论文部分内容阅读
文章提出了一种新的用遗传算法优化设计前向神经网络结构和权重矢量的方法。这种算法采用二进制与浮点数混合编码 ,对神经网络结构用二进制编码 ,对神经网络权重矢量用浮点数编码 ,并设计了与混合编码相对应的交叉、变异、选择算子 ,既保留二进制编码简单、易操作的优点 ,又具有浮点数编码搜索空间大、精度高、稳定性好、运算速度快的优点。优化算法包括两级级联的遗传算法。第一级遗传算法实现快速的局部寻优 ,而第二级遗传算法提高全局寻优能力。
In this paper, a new method to optimize the structure of forward neural network and weight vector using genetic algorithm is proposed. This algorithm uses mixed coding of binary and floating-point, binary coding of neural network structure, floating-point coding of neural network weight vector and crossover, mutation and selection operator corresponding to mixed coding, which not only preserves binary coding Simple, easy to operate advantages, but also has a floating-point code search space, high precision, good stability, the advantages of fast computing speed. The optimization algorithm includes two cascaded genetic algorithms. The first level of genetic algorithm to achieve rapid local optimization, and second-level genetic algorithm to improve global optimization ability.