摘要
月降水量的准确预测,对于国民生产、防灾减灾有重大意义。针对月降水量影响因素多,彼此之间关系复杂,预测难度大的特点,为提高预测精度,提出了一种基于相空间重构的BP神经网络月降水量预测方法。对一维月降水量时间序列进行参数提取,重构相空间,得到包含各态历经信息的多维序列,用训练神经网络进行月降水量预测,可以解决神经网络难以确定网络结构的问题。应用于杭州市月降水量预测,实验结果表明,改进方法具有较高的预测准确度,可以为干旱洪涝灾害预测提供依据。
Accurately prediction of monthly precipitation has significant meaning for the national production, dis- aster prevention and mitigation. Due to so many factors influencing monthly precipitation, the relationship between each other is very complex, and the monthly precipitation prediction is a difficult field in climate prediction. In order to improve the prediction accuracy, a monthly precipitation prediction method was proposed combining phase space reconstruction with BP neural network in this paper. Parameters of one-dimension monthly precipitation time series were extracted, one-dimension precipitation time series was developed to multi-dimension time series with recon- struction of phase space, and the multi-dimension series include ergodic information. It is conducive to the neural network training, and assisted to determine the structure of neural network. Applied to Hangzhou monthly precipitati- on forecasts, the experimental results show that a higher accuracy rate of prediction can be achieved, and provide the basis for drought and floods forecast.
出处
《计算机仿真》
CSCD
北大核心
2014年第1期352-355,共4页
Computer Simulation
基金
江苏省高校优势学科建设工程资助项目(PAPD)
中国气象局软科学研究课题(SK20120146)
南京市产学研资金项目(2012t026)
公益性行业(气象)科研专项资助项目(GYHY201106040)
关键词
相空间重构
神经网络
月降水量
预测
Phase space reconstruction
Neural network
Monthly precipitation
Prediction
作者简介
张颖超（1960-），男（汉族），江苏省沛县人，教授，博士生导师，主要研究方向为复杂系统建模与评估、模糊理论与应用、计算机网络等；
刘玉珠（1989-），女（汉族），河南省睢县人，硕士研究生，主要研究方向为预测技术。
