The Study Of Multilayer Feedforward Neural Network Structure

Intelligence control theory developed rapidly in recent years under the condition of more complex plant and higher control standards. Neural network methods have proven to be powerful tools in modeling of nonlinear process, however, some theoretical problems in conventional methods do not disappear accordingly, such as stability, convergence, and model structure etc. The network structure is significant in characterizing the performance of the feedforward network, such as network capacity, learning speed, generalization ability Standard back propagation performs gradient descent only in the weight space of a network with fixed topology, in general, it is useful only when the network architecture is chosen correctly With a bad structure we potentially run into problems like underfitting, overfitting or wasting computational resources, so algorithms than can find an appropriate network architecture automatically are highly attractive. In this thesis, several matching the network complexity to the real problem methods were compared and their weaknesses and strengths were pointed out In order to improve the CC methods, we proposed CBP methods in this paper. Through analysis, we show that CBP is computationally just as efficient as the CC algorithm even though we need to backpropagate the error through no more than one hidden layer. Further, CBP has the same constructive benefits as CC, but in addition CBP benefits from simpler implementation and the ability to use stochastic optimization routines. Moreover, we show how CBP can be extended to allow addition of multiple new units simultaneously. The simulation results show that CBP is more effective than CC...