Parameter Identification Based On Improved Neural Networks In Cable-Stayed Bridge

During the construction of the cable-stayed bridge, the actural value of the structural parameters are always differed from the designed one. Because of the existence of this errors, the configuration and internal force of the constructed bridge will be biased with the designed state. However, with the restriction of the existing measuring method, some structural parameter can not be obtained accurately. In consequence, backgrounded by the construction control of Jiashao Bridge, this paper applies the improved neural networks in parameter identification of cable-stayed bridge, the main works and some conclusions are as follows:(1) The specialties of artificial neural networks, grey system and wavelet transforms are respectively analyzed and the possible improvement forms of traditional neural networks are discussed. Then, this thesis focuses on respectively presenting the structures and algorithms of grey neural networks and wavelet neural networks in detail. At last, the superiorities of grey neural networks and wavelet neural networks are concluded by compared with the traditional neural networks.(2) Taking Jiashao bridge for background, the swatches for neural networks training and testing are obtained by theoretical calculation of FCM software—NLABS. Thus, the training swatches are trained respectively by grey neural networks and wavelet neural networks based on which the testing swatches are simulated in order to verify the feasibility and reliability of improved neural networks in parameter identification of cable-stayed bridge. The errors calculated by improved neural networks are separately compared with the one gained by the traditional BP neural networks. It can be concluded that the measure of improved neural networks have better reliability with good robusticity and satisfactory accuracy for parameter identification in the construction control of cable-stayed bridge.(3) The methods of initial parameters optimization in improved neural networks are discussed, then, this paper attempts to optimize the initial parameters of grey neural networks by the help of the theories of genetic algorithm and particle swarm optimization respectively. By the application of optimized grey neural networks for parameter identification in cable-stayed bridge, it is concluded that optimization of the initial parameters of neural networks could improves the accuracy of the net and the velocity of convergence to some degrees as well as solving the network shock caused by the unreasonable parameter initialization and decreasing the possibility of falling into partial optimal answer.