漂石河床扩大基础桥墩局部冲刷深度的人工神经网络解

以西南大漂石河床成桥后测试的桥墩局部冲刷深度结果为基础,以其中26个实测数据为样本,用BP人工神经网络对大漂石河床桥墩局部冲刷问题进行拟合.测试结果表明,用拓扑结构为3-30-1的BP网络,经学习40 000次后,随机测试样本局部冲刷深度其计算结果和测试结果的相对误差不超过2%;时于急流测深带来的不可避免的差错,采用先对所有样本同时作为学习样本和测试样本进行测试,再根据水文学原理剔除明显错误样本的方法,同时利用BP网络的容错功能,以确保结果的准确性.

The artificial neural network method used to calculate the depth of local scour of the expending base piers in the boulder riverbed

Based on the test results of the local scour depth near the piers in the boulder riverbed rivers in the southwest of China obtained after the constructions, and 26 test data as the samples, This paper simulates the local scour of the piers in the boulder riverbed rivers by the BP artificial neural network. By adopting the BP network with the structure 3-30-1 and after 40 000 times study, the test results of the network illustrates the error of the random samples does not exceed 2%. Considering the inevitable errors happened in testing the depth of the river in the rapids, the paper uses all samples as study samples and tests samples to test, then eliminates the conspicuous mistake samples with the principle of hydrology, and utilizes the ability of the BP network to tolerant mistakes to ensure the correctness of the results.