Modelling of spatial plasma by using neural network

A prediction model of spatial plasma was constructed by combining a radial basis function network and genetic algorithm. Multiparameterised widths were adopted and their effects on the model prediction were optimised by genetic algorithm. Spatial plasma data were collected by using a Langmuir probe in a Cl-2 inductively coupled plasma. For systematic modelling, plasma discharge process was characterised by a face centred Box Wilson experiment. Compared with statistical regression models, optimised radial basis function network model yielded an improved prediction of more than 45% for electron temperature. Electron density model revealed a noticeable increase in plasma density with increasing Cl-2 flow rate only at higher source powers or lower pressures as well as with decreasing the pressure only at higher Cl-2 flow rate. Also, electron temperature model showed a strong dependence on Cl-2 flow rate. Maintaining a higher Cl-2 flow rate was needed to make pressure (or source power) effect on plasma density (or electron temperature) significant.