随着国家大力推进能源供给侧结构性改革，新能源装机容量不断提升，电力市场竞争愈加激烈。另一方面，全球煤炭市场的复杂多变，导致以煤炭为能量来源的发电企业成本上涨。燃煤发热量是衡量煤质的重要评价标准之一，也是采购煤炭最重要的依据，对燃煤发热量进行准确预测能够有效地控制电厂运行采购成本。为了实现燃煤发热量的高效预测，采用Pearson系数对相关变量进行特征选取，采用基于密度的噪点空间聚类（Density-Based Spatial Clustering of Applications with Noise, DBSCAN）算法对某电厂自备煤厂近2年1 733条化验数据进行去噪，对去噪后数据进行谱聚类（Spectral Clustering, SC）分析。将分类后的子样本集采用极致梯度提升（Extreme Gradient Boosting, XGBoost）算法分别建立预测模型，并与最小二乘法回归（Ordinary Least Squares, OLS）、支持向量机（Support Vector Machines, SVM）模型进行性能比较。结果表明，基于XGBoost的电站燃煤发热量预测模型相较于其他算法准确性有明显提升，泛化能力更强。对经过SC算法分类后的燃煤分别建立预测模型能够进一步提高模型的精细化水平，为燃煤电站发热量预测提供一种可靠高效的方法。

With the country vigorously promoting structural reform on the supply side of energy, the installed capacity of new energy sources has been rising and competition in the power market has become increasingly fierce. On the other hand, the complexity and volatility of the global coal market has led to a rise in the cost of power generation enterprises using coal as their energy source. Coal heat value is one of the most important evaluation criteria for coal quality and is also the most important basis for coal procurement. Accurate prediction of coal heat value can effectively control power plant operation and procurement costs. In order to achieve efficient prediction of the heat value of coal, the Pearson coefficients were used to select the characteristics of the variables of interest, the DBSCAN algorithm was used to de-noise 1733 assay data of a power plant's own coal plant in the past two years, and spectral clustering (SC) analysis was performed on the de-noised data. The classified subsample sets were then used to build prediction models using the extreme gradient boosting (XGBoost) algorithm and compared with Ordinary Least Squares (OLS) and support vector machines (SVM) models. The performance of the models was compared with that of OLS and SVM. The results show that the accuracy of the XGBoost-based coal-fired heat value prediction model for power stations is significantly better than that of the other algorithms, and the generalization ability is stronger. The prediction model can further improve the refinement level of the model and provide a reliable and efficient method for coal-fired power station heat value prediction.