摘要
有效的短期电力负荷预测模型有利于保障电力系统稳定且高效地运行。为此,首先提出了一种具有相邻反馈的混合回声状态网络(hybrid echo state network with adjacent-feedback loop reservoir,HALR)模型,用以避免传统浅层模型使用单一类型神经元易产生奇异解的问题。然后,基于深度信念网络(deep neural network,DBN)和HALR模型提出了一种深度混合储备池计算(deep hybrid reservoir calculation,DHRC)模型,以提高传统模型的预测精度和效率,该模型实现了DBN优秀特征学习能力和HALR强大逼近性能的结合。将DHRC模型应用于比利时蒙斯大学采集的某地区电力负荷数据集,最终的X_(NRMSE)、X_(RMSE)和X_(MAPE)分别为0.6591、0.0541和4.8523%。最后,在西北某电网供电公司的实际应用中再次证明了DHRC模型的有效性。实验结果表明,与预测效果最佳的浅层模型HALR相比,DHRC的X_(NRMSE)、X_(RMSE)和X_(MAPE)分别降低了65.1685%、65.1079%和60.0954%;与预测效果较好的深度模型LSTM和DBEN相比,DHRC模型的预测效率分别提高了36.5566%和9.4276%。
An effective short-term electrical load prediction model is beneficial to ensure the steady and efficient functioning of the electrical system.To this end,this paper firstly proposes a hybrid echo state network with the adjacent-feedback loop reservoir(HALR) model In order that it can avoid the problem of easily yielding singular solutions by using a single type of neurons in the traditional shallow models.Then,based on the deep belief network(DBN) and the HALR,a deep hybrid reservoir calculation(DHRC) model is suggested to promote the prediction accuracy and efficiency of the traditional models,realizing the combination of the DBN’s great feature learning capabilities with the HALR’s powerful approximation capabilities.The DHRC is applied to a regional electrical load data set collected by the University of Mons in Belgium,and the final X_(NRMSE),X_(RMSE) and X_(MAPE) are 0.6591,0.0541 and 4.8523%,respectively.Following that,the DHRC’s validity is demonstrated once more in the practical application of an electrical grid electrical supply company in the northwest of China.The experimental results demonstrate that when compared to the shallow model HALR with the best prediction effect,the X_(NRMSE),X_(RMSE) and X_(MAPE) of DHRC are reduced by 65.1685%,65.1079% and 60.0954%,respectively.As compared to the well-predicted deep models LSTM and DBEN,the prediction efficiency of the DHRC is increased by 36.5566% and 9.4276%,respectively.
作者
张明辉
周亚同
孔晓然
ZHANG Minghui;ZHOU Yatong;KONG Xiaoran(School of Electronics and Information Engineering,Hebei University of Technology,Beichen District,Tianjin 300401,China)
出处
《电网技术》
EI
CSCD
北大核心
2022年第12期4751-4761,共11页
Power System Technology
基金
京津冀基础研究合作专项(H2021202008,J210008)
内蒙古自治区纪检监察大数据实验室开放课题(IMDBD202105)
河北省博士在读研究生创新能力培养项目(CXZZBS2022040)。
关键词
深度信念网络
储备池计算
短期电力负荷预测
短期记忆能力
deep belief network
reservoir calculation
short-term electrical load prediction
short-term memory ability