Research On Damage Identification Of Structure Based On Ga-bp Neural Network

At present, the constructions of power engineering are required to be able to usemore than60years. In the long process of use of the power engineering buildings,effective real-time monitoring of the building, ensuring its durability and safety, has animportant significance to the whole life cycle management of electric power project. Oneof the core technologys of health monitoring system is the damage identificationtechnology. the further in-depth study on the damage identification method, which can bewell fit for the real-time structural health monitoring system of electric power civilengineering structure, has an important theoretical significance and engineeringapplication value.The problem of structural damage identification essentially can be regarded as aproblem of pattern recognition, which is a problem to find the contact of the damagecharacteristic indexes and damage state. The neural network has obvious advantages inthe field of pattern recognition and has been used in the field of damage identification bymore and more scholars and researchers. However, BP algorithm is a kind of methoddepending on the gradient descent, and it is hard for it to avoid the problem of slowconvergence and falling into local minimum value. It is unable to meet the needs ofpractical application.This paper studies the basic principle of BP algorithm and genetic algorithm,analyzes the necessity of the fusion of the genetic algorithm and BP neural networkalgorithm, the basic thought and the method to determine the basic elements in the fusion,then combines the genetic algorithm which is good at overall search with BP algorithmwhich has much strong local optimizing ability. According to the characteristics of thecross operator, variation operator and choice operator of GA that these operators searchthe overall solve with great probability in the whole variable space and converge fast andaccurate near the solve. We use the GA to optimize the weights of the neural network,and then we use the parameters adjusted as the initial weights of the BP network.Through this method, the problem of easily trapped in local minima can be improved. Atthe same time, the speed of convergence is improved.In this paper, the top five order natural frequency is used as the damageearly-warning index of the frame structure. Combining with an instance of a framestructure of a certain power plant, the damage location and quantitative indicators areselected. With the optimized GA-BP algorithm, the power engineering frame structuredamage identification system is established. The experimental results show that thedamage identification effect of the frame structure damage identification system based onGA-BP algorithm is good, and it can satisfy the precision requirement in the practical engineering.