This paper presents a new cellular automata (CA) model which uses artificial neural networks for both calibration and simulation. A critical issue for urban CA simulation is how to determine parameter values and define model structures. The simulation of real cities involves the use of many variables and parameters. The calibration of CA models is very difficult when there is a large set of parameters. In the proposed model, most of the parameter values for CA simulation are automatically determined by the training of artificial neural networks. The parameter values from the training are then imported into the CA model which is also based on the algorithm of neural networks. With the use of neural networks, users do not need to provide detailed transition rules which are difficult to define. The study shows that the model has better accuracy than traditional CA models in the simulation of nonlinear complex urban systems.
A dynamically reconfigurable logic cell: from artificial neural networks to quantum-dot cellular automata