基于改进LSTM的电力调度数据预测模型设计与仿真  

Design and simulation of power dispatching data prediction model based on improved LSTM

作　　者：曹帅 李晓君 贺成铭 程方亮 吴鑫 CAO Shuai;LI Xiaojun;HE Chengming;CHENG Fangliang;WU Xin(Shenyang Institute of Computing Technology Co.,Ltd.,Chinese Academy of Sciences,Shenyang 110168,China;Xinjiang Electric Power Trading Center Co.,Ltd.,Urumqi 830063,China)

机构地区：[1]中国科学院沈阳计算技术研究所有限公司,辽宁沈阳110168 [2]新疆电力交易中心有限公司,新疆乌鲁木齐830063

出　　处：《电子设计工程》2024年第19期173-177,共5页Electronic Design Engineering

摘　　要：为了提高新型电力系统下负荷数据的预测精度,文中对基于融合信息的深度学习网络展开了研究。通过将电力负荷的预测抽象为时间序列处理问题,并以长短期记忆单元(LSTM)替代传统的神经元结构,进而提升了网络的记忆能力和长序列处理能力。由于所提网络以气象及能源交易信息等多源数据为训练集,因此引入了一种Attention机制。该机制在增强有用信息权重的同时还能降低LSTM网络对次要信息的关注力,从而提升网络的特征提取能力。为验证算法的改进效果,以采集到的负荷、电价与气象数据等为基础建立了训练、验证和测试数据集。仿真结果表明,改进LSTM算法的负荷预测结果更接近于实际情况,且算法的性能指标均有明显改善,其中MAE和RMSE分别下降了24.13%及23.13%。In order to improve the forecasting accuracy of load data in the new power system,this paper studies the Deep Learning network based on fusion information.The prediction of power load is abstracted as a problem of time series processing,and the traditional neuron structure is replaced by Long Short⁃Term Memory(LSTM),which improves the memory ability and long series processing ability of the network.Since the network uses meteorological data,energy transaction data and other multi⁃source data as training sets,an Attention mechanism is also introduced to enhance the weight of useful information while reducing the attention of LSTM network to secondary information,thus improving the network's feature extraction ability.In order to verify the improvement effect of the algorithm,the training,verification and test data sets are established based on the collected load,electricity price and meteorological data.The simulation results show that the load forecasting results of the improved LSTM algorithm are closer to the actual situation,and the performance indicators of the algorithm are significantly improved,with MAE and RMSE descend by 24.13%and 23.13%respectively.

关 键 词：深度学习 融合信息 电力负荷预测 改进LSTM Attention机制 

分 类 号：TP311[自动化与计算机技术—计算机软件与理论]