To establish an accurate arc model, this paper researches the simplified Schavemaker model under the background of industrial, commercial and residential electrical systems. A calculation method based on neural network is proposed, which uses a calculation method based on experimental data to obtain training samples and build a neural network between model parameters and working conditions. The built neural network can directly predict the model parameters’ values under certain working condition when the experimental data of this working condition cannot be obtained. The final verification results show that the proposed model parameter calculation method based on neural network has high accuracy and superiority compared with the existing methods. It also shows that the established AC arc model can realize arc simulation and can be used for further AC arc feature learning and detection algorithm design. This work is supported by National Natural Science Foundation of China (No.51277093).