基于ARMA-BP集成的藻类叶绿素a预测研究

Prediction on algal chlorophyll “a” by integrating ARMA and BP neural network

藻类叶绿素a浓度是反映太湖水体富营养化程度的重要参数指标。以太湖2010—2011年5—10月旬平均叶绿素a浓度和气象资料数据作为建模样本,通过对气象资料进行主成分分析,得到4种主要气象因子作为输入,建立时间序列ARMA预测模型与BP神经网络预测模型,并对2012年数据进行预测。利用两种模型在线性空间和非线性空间的预测优势,将叶绿素a数据结构分解为线性自相关主体和非线性残差两部分。首先用ARMA模型预测序列的线性主体,然后用BP模型对其非线性残差进行估计,最终集成整个序列的预测结果,建立了ARMA-BP预测模型。3种模型的预测效果为ARMA-BP>BP>ARMA。

The concentration of algal chlorophyll ＂a＂is an important parameter index reflecting the degree of eutrophication in Taihu Lake. By taking the mean dekad chlorophyll＂a＂concentration and meteorological data of Taihu Lake from May to October during 2010-2011 as the training samples and the four main meteorological factors derived by principal component analysis of the meteorological data as the input,the ARMA time series and BP neural network prediction models were built to predict the data in2012. The prediction advantages of the two models in linear and nonlinear space were used to decompose the structure of chlorophyll ＂a＂into linear autocorrelation and nonlinear residuals. Firstly,ARMA model is used to predict the linear component of the sequence; secondly,the nonlinear residuals component is predicted by BP model; finally,the predicted results of the entire series were integrated and establish the ARMA-BP prediction model on the data. The predicting effects of the three models can be expressed in a simple way as follows ARMA-BP BP ARMA.