Volterra神经网络水文模型及应用研究

本文在对Volterra泛函模型与ANN模型进行一致性研究的基础上,提出了基于Volterra泛函结构的神经网络水文模型(VNNH)。VNNH模型吸取了Volterra模型和ANN模型的优点,克服了它们的不足之处。VNNH模型设计了一种多项式的激活函数,克服了Volterra模型求解高阶核函数的困难。并利用自组织神经网络算法确定VNNH模型隐含层神经元的数目;根据Volterra模型的结构由流域的单位线确定VNNH模型的初始权值。将VNNH模型应用于两个实际流域的洪水模拟和预测中,取得了较为满意的结果。

Hydrologic model of Volterra neural network and its application

This paper discusses the relations between Volterra model and ANN model,and proposes hydrologic model of Volterra neural network(VNNH) based on studying in consistency between Voltera model and ANN model.VNNH model has been designed with a kind of polynomial activation function.It is useful and effective for high-order nonlinear systems.The number of hidden units of VNNH model is estimated by adaptive neural network algorithm,and the numbers of hidden units equal to the rank of Volterra model.The initial weight values of VNNH equal to unit hydrograph of watershed by the structure of Volterra model.Numerical simulations in Qingjiang watershed and Ouyanghai watershed have been made and used to test the VNNH model.Comparisons of VNNH model with Volterra model show that the proposed VNNH model is much more effective for hydrological nonlinear system.