论文部分内容阅读
在介绍广义回归神经网络(GRNN)基本算法、网络结构及平滑参数确定方法的基础上,提出将误差序列的均方值作为网络性能的评价指标并采用最小误差对应的平滑参数,建立了GRNN的预测模型。提出了确定输入神经元数目的方法:根据自回归模型阶次的选择经验初步确定输入神经元数目m;在m值附近进行搜索,对于每一个m值,确定平滑参数后,计算网络对学习样本的预测误差;根据BIC准则评价指标的最小值确定输入神经元数目。将模型应用于某地中长期电力网负荷预测,分别进行了单步预测和多步预测。与BP神经网络模型的预测进行比较,结果表明,采用该方法的预测精度明显高于BP模型,即使在训练集样本数据较少时,该方法的预测准确度仍然很高。
After introducing the basic algorithm of GRNN, the network structure and the method of determining the smoothing parameters, we propose to use the mean square value of the error sequence as the evaluation index of the network performance and the smoothed parameters corresponding to the minimum error to establish the GRNN Predictive model. A method of determining the number of input neurons is proposed. The number of input neurons is initially determined according to the choice experience of autoregressive model order. The number of input neurons m is searched around m value. For each m value, after the smoothing parameters are determined, Of the prediction error; according to the minimum evaluation index BIC criteria to determine the number of input neurons. The model is applied to the load forecasting of medium- and long-term power grid in a certain area, and the single-step prediction and multi-step prediction are respectively carried out. Compared with the BP neural network model, the results show that the prediction accuracy of this method is obviously higher than that of the BP model, and the prediction accuracy of the method is high even when the training set sample data is small.