结合注意力机制与BIM特征的电力能耗预测  

ELECTRICITY CONSUMPTION FORECASTING BASED ON BIM AND ATTENTION MACHANSIM

作　　者：田英杰 郭乃网 徐东辉 庞悦 周向东[2] 施伯乐[2] Tian YingjieGuo Naiwang;Xu Donghui;Pang Yue;Zhou Xiangdong;Shi Baile(State Grid Shanghai Electric Power Research Institute,Shanghai 200437,China;School of Computer Science and Technology,Fudan University,Shanghai 200433,China)

机构地区：[1]国网上海市电力公司电力科学研究院,上海200437 [2]复旦大学计算机科学技术学院,上海200433

出　　处：《计算机应用与软件》2021年第6期39-45,51,共8页Computer Applications and Software

基　　金：国家自然科学基金项目(61370157);国家电网公司总部科技项目(52094017001X)。

摘　　要：能耗预测中引入外因序列有助于目标序列的预测,但在实际应用中各个外因序列与目标序列之间的相关程度往往不清晰,导致无法有效利用多外因序列辅助预测。针对该问题,提出一种结合注意力机制与BIM特征的电力能耗预测模型——BIMAttenNN(BIM Attention Nerual Network)。通过结合注意力机制与BIM特征对外因序列自动选取并重构,将重构特征通过编码器-解码器结构的深度神经网络和线性回归分支准确预测未来电力能耗。实验结果表明,BIMAttenNN能够结合BIM特征自动捕捉电力序列间关系,与传统方法相比具有更高的预测精度。The valuable extrinsic sequences make positive contribute to the forecasting of target sequences,but in practice,the degree of correlation between each extrinsic sequence and target sequences is often unclear,resulting in the inability to effectively utilize the multiple extrinsic sequence to help improve forecasting accurary.In response to this problem,a BIM attention nerual network(BIMAttenNN)that combined BIM technology and attention mechanism is proposed.The model first automatically selected and reconstructed the extrinsic sequences by combining BIM and attention mechanism,and then accurately forecasting the future electricity consumption through the deep neural network of the encoder-decoder structure and linear regression branch.The experimental results show that BIMAttenNN can automatically capture the relationship between multi-variants sequences with BIM features.Compared with the traditional method,the proposed electricity consumption forecasting model has higher forecasting accuracy.

关 键 词：电力能耗预测 BIM 注意力机制 深度学习 

分 类 号：TP3[自动化与计算机技术—计算机科学与技术]