基于优化聚类分解与XGBOOST的超短期电力负荷预测

Ultra-short-term power load forecasting based on optimal cluster decomposition and XGBOOST

电力负荷预测是制定发电计划和确保电网稳定运行的基础,为提高电力负荷预测精度,提出一种基于优化聚类分解与极限梯度提升(XGBOOST)的超短期电力负荷预测方法。一方面针对模糊C均值聚类(FCM)不能自动选择聚类数问题提出一种均值漂移(Mean Shift)优化FCM的优化聚类(MF);另一方面为减小电力负荷数据随机性对电力负荷预测的影响,提出一种结合MF、自适应噪声的完全集成经验模式分解(CEEMDAN)、XGBOOST的MFCX(MF-CEEMDAN-XGBOOST)的超短期负荷预测模型。首先使用Mean Shift搜寻到的最佳聚类数和聚类中心替换FCM的聚类数和初始聚类中心,对负荷数据聚类,然后采用CEEMDAN分解得到较为平稳的负荷分量,最后使用XGBOOST对新的负荷序列分别预测后进行模态重构得到最终预测结果。使用Python语言搭建模型进行实例分析与不同模型对比,MFCX有较低的预测误差,从而验证了模型的有效性。

Power load forecasting was the basis for formulating power generation plans and ensuring the stable operation of power grids.In order to improve the accuracy of power load forecasting,an ultra-short-term power load forecasting method based on optimal clustering decomposition and extreme gradient boosting(XGBOOST)was proposed.On the one hand,to solve the problem that fuzzy C-means clustering(FCM)cannot automatically select the number of clusters,an optimized clustering(MF)of Mean Shift optimized FCM was proposed;on the other hand,to reduce the influence of randomness of power load data on power load prediction,an ultra-short-term load forecasting model of MF-CEEMDAN-XGBOOST(MFCX)combining MF,CEEMDAN,and XGBOOST was proposed.First,Mean Shift was used to search for the optimal number of clusters and cluster centers.The number of clusters and the initial cluster centers of FCM were replaced to cluster the load data,and then CEEMDAN was used to decompose the load components to obtain a relatively stable load component.XGBOOST predicts the new load sequence separately and then performs modal reconstruction to obtain the final prediction result.Python was used to build the model for instance analysis and compared with different models.The prediction error of MFCX was lower than other models,thus verifying the validity of the model.