机构地区: 华南农业大学工程学院南方农业机械与装备关键技术省部共建教育部重点实验室
出 处: 《计算机工程与应用》 2011年第34期228-230,248,共4页
摘 要: 碳通量(FC)作为全球二氧化碳循环与排放的重要指标,同各种生态因素有着密切的关系,因此可以通过各种生态因素预测碳通量,但迄今还缺乏有效的预测方法。研究脊波和神经网络结合的模型在碳通量预测中的应用,利用脊波处理碳通量数据的超平面奇异特性,从隐含层节点个数、误差、相关性等方面和小波网络进行了比较。实验结果表明,所采用的模型隐含层节点个数更少,拟合精度更高,预测能力更强,收敛速度更快。 Carbon Flux(FC) is an important index for global carbon dioxide circulation and emission.It is closely related to various ecological factors which can predict carbon flux.However,there have been no effective methods for prediction of FC so far.This paper researches the application of model combined with ridgelet and neural network in predicting FC.This model deals with hyperplane singularities of FC data with ridgelet.The ridgelet network is compared with wavelet network on number of hidden layer nodes,error,and correlation and so on.The experimental results show that ridgelet network has fewer number of hidden layer nodes,higher fitting precision,stronger predicted ability and faster convergent rate.
领 域: [自动化与计算机技术] [自动化与计算机技术]