提升火电机组的一次调频能力辨识有助于辅助电网的调度，保证电网的安全稳定运行。提出一种基于贝叶斯优化算法(Bayesian optimization， BO)的长短期记忆网络(long shortterm memory， LSTM)一次调频能力辨识方法，实现火电机组的一次调频能力精确建模。首先对机组机理及参数之间的相关性进行分析，确立模型的输入特征变量，再利用贝叶斯算法对LSTM网络结构进行优化，得到一次调频能力辨识模型。以某600 MW燃煤火电机组为研究对象，将该模型与传统BP神经网络模型、未优化LSTM网络模型进行对比。结果表明：所提出的网络模型均方根误差分别降低了66.51%和34.83%，具有更高的模型精度。

Improving the identification of primary frequency modulation capability of thermal power plants is helpful to assist the dispatching of power grid and ensure the safe and stable operation of the power grid. Therefore, this paper proposed a method for identifying the primary frequency modulation capability of thermal power plants based on the Bayesian optimization (BO) algorithm and long shortterm memory (LSTM) network, which achieved accurate modeling of the primary frequency modulation capability of thermal power plants. Firstly, the input feature variables of the model were established through the mechanism analysis of plant and the correlation analysis of parameters. Then, the LSTM network structure was optimized by using Bayesian algorithm to obtain the primary frequency modulation capability identification model. Based on the operating data of a 600 MW coalfired thermal power plant, the proposed model was compared with the traditional BP neural network model and the unoptimized LSTM network model. The results show that the root mean square errors (RMSE) of the proposed network model in this paper are 66.51% and 34.83% lower than that of the traditional BP neural network model and the unoptimized LSTM network model, which has higher model accuracy.

TM621

国家电网有限公司总部管理科技项目(52060021N00P)