基于改进Prophet算法的短期日负荷预测方法研究  

Short-term power load forecasting based on improved Prophet algorithm

作　　者：刘涅煊 杨学良 陶晓峰 黄福兴 陆春艳 LIU Niexuan;YANG Xueliang;TAO Xiaofeng;HUANG Fuxing;LU Chunyan(State Grid Hectric Reseerch Institute(NARI Group Corporation),Nanjing 211106,China)

机构地区：[1]国网电力科学研究院有限公司(南瑞集团有限公司),南京211106

出　　处：《电力需求侧管理》2022年第5期58-63,共6页Power Demand Side Management

基　　金：国家自然科学基金资助项目(61833008);国网电力科学研究院科技项目(524608220039)。

摘　　要：将Prophet算法引入负荷预测领域,并结合XGBoost算法提升Prophet负荷预测准确性。Prophet算法基于时间序列分解及机器学习的拟合,将负荷数据分解为趋势项、周期项、随机波动项3部分,引入XGBoost算法改进Prophet算法对随机波动项的预测,将XGBoost算法对随机波动的预测结果与Prophet算法对趋势项和周期项的预测结果叠加,获得最终的预测结果。该算法适用于用电负荷这种具备一定周期变化特征的序列,易于理解,预测准确性较高。通过某地区用电信息采集系统提供的专公变用户日冻结数据实验验证,结果表明在相同条件下,改进后的算法预测的结果的平均绝对误差百分比较原始的Prophet算法可降低2.5%,同时均方根误差降低幅度可达30.79%,体现出显著的改进效果。A new power load forecasting method based on optimized Prophet algorithm is presented. A power load time series can be disassembled into trend term, seasonal term and random term by Prophet algorithm. In the predicted time series, trend and seasonal term are predicted by Prophet, while the random term is replaced by the prediction output by XGBoost algorithm. The model proposed is easy to understand and requires only power load data, whose parameters can be adjusted by the analysts intuitively.The effectiveness of the forecasting method is proved by conducting experiments on the power load data obtained from power consumption information acquisition system. The result shows that the mean absolute percentage error of the forecasting result is reduced by up to 2.5% comparing with the original Prophet algorithm. At the same time, the root mean square error is reduced by up to30.19%, which show its excellent improvement.

关 键 词：短期负荷预测 可分解模型 预测精度 PROPHET XGBoost 

分 类 号：TM714[电气工程—电力系统及自动化] TK011[动力工程及工程热物理]