基于BP神经网络模型对用电极(铜Cu)一材料(模具钢NAK80)电火花加工系统的一组参数进行了优化。计算结果表明,减少输入层神经元的数量可以加快收敛速度,但计算精度将受到影响。若保持隐层结构,仅由不同训练函数的学习结果改变权系数(映射)实现网络结构优化,则可能较好地预言加工效果。
Based on BP neural network, a group of parameters of electric discharge machining system by using electrode (Cu)-material (die steel NAK80) has been optimized. The calculating results has shown decreasing the amount of neuron in inputting layer can accelerate convergence with some affection on calculating precision. But if keeping the structure of hidden layer, a better prediction of processing effects would be attained by optimizing network structure only through changing weight coefficient (mapping) according to the studying results of different training function.