应用时空大数据的电力需求侧不平衡数据渐进学习算法

Progressive Learning Algorithm for Power Demand Side Unbalanced Data Using Spatio-Temporal Big Data

针对重采样过程中,不平衡数据分类结果G-mean值较低,数据类分布不平衡的问题,考虑到电力需求侧不平衡数据的独特性,提出应用时空大数据的不平衡数据渐进学习算法.基于物联网终端节点采集的动态时空大数据,设计的不平衡数据渐进学习算法,建立基于长短记忆网络的时空大数据处理机制.依据属性值域,选定合适的时空数据与电力需求侧数据进行合成处理,促进数据类分布平衡.按照KL距离计算理念,获取计算不同数据之间的KL距离,并依托于KL距离,设计半监督学习算法,求解与优化电力需求侧不平衡数据渐进学习过程.实验结果表明:所提算法的平均G-mean值为0.93,平均G-mean值最高提升了37%.

Aiming at the problems of low g-mean value of unbalanced data classification results and unbalanced distribution of data classes in the process of resampling,considering the uniqueness of unbalanced data on the power demand side,a progressive learning algo-rithm for unbalanced data using spatio-temporal big data is proposed.Based on the dynamic spatiotemporal big data collected by the terminal nodes of the Internet of things,an unbal-anced data progressive learning algorithm is designed to establish a spatiotemporal big data processing mechanism based on the long and short memory network.According to the at-tribute value range,the appropriate spatio-temporal data and power demand side data are selected for synthetic processing to promote the balance of data class distribution.According to the KL distance calculation concept,obtain and calculate the KL distance between differ-ent data,and based on the KL distance,design a semi supervised learning algorithm to solve and optimize the progressive learning process of unbalanced data on the power demand side.Experimental results show that the average g-mean value of the proposed algorithm is 0.93,and the maximum increase of the average g-mean value is 37%.