| 引用本文: | 谈林涛,李军良,任昺,何杨,高欣,徐建航,黄晴晴.基于RB-XGBoost算法的智能电网调度控制系统健康度评价模型[J].电力自动化设备,2020,40(2): |
| TAN Lintao,LI Junliang,REN Bing,HE Yang,GAO Xin,XU Jianhang,HUANG Qingqing.Health evaluation model of smart grid dispatch and control system based on RB-XGBoost algorithm[J].Electric Power Automation Equipment,2020,40(2): |
|
| 摘要: |
| 针对智能电网调度控制系统(D5000系统)健康度评价，基于专家经验的传统评价方法存在主观性较大的问题，机器学习多分类方法是提高评价客观性的一种有效手段，但健康度各等级样本数目间存在的不平衡问题导致分类准确率较低，为此提出一种基于随机平衡和极端梯度提升(RB-XGBoost)算法的D5000系统健康度评价模型。首先，针对系统各评价等级样本数目严重不平衡的问题，提出一种自适应随机平衡(RB)的混合采样方法，分别以等级间样本数目的最大值、最小值作为采样区间的上、下限，生成多个随机数对各等级样本数据进行欠采样或过采样，增加训练数据的多样性并降低其不平衡程度；然后，训练平衡后的样本数据，建立极端梯度提升(XGBoost)算法子模型，考虑到各子模型重要度的一致性，提出采用硬投票方式集成所有子模型，得到与D5000系统各子模块对应的评价模型；最后，根据该系统指标层级关系，在评价过程中采用并、串行结合的计算方式，构建包含17个RB-XGBoost模型的D5000系统整体健康度评价模型。8组KEEL数据库中多类不平衡数据集的实验结果表明，与现有同类典型方法相比，所提方法的平均分类准确率最高提升了6.79 %，平均提升了2.03 %；某网省级D5000系统的实时采集数据验证了所提方法的有效性。 |
| 关键词: 智能电网 D5000系统 健康度评价 多分类 自适应随机平衡采样 XGBoost算法 |
| DOI：10.16081/j.epae.202001024 |
| 分类号:TM734 |
| 基金项目: |
|
| Health evaluation model of smart grid dispatch and control system based on RB-XGBoost algorithm |
|
TAN Lintao1, LI Junliang2, REN Bing3, HE Yang3, GAO Xin3, XU Jianhang2, HUANG Qingqing3
|
|
1.Dispatching and Control Center, Central China Branch of State Grid Corporation of China, Wuhan 430077, China;2.State Grid Electric Power Research Institute, NARI Group Corporation, Beijing 100192, China;3.School of Automation, Beijing University of Posts and Telecommunications, Beijing 100876, China
|
| Abstract: |
| For the health evaluation of smart grid dispatch and control system(D5000 system),the traditional expert experience-based evaluation methods have the problem of subjectivity. The multi-classification method based on machine learning is an effective way to improve the evaluation objectivity. However, the data imbalance among the samples of different health levels leads to a low classification accuracy. Therefore, a health evaluation model of D5000 system based on RB-XGBoost(Random Balance and eXtreme Gradient Boosting) algorithm is proposed. Firstly, in view of the serious imbalance of sample number among evaluation levels, a hybrid sampling method based on adaptive RB(Random Balance) is proposed. The maximum and minimum sample number among levels are used as the upper and lower limits of sampling interval respectively, and multiple random numbers are generated to undersample or oversample the data of each level, which increases the diversity and decreases the imbalance degree of training data. Then, after training the balanced sample data, sub-models of XGBoost(eXtreme Gradient Boosting) algorithm are established. Considering the consistency of each sub-model’s importance, a hard voting method is proposed to integrate all sub-models, and then the evaluation model corresponding to each sub-module of D5000 system is obtained. Finally, according to the hierarchical relationship of the system indicators, a parallel-serial combined calculation method is adopted in the evaluation process. In this way, the overall health evaluation model of D5000 system including 17 RB-XGBoost models is constructed. The experimental results based on eight sets of KEEL multi-class imbalanced datasets show that the average classification accuracy of the proposed method increases by 6.79% at the most and 2.03% on average compared with the existing similar typical algorithms. And the effectiveness of the proposed method is verified by the actual sampling data of a provincial D5000 system. |
| Key words: smart grid D5000 system health evaluation multi-classification adaptive random balance sampling XGBoost algorithm |
