RBF神经网络在洪灾易损性分析中的应用

RBF Neural Network in the Application of Flood Vulnerability Analysis

针对水稻洪灾易损性与多种因素呈复杂的非线性相关问题,通过训练和检验样本的选择、网络程序的构建、神经网络的训练、最佳参数的确定、原始数据的检验等过程,以Matlab为平台建立RBF神经网络洪灾易损性模型,对水稻的洪灾易损性进行模拟,并与AGA-BP网络的训练和检验结果进行对比,以此验证RBF神经网络在洪灾易损性分析中的可行性和有效性。实例分析表明:RBF网络检验样本的平均相对误差是5.428%,比AGA-BP网络(检验样本的平均相对误差是8.652%)精度更高,检验效果更好,对易损性的模拟程度更佳,使洪灾易损性分析结果更加准确、可靠。

Aimed at the flood vulnerability,the vulnerability of rice flooding is simulated,and the AGA-BP neural network with the training and test results are compared to verify RBF neural network in the flooding of vulnerability analysis is feasible and effective.Our analysis shows that the RBF network test sample of average relative error is 5.428%,compared with AGA-BP network（test sample of average relative error is 8.652%） a higher precision,inspection effect is better,the simulation degree of vulnerability more makes flood vulnerability analysis result more accurate and reliable.