基于改进Prophet模型的用电量预测研究  

Research on Power Consumption Forecasting Based on Improved Prophet model

作　　者：吴文培 宋亚林[1] 魏上斐 WU Wen-pei;SONG Ya-lin;WEI Shang-fei(Institute of Intelligent Network System,Henan University,Kaifeng Henan 475000,China)

机构地区：[1]河南大学智能网络系统研究所,河南开封475000

出　　处：《计算机仿真》2021年第11期473-478,共6页Computer Simulation

基　　金：河南大学研究生教育创新与质量提升项目-研究生教育创新培养基地子项(SYL18020105)。

摘　　要：针对一些电量预测模型缺乏对季节、节假日等特殊时间节点处理的过程,以及单一Prophet模型在节假日等特殊时间点易陷入过拟合,导致预测置信值偏差较大的问题,提出了一种改进Prophet预测方法。方法在原有Prophet模型的基础上,增加了一个预测值优化模块,通过模块融合Prophet模型和XGBoost模型,利用XGBoost模型正则化项可有效防止过拟合的特点,对不同子模型的预测结果进行加权求和,减小预测误差。实验结果表明,与原始Prophet预测模型相比,改进后的预测模型X-Prophet有效减小了预测误差。Some power forecasting models lack the process of dealing with special time nodes such as seasons and holidays,and the single Prophet model is apt to fall into over-fitting at special time points such as holidays,which leads to the problem of large deviation of prediction confidence value.This paper presents an improved prediction method of Prophet,which is based on the original Prophet model and adds a prediction value optimization module,through which the Prophet model and the XGBoost model are fused.The regularization term of the XGBoost model can prevent over-fitting effectively,and the prediction results of different sub-models can be weighted to sum and the prediction error can be reduced.The experimental results show that compared with the original Prophet prediction model,the improved X-Prophet prediction model reduces the prediction error effectively.

关 键 词：特殊时间点 过拟合 模型融合 用电量预测 

分 类 号：TP391.9[自动化与计算机技术—计算机应用技术]