基于mRMR与多级LSTM网络的火电机组响应AGC调控能力评估
Assessment of AGC Regulation Capability of Thermal Power Units based on mRMR and Multi-level LSTM Networks
郝晓光 1金飞 1张庆浩 2张文彬1
作者信息
- 1. 国网河北能源技术服务有限公司,河北 石家庄 050021
- 2. 华北电力大学 控制与计算机工程学院,北京 102206
- 折叠
摘要
为准确预测火电机组响应自动发电控制(Automatic Generation Control,AGC)调控的能力,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mRMR-mLSTM预测模型: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以提高建模效率;其次,采用多级长短期记忆神经网络模型(Multi-stage Long Short Term Memory,mL-STM)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;最后,使用某600 MW机组实际运行数据进行验证,预测偏差在10 MW以内.结果表明:本文所提模型的预测精度相较于未进行特征提取的模型和单一LSTM模型分别提高了21%和40%,证明该模型可精确评估深度调峰下火电机组响应AGC调控的能力.
Abstract
In order to accurately predict the automatic generation control(AGC)regulation capability of thermal power units,this paper proposed a mRMR-mLSTM prediction model based on feature extraction and multi-stage deep learning for thermal power units to respond to the effect of AGC regulation.Firstly,the minimum redundancy maximum relevance(mRMR)algorithm was used to extract features from the u-nit operation data to obtain the relevant variable set affecting the effect of AGC regulation in order to im-prove the modeling efficiency;secondly,the actual power was predictd by using the multi-stage long short term memory(mLSTM)neural network model,to obtain the power curve for the future period of time and to compute the indexes of the AGC regulation capability combining with the power command curve;final-ly,the actual operating data of a 600 MW unit were used to verify the prediction deviation was within 10 MW.The results show that the prediction accuracy of the model proposed in this paper is 21%and 40%higher than that of the model without feature extraction and the single LSTM model,respectively,which proves that the model can accurately assess the AGC regulation capability of thermal power units under deep peak shaving.
关键词
火电机组/自动发电控制/响应能力预测/最大相关最小冗余法/长短期记忆神经网络Key words
thermal power units/automatic generation control/responsiveness forecasting/mRMR/LSTM引用本文复制引用
基金项目
国家电网河北省电力公司科技项目(TSS202104)
出版年
2024
NETL
NSTL
万方数据