【摘要】 由于BP神经网络存在收敛速度慢和易于陷入极小值等缺点,引入遗传算法(GA)对网络的权值和阈值加以优化,并采用不同组合的输入因子和样本数,对福建省罗源湾口的波浪进行模拟研究.对输入因子的分析结果表明,研究区域的波浪主要受台湾海峡波浪传播影响,与局地气象因子(风速、风向、海气温差)的相关性较弱.训练样本试验表明,30 d以上的波浪历史数据可使GA-BP神经网络充分学习研究区域的波浪特征,从而实现对波浪要素的高精度模拟.模拟结果显示,对春、夏季实测波浪数据的模拟效果均很好,其中相关性分别为0.967和0.938,均方根误差分别为0.112 m和0.107 m,表明GA-BP神经网络在近岸波浪模拟预报中有较广阔的应用前景.更多还原

【Abstract】 Back Propagation algorithm has the main problems of easily converging to local minimum point and slow convergence speed.To overcome the shortcomings,this paper introduced genetic algorithm to optimize the weights and the threshold value of the BP neural network globally,and discussed the selection of the optimum topology of input parameters and training samples to carry out the simulating study of the significant wave height of Luoyuan Bay in Fujian Province.The result of analyzing input parameters shows that the waves in the domain are mainly influenced by the waves propagating from Taiwan Strait and have weak correlation with meteorological information(wind velocity,wind direction,temperature difference between ocean and atmosphere).Sample training experiments show that 30 + days’historical wave data input training can make GA-BP neural network grasp the wave characteristics of the study area,and realize high precision simulation.The simulation results are in good agreement with true values of wave data in both spring and summer,with correlation coefficient values of 0.967and 0.938 respectively and with RMS error of 0.112m and 0.107m respectively,demonstrating a broad prospect of the GA-BP model in coastal wave simulation prediction.更多还原