基于思维进化算法的神经网络权值与结构优化

人工神经网络应用的关键在于权值和结构的优化。思维进化计算(MEC)是模拟人类思维进化过程的一种新的进化算法，具有极强的全局寻优能力，在数值优化和非数值优化方面均显示出明显的优越性。本文在思维进化计算框架的基础上，提出了一种用于人工神经网络权值与结构优化的思维进化算法，设计了有效的结构优化‘趋同’与‘异化’算
子；在局部范围内寻求局部最优解，然后使用异化算子跳出局部范围的约束，在整个解空间寻求全局最优解。仿真结果说明了方法的正确性与有效性。

Neural Network Weights and Structure Optimization Based on MEC Algorithms

The key to artificial neural network applications is weights and structure optimization. The Mind Evolutionary Computation(MEC) is a new evolutionary algorithm which simulates the process of human mind evolution, it has the powerful ability to find global optimum, and it also has much superiority in resolving the problem of numerical and non-numerical optimizations. In the paper, a MEC algorithm is presented based on the foundational MEC algorithm framework to optimize neural network structures and weights, in which effective similartaxis and dissimilation operators of structural optimization are designed. Through similartaxis operators, the local optimum is found, and by exceeding the restriction of local range through dissimilation operators, the global optimum is acquired in the global solution space. Simulation results show the effectiveness and correctness of the method.