Visual interactive meta-simulation using neural networks

This paper proposes the technique of visual interactive meta-simulation. Results from a previously validated visual interactive simulation model are fed to a neural network. After training, the neural network becomes a metamodel of the original system which then allows real time management interaction and visualisation of different decision scenarios. This paper compares different types of simulation metamodel for three problems and reports that metamodels developed using a neural network consistently gave the most accurate responses. Results in the form of 2D and 3D response surfaces are then shown. The paper concludes by considering issues that must be resolved if the technique of visual interactive meta-simulation, using neural networks, is to be used on a more routine basis.