基于模糊Bagging-GBDT的短期负荷预测模型研究

Study on Short-term Load Forecasting Model Based on Fuzzy Bagging-GBDT

为了提高常规梯度提升决策树GBDT算法的泛化性能,并实现并行计算,在GBDT的基础上,利用隶属度函数对气象数据进行模糊处理,同时引入Bagging算法,通过Bootstrap方式对原始数据进行多次抽样形成新的训练样本,分别训练模糊GBDT负荷预测子模型,提出了基于模糊Bagging-GBDT的短期负荷预测模型。算例分析结果表明,本文提出的预测模型相较于BP-NN和常规GBDT预测模型,7日平均绝对误差分别降低了1.44%和0.22%,模型具有良好的预测精度和稳定性。

To improve the generalization performance of the traditional gradient boosting decision tree(GBDT)and realize parallel computation,the membership function is applied to fuzzy processing of meteorological data on the basis of GBDT. Meanwhile,the Bagging algorithm is introduced. Using the Bootstrap method,the original data are sampled for multiple times to form new training samples,which are further used to train the fuzzy GBDT load forecasting sub-models. In this way,the short-term load forecasting model based on fuzzy Bagging-GBDT is proposed. The result of a numerical example indicates that compared with the BP-NN model and the traditional GBDT forecasting model,the 7-day mean absolute error obtained using the proposed forecasting model is reduced by 1.44% and 0.22%,respectively,showing that the novel model has satisfying forecasting accuracy and stability.