基于模型组合的电力负荷精准预测
Accurate Prediction of Power Load Based on Model Combination
仲浩帆 1黎雅红 2朱恩豪2
作者信息
- 1. 中南民族大学数学与统计学学院,湖北武汉 430074
- 2. 中南民族大学计算机科学学院,湖北武汉 430074
- 折叠
摘要
电力负荷的预测涉及环境和社会因素,因此,设计一种高效准确的预测模型是电力行业的重要任务.设计了一种基于传统机器学习模型组合的电力负荷预测模型.该模型从特征层面入手并采用Wrapper与梯度的特征排序选择方法对主要特征进行筛选和优化.在实验中,对预测结果进行了基于时序扩展窗口拆分的K折时序交叉验证和比较.结果表明,该模型能够有效预测电力负荷的短期时序变化,且预测效果比传统的单一机器学习模型好.通过贝叶斯超参数调优方法得到的超参数组合能够显著提高模型的准确性和泛化能力.这说明贝叶斯超参数调优方法能够在一定程度上解决模型过拟合和欠拟合问题,提高模型的稳定性和可靠性.
Abstract
Electricity load forecasting involves numerous environmental and social factors.Therefore,designing an efficient and accurate prediction model has been an important task in the power industry.In this paper,a power load forecasting model based on a combination of traditional machine learning models is designed.The model starts from the feature level and adopts the feature ranking selection method of Wrapper and Gradient to filter and optimize the main features.In the experiments,the prediction results are cross-validated and compared with the K-fold timing based on the temporal expansion window splitting.The results show that the designed model can effectively predict the short-term time-series variation of power loads,and the prediction results are better than the traditional single machine learning model.The combination of hyperparameters obtained through the Bayesian hyperparameter tuning method can significantly improve the accuracy and generalization ability of the model.This indicates that the Bayesian hyperparameter tuning method can solve the model overfitting and underfitting problems to a certain extent,and improve the stability and reliability of the model.
关键词
短期电力负荷预测/机器学习/贝叶斯超参数调优/时序交叉验证/时间序列预测Key words
short-term power load forecasting/machine learning/Bayesian hyperparameter optimization/time series cross validation/time series prediction引用本文复制引用
基金项目
国家级大学生创新创业训练计划(2022)(GCX2230)
出版年
2024
NETL
NSTL
万方数据