基于核主成分分析和CSM-PSCO优化LSSVR的短期负荷预测

Short-term Load Forecasting Based on Kernel Principle Component Analysis and Optimized LSSVR due to CSM-PSCO

分析了一种基于核主成分分析(kernel principle component analysis,KPCA)和混沌单纯形混合粒子群协同(chaos and simplex method-particle swarm coordinate optimization,CSM-PSCO)算法优化最小二乘支持向量回归机(least square support vector regression,LSSVR)的短期负荷预测模型。首先,采用KPCA对训练样本的输入个数进行降维优选,以较少输入代替原始大量输入,同时信息大部分得以保留;然后,采用LSSVR对训练样本进行回归训练,训练过程中采用CSM-PSCO对LSSVR的相关参数进行优化,得到满足要求的模型;最后,采用训练好的模型对未知负荷进行预测。算例表明该模型的预测精度和速度均能满足实际的预测需求。

This paper analyzes a kind of short-term load forecasting model based on kernel principle component analysis（KP-CA）and least square support vector regression（LSSVR）due to chaos and simplex method coordinating with particle swarm optimization. Firstly,KPCA was applied for dimension reduction optimization on input numbers of training samples so as to replace original great deal of input information by less input and retain most information. Then,LSSVR was used for regres-sion training on samples and CSM-PSCO was used for optimizing relevant parameters of LSSVR in process of training in or-der to obtain satisfied model. Finally,the well-trained model was used to forecast unknown load. Examples indicate that forecasting precision and velocity of this model could meet practical forecasting requirements.