基于CLPSO模型选择的SVM电力负荷预测模型研究

Power Load Forecasting of SVM Based on CLPSO Algorithm Model Selection Method

电力负荷数据具有非线性、非平稳的特点,基于机器学习的预测技术始终是负荷预测领域的研究热点。而模型选择,即包括参数优化、特征选择等一系列可以使模型达到最优状态的操作,是提高机器学习负荷预测模型性能的关键。为此,提出了一种基于二进制CLPSO算法的模型选择一体化策略,整合特征选择和参数优化过程,以提高SVM预测方法的自适应性。并以GEFCom2012竞赛的电力负荷数据为例进行实验,证明了所提出的一体化模型选择框架能够有效提高SVM模型的预测精确度。

Machine learning-based forecasting technology remains a research hotspot in the field of load forecasting due to the nonlinear and non-stationary characters of load data.Model selection,including parameter optimization,feature selection and other operations to optimize the model performance,is the key to improve the outcome of machine learning load forecasting model.In this study,a model selection integration strategy based on binary CLPSO algorithm is proposed to unify feature selection and parameter optimization processes,which promotes the adaptability of SVM prediction methods.Taking the power load data of GEFCom2012 competition as an instance,experiments prove that the proposed integrated model selection framework can effectively improve the prediction accuracy of SVM model.