基于结合聚类与SVM参数寻优的短期电力负荷预测方法

A Short-Term Power Load Forecasting Method Based on Combining Clustering and SVM Parameter Optimization

预先作出准确的负荷预测有助于稳定电网市场,是电网安全调度与平稳运行的基础。考虑到目前研究的单一预测方法存在的不足,文章探讨了负荷变化趋势与温度、日类型等特征之间的关系,并提出了一种K-Means聚类与交叉验证和网格优化相结合的支持向量机(support vector machine,SVM)短期负荷预测方法(STLF-SK)。所提方法包括负荷数据预处理、K-Means聚类、模型参数优化以及模型的训练等步骤。在实际电力负荷数据集下,与LSTM、决策树和线性回归算法进行对比,实验结果表明,提出的STLF-SK方法有更高的预测准确性和平稳性,取得了较好的预测效果。

Making accurate load forecasting in advance helps stabilize the power market and is the basis for the safe dispatching and smooth operation of the power grid.However,considering the shortcomings of the single prediction method currently studied,this paper explores the relationship between load variation trend and certain characteristics such as temperature and day types,and proposes a SVM(support vector machine)short-term load forecasting method with the combination of K-Means clustering and parameter optimization by means of cross-validation and grid search(STLF-SK).The process of this method includes load data preprocessing,K-Means clustering,model parameter optimization,and model training.The simulation is completed under the data set of an actual district,and LSTM,decision tree and linear regression algorithm are selected for the experiment to make a comparison.The results of all experiments show that the method(STLF-SK)is more stable and has higher forecasting accuracy to achieve better forecasting effect.