基于机器学习的二次电子发射唯象模型

基于机器学习和深度人工神经网络（artificial neural network，ANN）提出一种二次电子发射唯象模型。利用Vaughan模型生成先验数据集，用于训练生成描述二次电子发射一般规律的先验知识ANN模型，并在不同参数条件下验证了先验知识ANN模型的正确性。然后，分别利用银和铝合金材料的二次电子发射系数实验数据修正先验知识ANN模型，分别得到了描述两种材料的特异ANN模型。测试结果表明，特异ANN模型计算结果与实验结果相比的平均绝对误差较Vaughan模型和Furman模型降低了30%以上，与复合唯象模型精度相当或更高。在小样本条件下测试了二次电子发射ANN模型的正确性，验证了分步训练方式的有效性和二次电子发射ANN模型对于小样本集的适应性。提出的基于机器学习的二次电子发射唯象模型能够避免复杂的参数修正过程，能够基于先验知识提升模型对于小样本的适应性，能够实现二次电子发射系数的连续插值，适于在数值模拟软件中使用。

Phenomenological model for secondary electron emission based on machine learning

A new phenomenological model for secondary electron yield (SEY) was proposed based on machine learning and deep artificial neural network (ANN). The Vaughan model is used to generate a training data set, which is used to train an artificial neural network to generate a prior knowledge ANN model.The prior knowledge ANN model can be used to describe the general law of secondary electron emission. The correctness of the prior knowledge ANN model is verified under different conditions. Then, two specific ANN models for silver and aluminum alloy are obtained by update the prior knowledge ANN model using the experimental SEY data. The modified Vaughan model, the modified Furman model, the composite phenomenological model, and the specific ANN model are used to describe the SEY of silver and aluminum alloy, respectively, and compared with the experimental results. The comparison results show that the mean absolute error of the specific ANN modelis more than 30% lower than that of the Vaughan model and Furman model, and is comparable or better than composite phenomenological model. The effectiveness of the step-by-step training method and the adaptability of the specific ANN model for small samples isverified under different conditions.The phenomenological model for SEY based on machine learning proposed in this paper can avoid complex tuning process, realize the continuous interpolation of SEY based on the prior knowledge and is more suitable for numerical simulation.