| In the modern building construction, the pile foundation obtains the widespread using as a result of its own merits. It is inseparable with modern project technology and the economic development of our country. But in the construction work, the body of pile is easy to appear some damages, such as breaking, reducing, segregation and so on. For that reasons, pile foundation itself can not be able to achieve the anticipated supporting capacity or the structure having the differential settlement to cause the project accident the occurrence. So, how to determine the healthy of pile foundation reasonably has become a very important topic.Analysis and forecast of gathering data using the neural network is an effective method to this problem. This thesis was precisely in the basis of analyzing the characteristic of the neural network processing data , in view of existence question of RBF network designing and training ,fused many kinds of intelligent methods, designed one kind of improvement of DNA evolution computation, and optimized the RBF neural network using the improvement of DNA evolution computation to solve the above problems, to improve network architecture, to enhance the network's approach ability and exude ability, to carry on the health examination of pile foundation effectively.The major contributions are as follows:(1)The thesis has studied the neural network, the rough sets theory, the genetic algorithm, and the DNA computation and so on many kinds of intelligent methods, and has thoroughly analyzed each kind of intelligent method's good and bad points.(2)The thesis has studied pretreatment neural network training sample sets based on the rough sets. For rough sets has the characteristic of simplifying the information space expression dimension, set rough sets achievement neural network pretage system to simplify the information structure of training sample sets, and simplify the complexity of neural network.(3) The essence of optimizing the RBF neural network is a multi-objective question. This paper designed one kind of multi-objective DNA evolution computation to solve the RBF neural network's optimized problem. In this algorithm, determine individual sufficiency in the evolutional computation using sorting Pareto control sets, do hereditable operation using evolution operator such as overlapping, variation, inversion, duplication, find the Pareto optimal solution sets fast, and select the most superior RBF network. At the same time this article has given the algorithm the astringency proof.(4) Apply RBF Neural network based on DNA evolution computation for the pile foundation health examination. Through analyzing and comparing the measured data and the data forecasted by neural network, showed this network will have the good examination effect.The research's results showed that in theoretical analysis, rough sets pretreat neural network's training sample sets, simplify the complexity of neural network. Introducing Pareto optimal solution sets to the DNA evolution computation, solve the multi-objective optimization problem effectively, and realize the RBF neural network's optimization. From the simulation and actual experimental results, compare with the traditional RBF network, the approached ability and exudes ability of RBF neural network based on DNA evolution computation have enhanced remarkably, and the forecast effect has been more precise. |
