摘要
为了避免由因素冗余导致的预测精度下降,对比分析了6种集成机器学习模型的性能,发现梯度提升决策树回归模型性能最好。利用梯度提升决策树进行特征重要性排序,选出显著影响因素;然后通过计算偏依赖量来评估各影响因素与最大负荷之间的非线性关系;最后,运用长短期记忆预测模型对各个因素的组合进行验证。结果表明,利用梯度提升决策树可以有效捕捉最大负荷与各因素之间的非线性关系,且经过因素选择和考虑温度累积效应后,负荷预测准确度得到显著提高。
In order to avoid the degradation of prediction accuracy caused by factor redundancy,the performance of 6 integrated machine learning models was compared and analyzed,among which gradient boosting decision tree(GBDT)regression model had the best performance.In order to select the significant influencing factors,the gradient-lifting decision tree is used to rank the feature importance,and then the nonlinear relationship between the influencing factors and the maximum load is evaluated by calculating the partial dependence.Finally,the combination of these factors was tested by long and short-term memory prediction model.The results show that the non-linear relationship between the maximum load and the factors can be effectively captured by the decision tree with gradient elevation,and the accuracy of load forecasting improves significantly after factor selection and consideration for the accumulative affect of temperature.
作者
邹鑫
罗涓
ZOU Xin;LUO Juan(Department of Economic Management,North China Electric Power University,Baoding 071003,China)
出处
《电力科学与工程》
2024年第3期10-19,共10页
Electric Power Science and Engineering
基金
国家自然科学基金资助项目(72171081)
中央高校基本科研业务费资助项目(2023MS153)。
关键词
新型电力系统
负荷预测
梯度提升决策树
长短期记忆
非线性影响
new power system
load forecasting
gradient lifting decision tree
long and short-term memory
nonlinear influence