Application of RBF neural network based on adaptive hierarchical genetic algorithm in soft sensor modeling

A soft model based on improved RBF neural network (RBFNN) is built in this paper. In order to optimize the RBFNN, an adaptive hierarchical genetic algorithm (AHGA) codes the topology and the parameters together and regards them as one genome to be adjusted dynamically by genetic operations. By searching the excellent genome, the best RBFNN is built. AHGA is more scientific than other methods of setting up the topology based on experiences. The simulation results show that the accuracy and the overall converging speed are really improved. This model, which has good real-time property, good stability and high precision, can be applied to on-line measure the carbon content of molten iron.