粒子群算法在提高激光弯曲中神经网络泛化性上的应用

Application of Particle Swarm Algorithm in Improving Generalization of Neural Network for Laser Bending

采用径向基函数(RBF)神经网络建立了激光弯曲成形过程中激光功率、扫描速度、扫描次数与弯曲角度之间的预测模型。对TC4钛合金板材进行了激光弯曲成形试验,将试验数据作为训练样本对神经网络进行了训练,得到工艺参数与成形角度之间的映射关系。利用粒子群优化算法对RBF网络的参数进行寻优计算。结果表明:使用优化后的网络对测试数据预测时,误差由之前的4.714%减小到0.974%左右;使用粒子群优化算法能显著地提高RBF神经网络在预测激光弯曲成形角度时的泛化能力。

The prediction model among laser power, scanning velocity, scanning number and bending angle during laserbending forming process was established by using radial basis function （RBF） neural network. The laser bending formingexperiment of TC4 titanium alloy sheet was carried out. Taking test data as a training sample, the neural network was trained.The mapping relationship between process parameters and forming angle was obtained. The optimizing compute of theparameters of RBF network was carried out by using particle swarm optimization algorithm. The results show that duringpredicting the test data by using the optimized neural network, the error is reduced from previous 4.714% to 0.974%. Usingparticle swarm optimization algorithm can significantly improve the generalization ability of RBF neural network duringpredicting the angle of laser bending forming.