【摘要】 应用水动力学模型可以在典型情形条件下模拟和预测整个研究区域中潮流运动过程的各个细节,但是对于某些环境与工程问题,人们往往更加关注整个研究区域中某些局部敏感区域在任意条件下的潮流实时变化或快速反馈信息。为了探讨这类问题的解决途径,本文将典型情形下的水动力学模拟方法与前馈型神经网络方法结合,以典型条件下水动力学模拟的结果作为神经网络训练的基础,对深圳湾的潮流运动特性进行了研究。通过论证将神经网络用于二维潮流运动特征模拟的可行性,说明了人工神经网络作为一种非线性动力学系统,能够根据海湾开边界潮流特性有效地模拟研究区域中任一网格对应的水流特性,较水动力学模型更为方便迅捷。利用人工神经网络得出的模拟结果与经过实测资料验证的海湾二维潮流模型计算结果十分吻合。文中对遗传神经网络和其它两种经典的前馈网络(BP网络和RBF网络)的学习能力和模拟效率进行了比较。就三种前馈型神经网络从水动力学数值模拟结果中归纳和提取水动力学信息和数据的能力而言,基于遗传算法的神经网络综合性能最为优越。更多还原

【Abstract】 Hydrodynamic models have been widely used for simulation of tidal characteristics. For some engineering problems, people often have more interests in the realtime or rapid feedback information of hydrodynamic characteristics in specific or local areas rather than those in the whole computational domain. This paper presents the approach to solve such problems through the combination of a 2D hydrodynamic model for tidal flows and feedforward artificial neural networks. Different types of ANNs are investigated through their applications in Deep Bay, Shenzhen, in order to assess their ability and relative performance in encapsulating the generic hydraulic knowledge and sitespecific data in the modeled area. Among the three types of feedforward ANNs used in the study, the ANN based on Genetic Algorithms proved to have the best overall performance.更多还原