Identification Of Moving Loads On The Bridge Based On Neural Network

With the rapid development of transportation of China,the bridge is more and more important,the health monitoring of bridge structures in today's engineering research is a hot topic.As the foundation of health monitoring,moving loads identification is an important basis for the policy-maker carrying on the structure security evaluation and the transportation planning,which is a fundamental premise to ensure bridge structure safety is reliable,and has important theoretical significance and the application value.In this paper,we used the BP neural network theory to the identification of moving loads of the bridge.Firstly,we researched that the BP neural network method was used to the moving loads identification of the basic principles and the implementation steps.Using the orthogonal experiment method establish network's sample storehouse,the sample storehouse's information would be the most and the sample number which was as small as possible.When selected the training samples in the training,we used the K-means clustering and then Euclidean distance method,then the training samples distributed special evenly and they were very representative,the network generalization ability would be the best.Secondly,we introduced the optimization neural network's intelligent algorithms,including the Bee-Queen Genetic Algorithm,the Ant Colony Algorithm and the Particle Swarm Optimization,and brought forward a new Hybrid Particle Swarm Optimization in others work's foundation.This hybrid algorithm mixed the Bee-Queen Genetic Algorithm and the Ant Colony Algorithm's thought to the Particle Swarm Optimization.Test functions simulation results demonstrate that the hybrid algorithm sharpened the Particle Swarm Optimization's optimization ability. Finally,we analyzed the BP neural network in the moving load identification the application results.For orthotropic plate model,we compared the identification results in different optimization methods and response types.We saw that the hybrid algorithm was better and the network in response to acceleration input parameters of the identification results to be better.For those samples who were "polluted" by noise,we used "denoising" method and also obtained the very good identification results.Bridge is in the approximate model for the identification,we discussed the steel pipe concrete bridge and the suspension bridge respectively.From the identification results can be seen,BP neural network in moving load identification had practical significance and great potential.This paper proposed the research methods and the simulation results,have provided with certain reference for the bridge intelligent identification of moving loads,also have built up certain foundation for the health intelligent monitoring identification system's research.