免疫粒子群算法在神经网络训练中的应用

Application of immune particle swarm optimizer in neural network training

神经网络能够用来检测结构损伤,但是其训练方法容易陷入局部最优。粒子群算法具有全局搜索能力,将免疫系统中的抗体抑制机理引入粒子群算法以保持粒子多样性,采用免疫粒子群算法(ImPso)训练前向神经网络。计算机仿真结果显示,训练后的网络性能优于使用一般BP算法训练的网络。

Feedforward neural network can be used to detect structural damage,but the gradient descent method in traditional BP algorithm is vulnerable to local optimum.Particle Swarm Optimizer（PSO） can be capable of searching optimum in global scope,by introducing clone suppression in immune systems into PSO to maintain the diversity of particles.The hybrid algorithm（ImPso） can be used to train feedforward neural network.The computer simulation results show that the performance of neural network with the hybrid algorithm（ImPso） is better than the performance with traditional gradient descent method.