基于多核模糊C均值聚类的配电网短期负荷预测

Short-term Load Forecasting of Distribution Networks Based on Multiple Kernel Fuzzy C-means Clustering

精准、高效的短期负荷预测是电力系统运行与调度的基础,负荷-气象因素的强耦合关系使得负荷预测过程中必须考虑气象因素。首先从影响电力负荷波动的气象因素出发,分析负荷样本数据的气象因素相关性,通过构造多核模糊C均值聚类函数实现负荷、气象数据的低维非线性至高维线性空间映射,完成基于负荷影响因素的聚类划分,获得强相关气象因素。接着,在传统LSTM(长短期记忆)神经网络中引入反馈环节,融合前向和反向计算机制消除LSTM训练过程的累计误差,构建基于深度学习的多层堆叠模式并应用于负荷预测中。然后,以历史负荷数据的聚类结果为训练样本,深度挖掘负荷-气象因素的耦合特征,从而提高负荷预测精度。最后,通过实际运行数据验证提出方法的合理性和准确性。

Accurate and efficient short-term load forecasting is the basis of power system dispatching and operation.It is necessary to consider the meteorological factors in the process of load forecasting since the strong coupling relationship between load and meteorological factors. Proceeding from meteorological factors that influence power load fluctuations,the correlation between load data samples and meteorological factors is analyzed;besides,the lowdimensional nonlinearity of meteorological data is mapped into the high-dimensional space based on multiple kernel fuzzy C-means(MKFCM) clustering algorithm. Furthermore,the meteorological factors are divided into several types. Based on the traditional LSTM(long and short-term memory)neural networks, a feedback mechanism integrating forward and reverse calculation is introduced to eliminate the cumulative error of the LSTM training process.The multi-layer stacking mode of LSTM based on deep learning is used in load forecasting. The clustering results of historical load data are used as training samples to deeply mine the coupling characteristics of load and meteorological factors and thus improve the accuracy of load forecasting. Finally,the rationality and accuracy of the proposed method are verified by using the actual operational data.