为了提升火电机组一次调频能力，提出一种基于长短期记忆网络(Long Short Term Memory Network， LSTM)与量子粒子群算法(Quantumbehaved Particle Swarm Optimization， QPSO)的一次调频能力计算方法。将负荷指令、机组实发功率、主蒸汽压力、汽轮机总阀位开度和发电机转速作为特征变量，基于某600 MW燃煤火电机组调峰运行工况下的实际数据，构建一次调频能力计算模型。利用QPSO算法优化模型隐含层节点数、训练次数和学习率，解决了因网络结构及模型参数的不确定性产生的精度问题，并将该模型与传统的神经网络模型进行了对比。结果表明：本文所提出的方法具有更高的模型精度，从而能够为机组一次调频能力的限制因素分析和调频性能的优化提升提供模型基础。

In order to improve the primary frequency modulation capability of thermal power plants,this paper proposed a method for calculating the primary frequency modulation capability based on long shortterm memory network (LSTM) and quantumbehaved particle swarm optimization (QPSO). The parameters including load command, actual generating power, main steam pressure, opening of the turbine total valve position and generator speed were selected as feature variables. Based on the operating data in peak regulation condition of a 600 MW coalfired thermal power plant, a model was constructed to calculate the primary frequency modulation capability. The accuracy problem arising from the uncertainty of the network structure and model parameters was solved by using the QPSO algorithm to optimize the node number of hidden layers, training times and learning rate of the model. The model was compared to the traditional neural network model. The results indicate that the proposed method has higher accuracy, thus providing a basis for analyzing the limiting factors of the primary frequency modulation capability and optimizing the frequency modulation performance of the unit.

TM621