Compared with a regular power plant, the load and fuel composition of a captive power plant vary greatly, resulting in complex and variable working conditions, and for specific operating conditions, there is a lack of data. Considering the irregularity and incompleteness of the data, a data-mining-based boiler energy management analysis method was proposed in this paper. This study adopted the random forest and recursive feature elimination (RF-RFE) method to select the features of distributed control system (DCS) monitoring data. According to different load intervals, some data with better running conditions were selected for clustering. According to the clustering center, adjustable parameters were optimized, and the obtained optimization results were input to the support vector machine model for energy consumption comparison. The results show that the method used in this study can reduce the unit energy consumption by 3–5% by adjusting the controllable parameters.