【摘要】 提出使用最大化最小概率机器回归MPMR方法来解决电力负荷中期预测问题,结合EUNITE网络2001年举办的电力负荷预测比赛提供的数据进行相关数据分析,建立相应的电力负荷中期预测模型。在分析最大化最小概率机器MPM及用于回归的原理的基础上,使用MPMR方法对不同的训练样本集进行训练,并计算出预测期的预测值,取得了理想的预测结果。核函数形状参数θ及回归管道宽度ε的选取将直接影响MPMR回归技术的系统性能,试验中使用了交叉验证的方法确定这2个参数值。通过交叉验证求得,当核函数形状参数和回归管道宽度取值都为3时预测效果更为理想。文中对试验结果进行分析,并与相关向量机等其它方法比较,总结了MPMR预测方法的特点。更多还原

【Abstract】 To improve mid-term load-forecasting accuracy, using Minimax Probability Machine Regression (MPMR), a brand-new technology,is proposed. The regression function maximizes the minimum probability that future predication will be within anεto the true regressionfunction. With competition data from EUNITE Network, the model for load forecasting is built after analyzing data, and the experiment ofchoosing different train samples is done with good results. The experimental results with other methods are compared and some explanationsare presented. Kernel function shape parameter and regression tube value will influence the MPMR-based system performance. In experiments,cross validation to choose the two parameters is used. In the end of the paper, the features of MPMR, comparing with other methods, such asthe Relevance Vector Machine, etc. are summarized.更多还原