湖库富营养化人工神经网络评价模型

在分析现有应用人工神经网络评价模型局限性的基础上，根据湖库富营养化的评价标准，提出了生成BP神经网络训练样本、检验样本和测试样本的新方法，给出了区分湖库富营养化不同程度的分界值，论述了确定合理隐层及其节点数的方法，使得训练后的神经网络模型具有更强的泛化能力，不受初始连接权值的影响。训练后的评价模型应用于实例的评价结果表明，新的评价模型具有更好的客观性、强壮性、通用性和实用性。并且由于评价结果采用连续函数输出，能够比较精细地分析湖库的富营养化程度。

Eutrophication assessment model using artificial neural networks for lakes and reservoirs

Some faults such as too little training set data, lack of testing (verification) set data, too large network topology and lack of boundary values were found in models, using artificial neural networks (ANN), presented before. A new approach, generating training set data, verification set data and testing set data and boundary set data, was put forward in this paper. Furthermore, the principle of determining the number of hidden layers and their neurons was also discussed. And, the trained NN-based model presented in this paper possessed the capacity of higher generalization and was not affected by the initial values of connection weighs. The assessed results of cases showed that the new established NN-based model was more objective, robust, reliable, practicable, and fault-tolerant compared with other methods such as gray-clustering and gray situation decison-making method etc. It is possible to make analysis and forecast of the eutrophication trend of lake and reservoir using the new model.