基于三元回声状态网络的二氧化碳浓度预测

The Forecasting of Carbon Dioxide Concentrations Based on Triplet Echo State Networks

针对二氧化碳浓度序列较强的周期性、非线性、混沌性等特征,提出了一种新型的三元回声状态网络二氧化碳浓度预测模型。首先,采用回声状态网络实现二氧化碳浓度序列的初步预测;然后,结合生态学、社会学等学科的先验知识,估算区域碳排放情况,利用三元模型对回声状态网络的输出结果进行修正,从而获得更精准的区域二氧化碳浓度预测结果;最后,进行了三元回声状态网络与支持向量机、主成分分析支持向量机在二氧化碳浓度预测上的仿真对比实验,实验结果表明,提出的三元回声状态网络方法比其它两种方法具有更高的预测精度,更高的可靠性和适用性。

Contraposing the characteristics of periodicity,nonlinearity and chaos of carbon dioxide concentration series,the paper proposes a new forecasting model of carbon dioxide concentrations,which is triplet echo state networks（ triplet ESN）. Firstly,we adopt the echo state network to get a preliminary prediction of carbon dioxide concentration series. Then,we estimate regional carbon emissions,according to prior knowledge of ecology,sociology and other subjects. And we utilize triplet model to revise the outputs of echo state network so that we can obtain more accurate forecasting results of regional carbon dioxide concentrations. Finally,comparative simulation experiments are done to compare forecasting results of carbon dioxide concentrations among triplet echo state networks,support vector machine and principal component analysis support vector machine. And experimental results show that the proposed triplet ESN can get highest forecasting precision and performs best in reliability and applicability among these three methods.