A Bayesian–Gaussian neural network and its applications in process engineering

Recently, artificial neural networks have been widely applied in process engineering, where the back-propagation neural networks are most frequently used, while the recurrent neural networks and radial basis function neural networks are sometimes used. However, the intrinsic vulnerable points of these networks in long training time, local minima and lack of self-tuning ability impair their further, specifically on-line, applications. To this end, a Bayesian–Gaussian neural network is introduced in this paper. Simulation studies on its application to the dynamic behaviour prediction of a nonlinear single-input single-output system, as well as to the static performance and dynamic behaviour predictions of circulating fluidized bed boilers, are provided to assess the advantages of this network, the results of which indicate that the BGNN could be a good alternative in neural network model based applications in process engineering.