摘要
为提高水环境非线性时序预测模型的精度,用自相关技术分析水环境时间序列的延迟特性,确定径向基函数(RBF)网络的输入、输出向量,建立了水环境时间序列预测的高精度RBF网络模型。用32年海洋水温时间序列实测资料来训练和检验网络并用于预测。用该模型对长江流域望江楼站8年总硬度、高锰酸盐指数、五日生化需氧量、氨氮、溶解氧、挥发酚、镉、氯化物、硫酸盐等9种水环境要素时间序列进行预测。实例分析表明,所建模型预测误差均较小,好于门限自回归模型,BP神经网络模型和ELMAN神经网络模型。所建模型不仅精度高,而且收敛速度快。
In order to raise the precision of prediction model for water environment nonlinear time series, a high precision radial basis function （RBF） artificial neural network model is presented. The delay time of water environment time series is analyzed with auto-correlation technique. The input and output of this model are decided by this delay time. And this model is verified by two cases. First, the training and test are given by the recorded data of 32 years of marine water temperature, and the result shows that the error of every training sample is 0.00, and the relative error is 0.3875% of forecasting marine temperature based on the RBF network model. Then, the water environment elements, such as hardness, the salts of permanganic acid, BOD5, NH3-N, DO, phenol, cadmium, chloral and sulfate, are predicted at Wangjianglou in Changjiang river for eight years. The prediction precision of this RBF model is higher than that of the threshold auto-regression model, the BP artificial neural network model and the ELMAN artificial neural network model. The convergent speed of this new model is also faster than that of the BP model. It is a good nonlinear prediction model for water science time series.
出处
《水科学进展》
EI
CAS
CSCD
北大核心
2005年第6期788-791,共4页
Advances in Water Science
基金
国家重点基础研究发展计划(973)资助项目(2003CB415204)
国家科技攻关资助项目(2004BA611B020401)~~
关键词
水环境时间序列
非线性预测
RBF神经网络
精度
water environment time series
nonlinear prediction
RBF artificial neural network
precision
