摘要
采用BDS、Box-Pierce和Ljung-Box方法对1995-2005年渤海化学耗氧量(COD)浓度历史监测数据的独立性进行了检验,采用代替数据检验IAAFT、White和Terasvirta人工神经网络弱非线性检验、Hinich双谱检验以及无Fourier变换检验进行了相关性的非线性检验,判断出渤海COD不独立且线性相关。综合比较线性ARMA、局部线性、规则集成、随机森林、随机梯度Boosting、支持向量、人工神经网络、自适应样条8种预测方法,结果表明线性ARMA模型误差均值和方差最小,证明线性ARMA模型更适合分析渤海COD数据,同时证实了前述各种非线性检验结果的可靠性。
Dependence was tested in Bohai COD from 1995 to 2005 by BDS, Box-pierce and Ljung-Box independence test. Surrogate data test/AAFT for nonlinearity, White＇s and Teravirta＇s artificial neuron network test for neglected nonlinearity, Hinich＇s bispectrum test and a test without Fourier transfer were carried out, and no nonlinearity was made out. It was judged that Bohai COD concentration was of non-dependence and linear correlation. The mean errow and standard error of linear ARMA model were the minimum comparing local linear, rule ensemble, random forest, stochastic gradient boosting, support vector, artificial neuron network and multivariate adaptive regression splines. So linear ARMA model fitted the best to analyze Bohai COD data, which testified former tests result.
出处
《海洋通报》
CAS
CSCD
北大核心
2009年第1期96-101,共6页
Marine Science Bulletin
基金
国家自然科学基金资助项目(10772132)
博士点基金项目(20070056063)
