| 引用本文: | 庞传军,刘金波,张波,杨笑宇,余建明,刘艳.基于Shapley值的电力负荷预测结果溯源分析方法[J].电力自动化设备,2021,41(12): |
| PANG Chuanjun,LIU Jinbo,ZHANG Bo,YANG Xiaoyu,YU Jianming,LIU Yan.Traceability analysis method of power load forecasting results based on Shapley value[J].Electric Power Automation Equipment,2021,41(12): |
|
| 本文已被：浏览 8200次 下载 1149次 |
|
|
|
| 基于Shapley值的电力负荷预测结果溯源分析方法 |
|
庞传军1,2, 刘金波3, 张波1,2, 杨笑宇3, 余建明1,2, 刘艳1,2
|
|
1.南瑞集团有限公司(国网电力科学研究院有限公司)，江苏 南京 211106;2.北京科东电力控制系统有限责任公司，北京 100192;3.国家电网公司国家电力调度控制中心，北京 100031
|
|
| 摘要: |
| 针对由于机器学习的黑盒特性导致负荷预测结果不可溯源的问题，提出一种基于Shapley值的电力负荷预测结果溯源分析方法。阐述利用机器学习技术构建负荷预测模型的一般形式和基本过程；基于负荷预测模型，利用合作博弈论中的Shapley值计算各类负荷影响因素对负荷预测结果的影响；对利用梯度提升决策树算法训练的负荷预测模型的预测结果进行溯源分析。实验结果表明，利用所提方法可以洞察负荷预测过程，从而实现负荷预测结果的溯源分析以及考虑复杂非线性的负荷影响因素分析，也可以在构建负荷预测模型时指导特征选择提升模型的泛化能力。 |
| 关键词: 负荷预测 负荷影响因素 溯源分析 Shapley值 梯度提升决策树 机器学习 |
| DOI：10.16081/j.epae.202110001 |
| 分类号:TM73 |
| 基金项目:国家电网公司科技项目(5700-202055368A-0-0-00) |
|
| Traceability analysis method of power load forecasting results based on Shapley value |
|
PANG Chuanjun1,2, LIU Jinbo3, ZHANG Bo1,2, YANG Xiaoyu3, YU Jianming1,2, LIU Yan1,2
|
|
1.NARI Group Corporation Co.,Ltd.(State Grid Electric Power Research Institute Co.,Ltd.),Nanjing 211106, China;2.Beijing Kedong Electric Power Control System Co.,Ltd.,Beijing 100192, China;3.National Power Dispatching and Control Center of State Grid Corporation of China, Beijing 100031, China
|
| Abstract: |
| Aiming at the problem that load forecasting results cannot be traced due to the black-box characteristic of machine learning, a traceability analysis method of power load forecasting results based on Shapley value is proposed. The general form and basic process of building load forecasting model with machine learning technology are explained. On the basis of load forecasting model, the Shapley value in cooperative game theory is used to calculate the impact of various influencing factors of load on the load forecasting results. Traceability analysis of the forecasting results of load forecasting model trained by the gradient boos-ting decision tree algorithm is carried out. The experimental results show that the proposed method can gain insight into the load forecasting process, so as to realize the traceability analysis of load forecasting results and analysis of influence factors of load considering complex nonlinearity, and can guide feature selection to improve the generalization ability of model when building load forecasting model. |
| Key words: load forecasting influencing factors of load traceability analysis Shapley value gradient boosting decision tree machine learning |
