Development of neural networks in civil engineering applications.
Date of Issue2010
School of Civil and Environmental Engineering
The architecture of neural networks (NNs) has a significant impact on a network’s generalization ability. Ensemble neural networks (ENNs) are commonly used networks in many engineering applications due to their better generalization properties. An ENN usually includes several back-propagation networks in its structure, where the back-propagation network is a single feed-forward network trained with the back-propagation learning rule. In this thesis, the Akaike information criterion (AIC) and the entropy are used as the automating design tools for balancing the generalization against the parameters and finding the best combining weights of the ENNs. Two ENNs, namely, the AIC based ENN and the entropy based ENN are developed first. Since the AIC and entropy have their own merits for solving different problems, a new AIC-entropy based ENN is developed. Two analytical functions – the peak function and Friedman function are used first to assess the accuracy of the proposed ensemble approaches. The verified approaches are then applied to the civil engineering applications.