典型降水预报ANN多指标优选——以太湖湖西区为例

为提高多种典型人工神经网络应用于降水预报的精度与稳定性并做出优选,对太湖流域湖西区丹徒、丹阳、金坛、溧阳、宜兴5站的年降水量时间序列建立基于组成成分分析的人工神经网络模型,并通过平均相对误差、平均绝对误差、均方根误差及合格率4项评价指标对比分析预报效果。该模型采用Mann-Kendall法、秩和检验法、谱分析法进行组成成分分析;建立BP网络、小波神经网络、RBF网络、GRNN网络及Elman网络模拟并预测随机成分,与确定性成分叠加得年降水量预报结果。在湖西区的研究结果表明,基于组成成分分析的人工神经网络模型的拟合及预测精度高于原始人工神经网络和线性自回归模型,GRNN网络的预测精度与稳定性高于其他4类神经网络。

Multi-indexes Optimization of Typical Artificial Neural Networks for Rainfall Forecasting: A Case Study in West Taihu Lake Basin

In order to increase the precision of the typical artificial neural networks for rainfall forecasting and accomplish the optimization, this paper built the artificial neural networks based on the component analysis for the annual precipitation time series at 5 the representative rainfall stations in the west Taihu Lake basin, and evaluated the forecasting effects through mean absolute percentage error, mean absolute error, root mean square error and qualified rate. The model applied MK method, rank sum test and spectrum analysis to analyze the components. Then, BP neural network, wavelet neural network, RBF neural network, general regression neural network and Elman neural network were built to simulate and forecast the stochastic component. The annual rainfall forecasting results are the superposition of the stochastic and deterministic components. The results show that the fitting and predicting precision of the artificial neural networks based on the component analysis are superior than the original artificial neural networks and AR(p) model. Besides, the precision and stability of the general regression neural network are higher than the other four kinds of neural networks.