Foundation Settlement Prediction Method Research Based On Wavelet Neural Network

As the continuous development of economic and urban, high buildings rise rapidly in recent years. And also the accidents caused by being neglected the foundation settlement monitoring occur frequently. Therefore the safety of ground subsidence monitoring has become a hot industry, and the foundation security settlement has been pay attention. In recent years, in order to explore an effective way to predict ground settlement, many scholars have been carrying out a lot of explorations and researches from aspects of theory and practice, such as the gray theory and time series, etc. These methods have achieved some success, but still there are many problems and defects. However this paper, based on the characteristics of foundation settlement and the current hot methods in the predicting area, used the wavelet analysis, neural networks, intelligent optimization algorithms and other methods to predict the foundation settlement. And also for the shortcomings of some intelligent optimization algorithms which are researched widely currently, we carried out a wide range improvement and comparison. From the comparative experimental results which got by the improved methods and traditional methods, we can see that these improved methods have take a good results.In this paper, we took the foundation settlement data which got under the karsts geology condition as the research object. For the problem of foundation settlement observational data easy being mixed noise, firstly we used the wavelet to remove the noise from the settlement observational data, and then we divided the processed data into two categories. One we used it to treat neural network, and the other we to test the neural network's learning effect. To improve the neural network's learning speed and learning effect, we integrated many optimization algorithms, such as particle swarm optimization algorithm, simulated annealing algorithm, cloning immune optimization algorithm, chaos optimization algorithms and so on, to optimize the neural network parameters to improve the network's learning effect. In this paper, we also explored many kinds of improved methods for the particle optimization's shortcomings. These improved methods can be divided into four types. The first way which we improved the algorithm with is that we integrate with many kinds of optimization algorithms to optimization the same problem. A single algorithm is easy to expose its defects. This way many kinds of methods combined with each other help to make the algorithm learn respective advantages from each other. It is for this reason that we have considered using this improved method. The second way we improved is multi-species multi-strategy approach. That is to see in the algorithm the swarm was divided into multiple subgroup, and for the different swarm we took different update strategy. The third improved method base on the swarm's diversity. Inspired by the diversity control thought, we improved the particle update formula's inertia weight. The value of the inertia weight is determined by a nonlinear decreasing function and a random disturbance term. This disturbance term can be adjusted dynamically based on the current swarm's diversity. When the swarm's diversity is poor, the random disturbance term's value will be large, and contrary when the swarm's diversity is fine, the random disturbance term's value will be small. Therefore this will make the inertia weight have a large value when the swarm's diversity is poor and have a small value when the swarm's diversity is fine. A large inertia weight can make the particle search in the global area, thus the algorithm can have a large opportunity to jump out of the local optimal value. A small inertia weight is helpful to make the particle search in local area, thus at this time it can contribute to improve the algorithm's convergence speed and accuracy. To further improve the algorithm's convergence accuracy, at latter part of the algorithm, we also introduced the mutative scale chaos search algorithm which has strong global search ability. The fourth way is multi-model improved method. In this improved method we divided the particle into different model according to its second-order original moment. Under the different model we take different update formula. To improve the search accuracy we integrate the method with mutative scale chaos search algorithm. From a large number of the compared experiments which got form these improved methods in the above all, we can see that these improved methods all have achieved very good results.With the continuous development of bridge construction, functions and formats of bridge become more complicated day by day. The technology of repairing and strengthening of bridge has been emphasized. After bridges having been constructed and opened to traffic, their material will be deteriorated or aged gradually because the influence of the weather, environmental factors and their strength and stiffness will degrade with the time running for action of the static and active loads applying on them. Not only will this endanger the safety of the traffic, but also it will shorten the life span of the bridge and claimed the lives of people. Therefore, detection and inspection of bridges have become one important guarantee that ensures safety maintenance and normal use of it. So, how to perform quality detection and safety inspection with bridges has become a research hotshot of foreign academia and engineering.