基于ICPSO-XGBoost的非侵入式负荷辨识方法

A non-intrusive load identification method based on ICPSO-XGBoost

针对家庭用电负荷的电气特征相近导致基于电气量特征的非侵入式负荷辨识方法易产生误辨识的问题,提出以电器投切时间、运行时长和投切次数为代表的电器使用规律特征,结合传统电气负荷特征组合成为新的负荷特征标签。在此基础上,提出一种基于改进混沌粒子群优化的极端梯度提升树算法。在该算法中,利用回归树作为负荷特征的基分类器构建极端梯度提升树模型。进一步地,通过在目标函数中加入正则项,添加缩减系数等措施避免算法陷入过拟合。同时,将混沌思想应用于粒子群算法中提升其全局寻优能力,并得到基于改进混沌粒子群优化后的极端梯度提升树算法模型。在AMPds公用数据集上进行测试,通过对比分析测试结果,验证了文中所提出的负荷特征标签和负荷辨识算法对提升非侵入式负荷辨识的有效性。

Considering that the misidentification of non-intrusive load identification methods often occurs when the electrical features of household loads are similar,combining with the traditional electrical load characteristics,this paper proposes a novel type of load feature label,which is represented by utilization behavior of household,including household switching time,operation period and the numbers of switching.On this basis,this paper presents extreme gradient boosting algorithm based on improved chaotic particle swarm optimization(PSO).Firstly,extreme gradient boosting model with regression trees is built as the basic classifier.To avoid ease over fitting,regularization terms to object function and reduction coefficient are then introduced into our method.Alternatively,chaos thought is applied to particle swarm algorithm to find the global optimization ability,and the extreme gradient boosting algorithm model is obtained based on improved chaotic PSO.Finally,tests are carried out on the AMPds public data set,and the validity of the proposed load feature label and load identification algorithm to improve the non-intrusive load identification is verified through comparing and analyzing the test results.