Application Of Artificial Neural Network In Construction Control Of Composite Cable-Stayed Bridges

The main bridge of GuanYinyan Yangtze River Bridge in JiangJin is a composite cable-stayed bridge with steel I-beam and concrete deck plate, double pylons and cable planes with main span of 436m.In this thesis, taking the construction control of GuanYinYan Yangtze River as the project background, the method of application of Artificial Neural Network(ANN) for construction control of composite cable-stayed bridge is studied.Nowadays, the BP(back propagation) Neural Network has been one of the most widely applicated networks. But, attributing to the drawbacks of its algorithm principle and theory, the BP Neural Network has some faults, such as the possibility of falling into local minima, slow convergence and weak generalization ability. In order to solve these problems, the feasibility of the application of both Radial Basis Function(RBF) Neural Network and the BP Neural Network methods in the construction control of composite cable-stayed bridge is explored.Firstly, the algorithms and principles of BP network and RBF network are gone into details. Comparing the one to the other what has been discussed above, we may draw the conclusion that in theory, the RBF network, using local approximation approach, is superior to the BP network which uses universal approximation approach. Then, the methods and details of using BP network and RBF network for parameter identification are illustrated. Establishing the parameter identification model under the same parameters'value of networks, comparing the convergence time and convergence precision of both networks as stated above, we prove the feature that the RBF neural network is superior to the BP neural network in practice.Based on the indicating that the RBF neural network is superior to the BP neural network in practice and theory, the specific method and feasibility of the application of the RBF network method in the prediction of linear errors and in determining the cutting length of the mid-span closure segment are explored.Finally, using Artificial Neural Network to solve the three problems above, the feasibility of the application of the Artificial Neural Network method in the construction control of composite cable-stayed bridge is explored and summerized.