in prediction accuracy and stability, along with higher generalization capability and robustness. The dynamic CO₂ emission forecasting using the optimized AdaBoost algorithm achieves R² values greater than 0.99 on both the training and the testing sets, with a MAPE below 2%, which indicates that the algorithm is of high prediction accuracy, stability, generalization ability and robustness. The proposed multi-source data fusion method not only effectively overcomes the limitations of traditional methods in dynamic scenarios but also enables precise hourly CO₂ emission forecasting based on real-time data.