【摘要】 提出一种用于多元非线性函数逼近的小波神经网络的训练算法,设计了网络拓扑结构和对其参数进行估计的混合递阶遗传算法。该方法避免了经典BP算法容易陷入局部极小和训练效率低的缺点,保持了较高的泛化精度,且收敛速度快,进而提高了测量精度。实验表明:应用小波神经网络多元非线性回归方法的含水率测量值与真实值间相对误差均方差为0.084,相对误差均值为0.117,相对误差最大值为0.159。更多还原

【Abstract】 A regression is put forward based on a wavelet function network algorithm to modify the measurement result.The network’s topological structure and its parameters evaluation method are designed by bonding the gradient GA algorithm and multi-nonlinear regression so as to avoid getting into infinitesimal locally effectively and keep the merits of high prediction precision and rapid convergence so as to enhance the measurement accuracy.The results show that the root mean-square relatively error is 0.084,the mean absolute relatively error is 0.117,and the maximize absolute relatively error is 0.159 between the measurement value and the real one.更多还原