Comparison of algorithms for predicting plasma physical parameters of SF6–Cu mixtures at local thermodynamic equilibrium state via machine learning

In order to study the arc process of the SF6 circuit breaker during the current breaking process, it is necessary to calculate the physical parameters of the arc discharge plasma. However, the calculation of plasma physical parameters is very difficult and complicated and generally requires solving dozens of differential equations. Based on the machine learning method, this paper constructs a learning prediction model of physical property parameters in a local thermodynamic equilibrium state without solving a large number of differential equations so as to perform a rapid prediction of physical property parameters in other scenarios based on the existing physical parameter database. This paper uses the support vector machine, K-nearest neighbor algorithm, gradient boosting regression, decision tree, and random forest algorithm to predict and calculate the thermodynamic parameters and transport characteristics of SF6 at different pressures and temperatures. At the same time, this paper also predicts and calculates the parameters of the SF6–Cu mixed gas at different mixed concentrations. The results show that the machine learning algorithm can predict and generate consistent gas property parameter data, indicating that the model has good generalization performance. Finally, by comparing the error measures of the prediction results of various algorithms, the algorithm suitable for predicting the physical parameters is found to improve the prediction accuracy.