Neural network approach to energy confinement scaling in Tokamaks

Abstract-Empirical studies ofthe scaling ofTokamak energy confinement times with machine parameters constitute a useful point ofcontact with physics-bssed transport theories. They also form the basis for the design of next-step and reactor grade Tokamaks. In most cases a simple power law (or sometimes offset linea) functional farm is fitted to the data. Such linear regression techniques have the advantage of computational simplicity, but otherwise have little rr-priurijustification. Neural networks provide a powerful general purpose technique for nonlinear regression which exhibits no essential limi?ations on the functional form which can be fitted. In this paper we apply neural networks to the problem of energy confinement scaling and we illustrate the technique using data from the JET (Joint European Torus) Tokamak. The results show that the neural network approach leads to a substantial improvement in the ability to predict the energy confincment times as compared with linear regression. The significance of this result is discussed. I.