基于叠加马尔科夫链和BP神经网络模型的降水量预测研究  

The Prediction of Rainfall Based on Superimposed Markov Chains and BP Neural Network Model

作　　者：王宇博[1] 梁秀娟[1] 乔雨[1] 王亮[1] 徐海岩[1] 陈伟[1] 

机构地区：[1]吉林大学地下水资源与环境教育部重点实验室,长春130021

出　　处：《中国农村水利水电》2014年第9期80-82,86,共4页China Rural Water and Hydropower

基　　金：吉林省科技厅重点攻关项目(20100452);国家潜在油气资源产学研用项目专题(20100331-OSR01-7);吉林省重点科技攻关项目(20100452);教育部高等学校博士点基金项目(200801830044)

摘　　要：选用叠加马尔科夫链和BP神经网络模型,利用1951年-2013年的降水资料对长春市降水量进行预测。叠加马尔科夫链方法具有原理简单易懂,计算方法简便的优点。BP神经网络模型具有计算快速,占用内存小,还有很好的容错性。根据计算,BP神经网络预测误差均小于20%,叠加马尔科夫链方法仅有6个年份误差小于20%。通过对比分析预测结果可知:对于本研究区BP神经网络模型的预测精度较高。预测方法对数据的稳定性有要求,数据的波动越强,预测结果精度越低。对比可知,叠加马尔科夫链对数据的平稳性要求更高,BP神经网络模型的适用性更强。This paper superimposes Markov chains and BP neural network model to predict rainfall in Changchun City based on the precipitation data from 1951 to 2013 .The method of superimposed Markov chains has the benefits of being easy to understand and to calculate .The BP neural network model can give the result quickly and it＇s precise and credible .According to calculation ,BP neural network＇s predictions error are all less than 20% ,in the predictions of superimposed Markov chains ,only six years＇prediction errors are less than 20% .By comparing and analyzing the results of prediction ,BP neural network model＇s prediction accuracy is relatively high for this study area .It can be seen ,prediction methods require the stability of the data .The stronger the fluctuations in the da-ta ,the lower the prediction accuracy .In contrast ,superimposed Markov chain has higher requirements of stability on the data ,and BP neural network model is stronger in applicability .

关 键 词：降水量预测 叠加马尔科夫链 BP神经网络 

分 类 号：P338[天文地球—水文科学]