优化组合核函数相关向量机电力负荷预测模型

Relevance vector machine based on particle swarm optimization of compounding kernels in electricity load forecasting

在单一核函数相关向量机模型的基础上,构建高斯核函数分别与多项式核函数和张量积线性样条核函数进行线性组合的多种组合核函数相关向量机中期电力负荷预测模型,并利用粒子群优化算法对组合核函数的各参数进行优化选择。以2001年组织的国际电力负荷预测竞赛提供的公开数据为训练和测试样本,分别对多种核函数相关向量机中期电力负荷预测模型进行仿真预测计算。结果显示,虽然各模型都取得了较好的预测精确度,但是基于组合核函数的相关向量机在各项评价指标上都优于基于单一核函数的相关向量机。还利用相关向量机的概率预测优势得到了其他模式识别模型无法得到的预测误差范围。

Based on the single kernel function relevance vector machine（RVM） models,it constructs multi-ple middle-time-load-forecasting models.The RVM’s kernel functions were linearly compounded by Gauss-ian kernel with polynomial kernel and tensor product spline kernel,and the compounding kernels’parame-ters are optimized by algorithm of particle swarm optimization（PSO）.With the training and testing sample data of 2001 world-wide competition of electricity load forecasting,all the models’forecasting value were given.The results show,although the every model has a good accuracy,all the multi-linearity-compoun-ding kernels RVM models give the better accuracy than the single kernel ones.Besides the forecasting error bar also be given based on the exclusive probability character of relevance vector machine.