基于遗传神经网络的AZ91镁合金力学性能预测

Prediction of mechanical properties of AZ91 magnesium alloys based on genetic neural network

在合适的参数条件下径向基函数神经网络能够以任意的精度来逼近任意的函数,遗传算法是一种高效的全局寻优的搜索方法.将遗传算法和径向基函数神经网络相结合,建立遗传神经网络,运用到合金设计的性能预测方面.使用简单易行的二进制编码方法,在寻求径向基函数神经网络隐含层神经元最优中心矢量的同时确定其最优个数,通过设定合理的目标函数解决网络函数逼近能力与泛化能力之间的矛盾.试验证明,该方法在合金性能预测方面有较好的效果,能够成为合金设计有力的辅助手段.

It has been shown that the radial basis function neural network with appropriate parameters is a universal approximator for continuous functions. The genetic algorithm is efficient in seeking for a global solution. A net was built with genetic algorithm and RBF neural network combined together, which was used to predict properties of alloys. The optimum number of the neuron was confirmed when these neurons had been found by using a simple bit string coding method. A reasonable fitness function was designed to settle the contradiction between approximative capability of the neural network and using capability. According to the result of the prediction of the mechanical properties of AZ91, this method can be considered as a potent assistant instrument to those non-linear systems.