Research On The Prediction Of Surface Subsidence By Improving Genetic Bp Network

With the development of economic and population growth, the infrastructure construction is gradually improving and optimizing in our country, particularly in the construction of modern transportation, each year, the state has invested enormous manpower, financial and material resources to ensure safety of people, convenient and comfortable travel. However, in major cities such as Beijing, Shanghai, Guangzhou, Tianjin, Shenzhen, the public transport on the ground is still too heavy to meet the people’s commute requirements, to ease traffic pressure, China’s construction of the subway is in full swing. First of all, the construction and operation of the subway in the underground space disperses the ground mass traffic, greatly spread the traffic pressure on the ground of public; secondly, for the flow of dense population in city, the subway is superior to the ground public transportation in the speed, stability, convenience and transport capacity and other aspects; last but not the least, it is important to rely on electricity to drive, the subway, which not only can save coal and oil and other non-renewable energy, reduce the pollution, but also to meet the the national advocate "low carbon life, green travel", this also is precisely consistent with the Chinese modernization transportation goals. However, in recent years due to the surface damage of geological environment and other causes of settlement did a great harm to safe operation of the subway, therefore, research on the subway surface subsidence is particularly necessary and important.Metro surrounding surface subsidence is a comprehensive problem related to variety of disciplines such as mapping, geotechnical, hydrological, geological and mechanical, and deformation monitoring data are highly susceptible to geological conditions, climate change and other factors, there are still problems such as lacking references and unknown mechanism of action and so on, the use of traditional conventional modeling methods are not efficient and accurate way to predict and analyze ground settlement.In this paper, through the genetic algorithm’s selection operator, crossover and mutation operator to improve the weights and thresholds of BP neural network, optimizing BP network topology and make full use of the BP neural network model,which has a more high fault tolerance, adaptability and shows the strong non-linear mapping ability in dealing with a non-structured, non-accuracy data etc. but at the same time in order to avoid the disadvantages of standard BP neural network with random initialization, slow speed of convergence, trending to fall into local optima easily in training process, the application of genetic algorithms to adaptive threshold parameters for BP neural network model of global optimization, combined with Suzhou Metro Line Riverside Road Station exit 4 of surface subsidence engineering, the improved BP neural network model and the traditional conventional gray Verhulst model and BP neural network model were compared quantitatively analyzed the prediction accuracy of the three models.The results prove that the improved genetic algorithm BP neural network model can not only make better use of raw monitoring data for complex learning and information processing, and has a high fault tolerance and robustness, but also show that the method can comprehensively take a variety of factors into account,which can be applied to the actual deformation monitoring and it is a worth using model.