Research On Adaptive Structure Model Of Multi-layer Feedforward Small-world Neural Networks

Multi-layer forward neural network is a kind of classical network of artificial neuralnetwork, is currently the most successful and most widely used network model. But WSmulti-layer neural network and NW multilayer neural networks to the small-worldnetwork during the build process is the use of the randomization across-layer and edge tocreate a short-cut, so small world system generated neural network lost importantinformation, Presented on the basis of merit thoughts multilayer forward small worldstructure adaptive neural network model.First study based on merit type WS multilayer forward small world structureadaptive neural network model, WS multilayer forward small-world structure adaptiveartificial neural network, the two networks and the original multilayer neural networks tothe small world before do the comparison of performance is convenient, the study foundthat under the same precision, p=0.05network nonlinear function approximation effectis best. After the merit of NW multi-layer forward small world structure adaptive neuralnetwork model with the original NW multilayer forward in small-world network underthe condition of increasing precision and the convergence rate of the two kinds ofnetwork also showed a trend of close to, but after the merit is obviously better than theoriginal network. Experiment has proved that, after the merit of multilayer forward smallworld structure adaptive neural network model in the network performance is better thanthe original small multi-layer forward neural network in the world.Finally model of NW multilayer forward small world structure adaptive neuralnetwork to predict billet surface temperature, given the billet reheating furnace heattransfer process is very complex, which involves many factors, so the mechanismmodeling method to set up the prediction model of slab surface temperature is extremelydifficult. Results show that NW small-world structure of neural network training speedand precision of adaptive model is superior to multi-layer forward neural network withthe size of the rules, can meet the actual demand forecast.