基于粗糙集的模糊神经网络降水预报模型研究

A Fuzzy Neural Network Precipitation Model Based on Rough Set

为了提高模糊神经网络预测模型的预测能力,提出了基于粗集理论方法选择自变量作为模型输入的模糊神经网络预报建模方法。以短期降水预报作为研究对象,利用粗集理论的条件属性约简计算分析方法,对初选得到的预报因子矩阵进行属性约简,剔除不相关的属性,找出与预报量直接相关的预报因子,建立模糊神经网络的降水预报模型。实际的预报试验结果表明,该预报方法的预报精度明显高于由逐步回归方法选择预报因子作为模型输入的模糊神经网络预报模型及中国气象局T213数值预报模式的预报结果。

To improve the predictive ability of a fuzzy neural network （FNN） prediction model, the paper presents a prediction modeling methodology which combines the rough set attribute reduction with FNN. Taking short - term precipitation prediction as research objects and using attribute reduction to compute and analyse the original selected matrix of forecast factors, the scale of network is effectively reduced and the prediction ability of the FNN prediction model is evidently enhanced. Result shows that the FNN model based on rough set is superior to the conventional regression and FNN prediction model based on regression and CMA T213 numerical prediction model. Eventually, the merits of the rough set attribute reduction and FNN techniques are explained.