基于RBFNN的短期电力负荷混沌局域预测法

短期负荷序列数据表现出混沌特性,可以使用混沌时序局域方法进行预测。在混沌时序重构相空间中预则中心相点和趋势相点之间的映射关系不是单纯的线性关系,而常用的线性回归预测模型只能逼近线性映射。提出利用径向基函数神经网络(RBFNN)来建立预测模型,可以更加精确逼近预则中心相点和预测相点之间的映射关系,并用欧氏距离和关联系数联合方法选取近邻相点,选取的近邻相点与预测中心相点的关联性更好。利用西北电网的负荷数据所做的实验证明,本文提出的基于RBFNN的局域预测法比线性局域预测法获得了更为满意的预测精度。

Chaotic load series local forecasting based on RBFNN

Short-term load series possess the chaotic characteristic,and can be forecasted with chaotic series local model.The mapping between prediction phase point and future counterpart point is not absolute linear,but the common linear regression models can only approximate linear mapping. The paper puts forward a novel model based on Radial-Basis-Function Neural Network(RBFNN),which has higher performance on approximating the real mapping,and both Euclidian distance and correlative coefficients approaches are employed to select the neighbourhood phase points,which are more close to the prediction point.The experiment with load series with 24 points one day from Northwest Power Grid demonstrates that the novel model achieves better predication precision than the linear model.