随着新能源发电的比例不断增加,火电机组响应AGC的调节过程对电网频率安全的影响越来越大,因此,火电机组响应AGC调控能力的准确预测对电网的安全稳定运行非常重要。本文提出一种基于特征提取和多级深度学习的火电机组响应AGC调控效果的预测框架,首先采用最大相关最小冗余算法(minimum redundancy maximum relevance, mRMR)对机组运行数据进行特征提取,获得影响AGC调控效果的相关变量集。其次采用多级长短期记忆神经网络(multi-stage long short term memory, mLSTM)模型对实发功率进行预测,得到未来一段时间的功率曲线图,结合功率指令曲线计算AGC调节能力指标。使用某600MW机组实际运行数据进行验证,预测偏差在10MW以内。结果表明：本文所提方法可有效实现对火电机组出力的预测,进而精确评估深度调峰下火电机组AGC调控能力,为电网安全稳定运行提供技术支撑。

With the increasing proportion of new energy generation, the regulation process of thermal power units responding to AGC has an increasing impact on the frequency safety of the grid. Therefore, the accurate prediction of the regulation ability of thermal power units responding to AGC is very important for the safe and stable operation of the grid. In this paper, we propose a prediction framework based on feature extraction and multi-stage deep learning for the regulation effect of thermal power units in response to AGC. Secondly, a multi-stage long short term memory (mLSTM) model was used to predict the actual power generation and obtain the power curve for a period of time in the future, which was combined with the power command curve to calculate the AGC regulation capability index. The prediction is validated using actual operating data of a 600MW unit, and the deviation of the prediction is within 10MW. The results show that the proposed method can effectively predict the output power of thermal power units, and then accurately assess the AGC regulation capability of thermal power units under deep peaking, providing technical support for safe and stable grid operation.

TK221

国网河北省电力有限公司科技项目