改进遗传BP网络的地表沉降预测方法研究

发布时间：2018-04-14 06:11

本文选题：沉降监测 + BP神经网络　；参考：《江西理工大学》2015年硕士论文

【摘要】：随着经济的发展与人口的增长,我国的各项基础设施建设正在渐渐地完善和优化,尤其是在现代化的交通建设方面,国家每年都投入巨大的人力、财力和物力,以保障人们安全、便捷、舒适的出行。然而,在北京、上海、广州、天津、深圳等各大城市,地面上的公共交通依然较难满足市民们的通勤要求,为缓解交通压力,我国地铁的修建正紧锣密鼓地进行。首先,地铁的建设与运行都在地下空间,将大量的地面交通量分散到地下,极大地缓解了地面的公共交通压力;其次,在人口流动密集的城市,地铁在速度、稳定性、便捷性和运输力等方面都要优于地面的公共交通;最后,重要的是,地铁依靠电力来驱动,这不但可以节约煤炭和石油等不可再生能源、减少环境污染,而且还符合国家倡导的“低碳生活,绿色出行”的理念,这也正好符合我国建设现代化交通的目标。但是,近年来由于地质环境的破坏和其他原因引起的地表沉降都对地铁的安全运行产生了巨大的危害,因此,地铁地表沉降问题的研究尤为必要和重要。地铁周边的地表沉降是一个涉及到测绘、岩土、水文、地质和力学等各种学科交错的综合性问题,而变形监测的数据极易受到地质条件,气候变化等因素的影响,还存在着参考资料不足、作用机理不明等问题,采用传统常规的建模方法没有办法高效准确地进行预测和分析。本文采用遗传算法中的选择算子、交叉算子和变异算子来对BP神经网络进行权值和阈值的改进,并将BP网络的拓扑结构进行调整优化,充分利用了BP神经网络模型具有较高容错性、自适应性和处理具有非结构性、非精确性规律的数据时表现出的超强的非线性映射能力等优点,同时针对标准BP神经网络初始化具有随机性、训练过程收敛速度慢、结果易陷入局部最优等缺点,应用自适应的遗传算法来对BP神经网络模型的阀值参数进行全局优化,并结合苏州地铁一号线的滨河路站4号出口的地表沉降工程,将改进的BP神经网络模型与传统常规的灰色Verhulst模型以及BP神经网络模型进行了对比,定量地分析了三种模型的预测精度。结果证明,采用遗传算法改进的BP神经网络模型不仅能够较好地利用原始监测数据进行复杂的学习和信息处理,并且具有较高的容错性和鲁棒性,同时也表明了该方法能够综合考虑多种因素的影响,可以将其应用到实际的变形监测中,是值得采用的一种模型。
[Abstract]:With the development of the economy and the growth of the population, the infrastructure construction of our country is gradually improving and optimizing, especially in the modern transportation construction, the country invests huge human, financial and material resources every year.In order to ensure the safety of people, convenient, comfortable travel.However, in Beijing, Shanghai, Guangzhou, Tianjin, Shenzhen and other major cities, the public transport on the ground is still difficult to meet the commuting requirements of citizens.First, the construction and operation of the subway are in underground space, dispersing a large amount of surface traffic to the ground, greatly relieving the pressure of public transport on the ground; secondly, in cities with dense population mobility, the subway is at speed and stability.In terms of convenience and transport power, it is better than public transport on the ground. Finally, it is important that the subway is driven by electricity, which not only saves non-renewable energy sources such as coal and oil, but also reduces environmental pollution.It is also in line with the concept of "low carbon life, green travel" advocated by the state, which coincides with the goal of building modern transportation in our country.However, in recent years, the ground subsidence caused by the destruction of geological environment and other reasons has caused great harm to the safe operation of subway. Therefore, the study of subway surface subsidence is particularly necessary and important.The ground subsidence around the subway is a comprehensive problem involving surveying, mapping, geotechnical, hydrology, geology and mechanics, and the deformation monitoring data are easily affected by geological conditions, climate change and other factors.There are still some problems such as lack of reference data and unclear mechanism of action. There is no way to predict and analyze efficiently and accurately by using conventional modeling methods.In this paper, the selection operator, crossover operator and mutation operator in genetic algorithm are used to improve the weight and threshold of BP neural network, and the topological structure of BP neural network is adjusted and optimized.The BP neural network model has the advantages of high fault-tolerance, self-adaptability and the ability of super-strong nonlinear mapping when dealing with data with imprecise laws and so on.Aiming at the randomness of the initialization of the standard BP neural network, the slow convergence speed of the training process and the easy to fall into the local optimum, the adaptive genetic algorithm is used to optimize the threshold parameters of the BP neural network model globally.The improved BP neural network model is compared with the traditional grey Verhulst model and the BP neural network model, combined with the surface settlement project of Binhe Road Station No. 4 of Suzhou Metro Line 1, the improved BP neural network model is compared with the traditional grey Verhulst model and the BP neural network model.The prediction accuracy of the three models is analyzed quantitatively.The results show that the BP neural network model improved by genetic algorithm can not only make use of the original monitoring data for complex learning and information processing, but also have high fault tolerance and robustness.It is also shown that the method can comprehensively consider the influence of many factors and can be applied to the actual deformation monitoring. It is a model worthy to be adopted.
【学位授予单位】：江西理工大学
【学位级别】：硕士
【学位授予年份】：2015
【分类号】：P642.26;TP18

【共引文献】