Study And Application Of ANN Optimization Model On Dam Deformation Prediction

Dam is a building which is mainly built in complex hydrology、engineering geological conditions and bears huge load. Having a real-time monitoring, using a large amount of deformation observation data, analyzing and accessing its safety state then making a forecast to it are the important means to ensure its safety. Meanwhile, since affected by so many factors, such as water pressure, uplift pressure, temperature, time-dependent and many uncertainties, moreover, with strong randomness and complex mutual relationship of these factors, which make it difficult to describe the exactly quantitative relationship between these factors and Dam displacement when using the traditional mathematical model who is found on the conditions of independent observations and zero expected value. Especially when having less observation data or large observation noise error, the traditional deformation analysis models are usually limited. Therefore, studying the fusion of multi discipline knowledge and technology method and establishing a suitable combination of deformation analysis and prediction model to analyze the deformation trend become an important subject. Started from this point, the paper introduces Neural network theory, Genetic algorithm,Particle swarm optimization algorithm and its improved algorithm, then studies the feasibility of these mixed intelligent models. Combined with concrete dam engineering, this paper has an application and comparison of these mixed intelligent models.The main research contents are as follows:1) To have a study of the Neural network, Genetic algorithm, PSO algorithm, based on the fact that Neural network has a strong randomness when initialization, slow convergence rate, easy to fall into local minimum, the paper uses Genetic algorithm and PSO algorithm to optimize the weights and thresholds between neurons and has a nonlinear improvement of the weight of the standard PSO algorithm.2) Using variable selection method based on ANN to calculate all factors’ contribution rate, according to it, we make a choice of these impact factors and determine the final factors of the dam, then normalize all observation data and these impact factors.3) Building dam deformation prediction model based on GA, PSO and Improved PSO algorithm respectively. BY programing corresponding model in MATLAB environment and applying these models to some typical dam deformation analysis and prediction, comparing and analyzing the traditional mathematical statistics, BP model, GA-BP and IPSO-BP, this result shows,the prediction accuracy of dam deformation models based on IPSO-BP is higher than traditional mathematical statistics model, BP model, and GA-BP model.