基于DGA的BP神经网络及其在一维河网模拟中的应用

考虑到BP神经网络的计算精度和稳定性依赖于初始权、阈值,首先对标准遗传算法进行改进,然后用改进的遗传算法优化BP神经网络的初始权、阈值。将遗传算法和神经网络结合起来建立河网BP模型,把实测资料或者水动力数学模型的计算结果作为学习样本对模型进行训练。将河网BP模型运用于西江三角洲河网,发现该模型与水动力模型的计算结果吻合较好,表明优化后的BP网络用于河网水力模拟是可行的。

BP Neural Network based on DGA and its application in one-dimensional river network simulation

Back propagation neural networks(BPNN) are non-linear, self-adaptive and high-speed,and therefore can be used in complicated river network simulation.Considering that the precision and stability of BPNN depend on initial weights,some improvements are made on the standard genetic algorithm.Then,with the developed genetic algorithm(DGA),the initial weights of BPNN are optimized.A BP model for a river network is presented by combining genetic algorithm and BPNN.The model is trained with the measured data and a hydrodynamic module.The model was used to simulate the river network of the Xijiang Delta.Sound agreement was obtained between the results of the BP model and those of the hydrodynamic module,which suggests that the BP model is particularly suited for hydraulic simulation of river networks.