P. Koduru, A. Davari, L. Shadle, and L. Lawson (USA)
Circulating Fluidized Bed, Neural Networks, Multi-Layer Perceptron, Backpropagation, Levenberg-Marquardt Algorithm, Modeling
Fluidized bed technology brings solid particles into contact with a gas phase in very controlled conditions. Compared with conventional fluidized beds, Circulating Fluidized Beds (CFB) have many advantages including better interfacial contacting and reduced back mixing. Understanding the CFB technology is not a quick process. The recycle nature of CFB allows for a better process, but also makes modeling and understanding it many times more difficult. Currently there is no way to construct a reliable model of such a complex system using traditional methods. Neural Networks (NN) provides an easier way to construct such a dynamical model. This stems form their proven ability to approximate arbitrary nonlinear mappings. The main objective is to train a NN model to simulate the CFB operation. In this paper we present a model of CFB that predict not only the mass flow rate but also the differential pressures in the plant. In other words we present a Multiple Input Multiple Output (MIMO) model of the CFB. Simulation results are presented and discussed.
Important Links:
Go Back
