基于LSTM-XGBoost和多模型算法的短期负荷预测  

Short-term load predicting based on LSTM-XGBoost and multi-model algorithm

作　　者：邵必林[1] 庄雪莉 曾卉玢 Shao Biin;Zhuang Xuei;Zeng Huibin(School of Management,Xi'an University of Architecture and Technology,Xi'an,Shaanxi 710055,China)

机构地区：[1]西安建筑科技大学管理学院,陕西西安710055

出　　处：《计算机时代》2023年第12期49-54,共6页Computer Era

基　　金：国家自然科学基金面上项目“面对不确定因素的天然气负荷预测及用户行为检测方法研究”(No.62072363)。

摘　　要：针对负荷数据波动性强、特征存在冗余而导致使用单一模型预测短期负荷时精度较低的问题,提出一种融合梯度提升树(GBDT)、自适应噪声完备集合经验模态分解(CEEMDAN)、长短期记忆(LSTM)和极端梯度提升(XGBoost)的短期负荷预测组合方法。首先利用GBDT对负荷数据集进行特征选择,筛选出重要特征;然后使用CEEMDAN将负荷序列分解后合并为低频分量和高频分量;再将低频分量输入到LSTM中进行预测,将高频分量输入到XGBoost中进行预测;最后,短期负荷的最终预测结果由两个模型的预测结果进行叠加而成。与单一预测模型相比,所提方法在短期负荷方面具有更高的准确性。To address the problem of low accuracy in predicting short-term load using a single model due to the high volatility of load data and redundancy of features,a combined method for short-term load prediction is proposed that combines GBDT,CEEMDAN,LSTM,and XGBoost.Firstly,GBDT is used to feature select the load dataset to filter out important features.Next,the load sequence is decomposed and merged into low-frequency and high-frequency components using CEEMDAN.Then,the low-frequency components are input into LSTM and the high-frequency components are input into XGBoost for prediction.Finally,the final prediction results of short-term load are made by superposing the prediction results of the two models.Compared with the single prediction model,the proposed prediction method has higher accuracy in the short-term load.

关 键 词：长短期记忆 极端梯度提升 短期负荷预测 自适应噪声完备集合经验模态分解 梯度提升树 

分 类 号：TM715[电气工程—电力系统及自动化]