Abstract
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 Cl2 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 Cl2 flow rate only at higher source powers or lower pressures as well as with decreasing the pressure only at higher Cl2 flow rate. Also, electron temperature model showed a strong dependence on Cl2 flow rate. Maintaining a higher Cl2 flow rate was needed to make pressure (or source power) effect on plasma density (or electron temperature) significant.