极端温度气象日母线日前负荷预测

Day-ahead Bus Load Forecasting in the Extreme Temperature Meteorological Days

极端高温、低温等气象条件下,母线负荷历史数据少,且受用户用电行为影响大,预测困难。为提高极端高、低温环境下母线日前负荷预测精度,提出一种计及多因素差异化影响的改进极端梯度提升(Xtreme Gradient Boosting,XGBoost)极端温度气象日母线日前负荷预测新方法。首先,以原始高维度特征集合构建XGBoost预测器,并在训练过程中获得特征重要度,分析极端温度气象日下母线与温度、降雨量等气象因素相关性;然后,构建基于XGBoost的母线日前负荷滚动预测器,以预测精度为决策变量确定母线日前负荷预测最优特征子集;最后,根据最优特征子集,针对性构建最优母线日前负荷预测模型。以东北某地区多条实际母线负荷数据验证新方法有效性,结果表明,新方法显著提高极端温度气象日下预测精度。

Under extreme high temperature,low temperature and other meteorological conditions,the historical data of bus load is few,and is greatly affected by the user’s electricity consumption behavior,so it is difficult to predict.In order to improve the day-ahead load forecasting accuracy of buses in extreme high and low temperature environments,a new method for day-ahead load forecasting of XGBoost extreme temperature meteorological buses is proposed,which takes into account the influence of multi-factors differentiation.Firstly,the XGBoost predictor is constructed based on the original high-dimensional feature set,and the feature importance is obtained during the training process.Then,the correlation between extreme temperature meteorological busbar and temperature,rainfall,wind speed and other meteorological factors is analyzed,and the order of feature importance is obtained;On this basis,a rolling bus day load predictor based on XGBoost is constructed by using forward feature selection method and different feature subsets.The optimal feature subset of bus day load forecasting is determined by taking prediction accuracy as decision variable;In the process of feature selection,the XGBoost parameters are optimized by grid search to avoid the influence of unreasonable parameters on the prediction accuracy of each predictor;Finally,according to the optimal feature subset of buses,a day-ahead load forecasting model of the optimal buses is constructed.The validity of the new method is validated by the actual bus load data under extreme temperature meteorological conditions in a certain area of Northeast China.The results show that the new method is more accurate than other methods in extreme temperature meteorological day prediction.