基于并行数据计算的网格化电网空间负荷预测方法研究  

Research on grid power grid spatial load forecasting method based on parallel data computing

作　　者：黄存强 赵雪 张祥成 李绚绚 李跃辉 HUANG Cunqiang;ZHAO Xue;ZHANG Xiangcheng;LI Xuanxuan;LI Yuehui(Economic and Technological Research Institute,State Grid Qinghai Electric Power Company,Xining 810008,China;Tiandi Dianyan(Beijing)Technology Co.,Ltd.,Beijing 102206,China)

机构地区：[1]国网青海省电力公司经济技术研究院,青海西宁810008 [2]天地电研(北京)科技有限公司,北京102206

出　　处：《电子设计工程》2023年第9期123-127,共5页Electronic Design Engineering

基　　金：国网青海省电力公司经济技术研究院专题研究项目(SGQHJY00GHJS2100228)。

摘　　要：针对传统电网空间负荷预测算法中存在主观性判断较强且处理海量化电网数据效率偏低的缺点,文中对循环神经网络进行了改进。采用包含更新门与重置门的Bi-GRU算法使模型具有处理时间序列数据的能力,并利用叶子生长策略及直方图算法改进了GBDT算法,使其具有更高的执行效率。通过将算法数据部署在Spark云计算集群,有效提升了算法的训练速度。实验结果表明,所设计算法并行化后的运行时间与部署节点数量成反比,而RMSE及MAPE两项指标优于其他同类型算法,相比LSTM分别降低了1.5%和0.093%,证明了该算法的预测精度较为理想。同时,该算法的运行时间较短,说明其收敛度良好,故可应用于电网空间负荷预测系统中。Aiming at the shortcomings of strong subjective judgment and low efficiency in dealing with massive power grid data in the traditional power grid spatial load forecasting algorithm,the cyclic neural network is improved in this paper.Bi-GRU algorithm including update gate and reset gate is adopted to make the model have the ability to process time series data,and GBDT algorithm is improved by using leaf growth strategy and histogram algorithm to make it more efficient.By deploying the algorithm data in Spark cloud computing cluster,the training speed of the algorithm is effectively improved.The experimental results show that the running time of the parallelized algorithm is inversely proportional to the number of deployed nodes,while the two indicators of RMSE and MAPE are better than other similar algorithms,which are 1.5%and 0.093%lower than LSTM,respectively,which proves that the prediction accuracy of the algorithm is relatively ideal.At the same time,the shorter running time of the algorithm shows that its convergence is good,so it can be applied to the power grid spatial load forecasting system.

关 键 词：电网负荷预测 Bi-GRU 轻量化GBDT Spark云计算 智能电网 数据处理 

分 类 号：TP391[自动化与计算机技术—计算机应用技术] TN99[自动化与计算机技术—计算机科学与技术]