海南岛秋季旱涝的模糊神经网络预测模型研究

Research on Fuzzy Neural Network Prediction Model for Autumn Drought-flood in Hainan

利用1961～2008年海南岛18个市县观测站的秋季降水量和NCEP再分析资料,通过合成分析和相关分析方法,选取前期海温场、位势高度场、风场、相对湿度场等环境场中影响海南岛秋季旱涝关键区作为因子,经过因子的降维处理,建立海南岛秋季旱涝的模型。结果表明,综合考虑多个环境场因子构建的模型优于单个或少数环境场因子构建的模糊神经网络预测模型,其对历史样本的拟合精度高,独立样本对趋势预测的准确率高而稳定,对特旱和特涝年也有较好的预测能力,在新评分标准下2004～2008年的独立样本预测检验评分达86,可以投入业务使用。

With the data of autumn rainfall in 18 meteorological observatories in Hainan and reanalyzed data of NCEP of 1961-2008,using comprehensive and correlation analysis,after dimension reduction,a fuzzy neural network prediction model for autumn drought-flood in Hainan was developed taking previous SST,geopotential height,wind and relative humidity that effect the autumn drought-flood as factors.The results showed that the fuzzy neural network prediction model with multi-factors is superior to those with single or several factors,and it shows advantages such as high fit precision and stability to history samples,especially for the independent samples,and with good prediction ability for severe drought and severe flood years.The tests for independent samples in 2004-2008 with new criterion reaches 86 and it is of great help for operational prediction.