基于高斯过程的变形预测算法研究

发布时间：2018-03-09 23:14

本文选题：变形预测　切入点：高斯过程　出处：《东华理工大学》2013年硕士论文　论文类型：学位论文

【摘要】：随着经济的快速发展，工程建设也日益兴起。工程建筑物的兴建，从施工开始到竣工，以及建成后整个运营期间都要不断地监测，以便掌握变形的情况，及时发现问题，保证工程建筑的安全。因此，对大型建筑物进行变形监测并且对其数据进行处理就尤为重要。目前，国内外研究变形分析模型算法有很多，尤其是以神经网络的智能算法模型。近些年来，“核学习”是机器学习领域的热点问题，其最具代表性的是支持向量机、高斯过程。高斯过程，作为新兴的机器学习方法，提供了一个原则性的、实用性的、概率的核机器学习方法。它给预测模型提供解释，并提供了模型的选择和学习框架结构。不断发展的理论与实践的应用使高斯过程成为近年来监督学习应用中强有力的竞争者，并在各个领域有着广泛的应用。本文以该角度为出发点，研究高斯过程在变形监测数据处理中的运用，主要研究内容和成果如下： 1)首先系统阐述高斯过程理论、原理和思路，运用高斯过程理论对变形监测数据进行分析，，通过实例表明高斯过程回归在变形监测的数据处理方面精度高，程序简单。 2)高斯过程模型中的超参数主要是由传统的优化方法(共轭梯度法)获得，但共轭梯度法在优化过程中存在依赖初始值、迭代次数难以确定以及局部优化等弊端。针对传统方法存在的缺陷，运用粒子群算法与高斯过程融合，建立粒子群算法高斯过程隧道位移模型。 3)通过某隧道工程进行实例分析，对变形监测数据分别采用高斯过程模型、粒子群高斯过程模型，BP模型进行处理。通过一定的误差指标评价模型的精度，得到粒子群算法高斯过程模型处理结果好，有一定的适用性。
[Abstract]:With the rapid development of the economy, the construction of engineering is also rising day by day. The construction of engineering buildings should be continuously monitored from the beginning of construction to the completion of construction, as well as throughout the period of operation after completion, in order to grasp the deformation situation and find problems in time. Therefore, it is very important to monitor the deformation of large buildings and process their data. At present, there are a lot of deformation analysis model algorithms at home and abroad, especially the intelligent algorithm model based on neural network. In recent years, "kernel learning" is a hot issue in the field of machine learning, the most representative of which is support vector machine (SVM). Gao Si process. Gao Si process, as an emerging machine learning method, provides a principled, practical, probabilistic approach to nuclear machine learning. It provides an explanation for the prediction model. It also provides the choice of model and the structure of learning framework. With the application of theory and practice, Gao Si process has become a strong competitor in the application of supervisory learning in recent years. And has been widely used in various fields. This paper studies the application of Gao Si process in deformation monitoring data processing from this point of view. The main research contents and results are as follows:. 1) firstly, Gao Si's process theory, principle and train of thought are expounded systematically, and then the deformation monitoring data are analyzed by the use of Gao Si process theory. It is shown by an example that Gao Si process regression has high precision and simple procedure in the data processing of deformation monitoring. 2) the superparameters in Gao Si's process model are mainly obtained by the traditional optimization method (conjugate gradient method), but the conjugate gradient method depends on the initial value in the optimization process. It is difficult to determine the number of iterations and local optimization. Aiming at the defects of the traditional methods, the particle swarm optimization algorithm is combined with Gao Si process to establish the tunneling displacement model of the particle swarm optimization algorithm. 3) through the analysis of a tunnel project, the deformation monitoring data are processed by Gao Si process model and particle swarm Gao Si process model and BP model respectively. The accuracy of the model is evaluated by certain error indexes. The result of processing Gao Si process model of particle swarm optimization algorithm is good and has certain applicability.
【学位授予单位】：东华理工大学
【学位级别】：硕士
【学位授予年份】：2013
【分类号】：P227

【参考文献】