小样本激光切割工艺参数预测模型优化

Optimization of prediction system for small sample laser cutting process parameters

神经网络强大的非线性拟合能力使其在参数预测方面的应用越来越广泛,在神经网络预测激光切割工艺参数过程中,使用小样本数据训练的神经网络模型存在预测误差大、鲁棒性较低等缺点。因此,提出了一种基于生成式对抗网络的小样本数据处理方法,设计了适用于此问题的基于生成式对抗网络的数据生成模型,对激光切割工艺参数的小样本数据进行扩充,运用显著性检验法对扩充的数据进行了可靠性验证。利用BP神经网络构建工艺参数预测模型,分别使用原数据和扩充后数据训练参数预测神经网络,通过实验对|比,优化后的模型具有更好的准确性和泛化性。

Neural network is widely used in parameter prediction because of its strong nonlinear fitting ability,in the process of using neural network to predict laser cutting process parameters,the neural network trained with small sample data has the disadvantages of large prediction error and low robustness.In this paper,a small sample data processing method based on generative countermeasure network was proposed,and a data generation model based on generative countermeasure network was designed.The small sample data of laser cutting process parameters were expanded,and the reliability of the expanded data was verified by significance test.The BP neural network was used to construct the process parameter prediction model,and the original data and the expanded data were used to train the parameter prediction neural network.Through experimental comparison,the expanded training model has better accuracy and genera lization.