随着新能源的大比例上网，燃煤机组深度调峰已成为新常态。为实现火电机组的节能提效，开展不同调峰深度下火电机组能效研究势在必行。本文首先构建了机组能效状态指标体系，通过对火电机组的运行数据进行分析，采用基于K-medoids多指标同步聚类算法获得供电煤耗率辅助基准值；其次，建立各工况能效特征指标与供电煤耗率的长短期记忆LSTM神经网络非线性模型，进而获得更具泛化性的机组各工况供电煤耗率最优基准值；最后，建立了多状态指标融合的能效评价模型，形成对火电机组能效状态的实时评价结果。以660MW和1000MW机组为例，对所提的火电机组能效状态实时评价方法进行了验证。结果表明，该方法实现了对火电机组能效状态的在线监测和评价，可为火电机组能效指标排查及节能优化调整提供指导。

With the large proportion of new energy into the grid, deep peak shaving of coal-fired units has become the new normal. In order to achieve energy saving and efficiency improvement of thermal power units, it is imperative to carry out energy efficiency research of thermal power units under different peak load depth. In this paper, the unit energy efficiency state index system is constructed, and the auxiliary reference value of power supply coal consumption rate is obtained by using K-medoids multi-index synchronous clustering algorithm through analyzing the operation data of thermal power units; Secondly, the LSTM neural network nonlinear model of energy efficiency characteristic index and power supply coal consumption rate in each working condition is established, and then the optimal reference value of power supply coal consumption rate in each working condition is obtained with more generalization; Finally, the energy efficiency evaluation model of multi-state index fusion is established to form the real-time evaluation results of the energy efficiency state of thermal power units. The proposed real-time evaluation method for energy efficiency status of thermal power units was validated using 660MW and 1000MW units as examples. The results show that this method achieves online monitoring and evaluation of the energy efficiency status of thermal power units, which can provide guidance for the investigation of energy efficiency indicators and energy-saving optimization adjustment of thermal power units.

国网湖南省电力有限公司科技项目（NO.5216A521N00H）；国家自然科学青年基金项目(NO. 51706022)