PSOTrainBP

  BP神经网络容易陷于局部极小值，PSO算法在无约束非线性函数优化方面性能优越，通常可以直接找寻到全局最优解，即使不能搜多到全局最优解，也距离全局最优点不远。当然，基本PSO算法陷入局部极值也是有的。对于这个缺点目前还没有找到比较有效、省市的解决方案。本案例实现利用PSO算法和BP算法共同训练神经网络，先将网络进行PSO算法训练，然后BP算法接着进行小范围精细搜索，PSO算法训练神经网络的本质就是将输出误差函数（即能量函数）看成目标函数，PSO对能量函数进行全局寻找最小值。(BP neural networks are prone to local minimum values. The PS algorithm has superior performance in the optimization of unconstrained nonlinear functions. It can usually find the global optimal solution directly. Even if it can not find more global optimal solutions, it is not far from the global best. Of course, there is also a local extreme of the basic PO algorithm. For this shortcoming, there is no more effective, provincial and municipal solution. This case realizes the use of the SO algorithm and the BP algorithm to train the neural network. First, the network is trained in the SO algorithm, and then the BP algorithm is followed by a small range of fine searches. The essence of the PO algorithm training neural network is to regard the output error function(ie, the energy function) as the objective function, and the PO seeks a global minimum value for the energy function.)

PSOTrainBP\PSOBP502 (1).m, 7009 , 2018-08-06
PSOTrainBP\PSOTrain.m, 3974 , 2018-08-06
PSOTrainBP, 0 , 2018-08-06