改进BP算法在海堤渗压多测点监测预报中的应用研究

发布时间：2018-06-12 15:22

本文选题：海堤渗压 + 多测点监测　；参考：《合肥工业大学》2014年硕士论文

【摘要】：沿海地区经济发展迅猛，，为保障堤内人们的生命财产安全，海堤的重要性越来越受到重视，相关部门也加强了对海堤的保护措施。海堤是沿海而建的堤防工程，堤身一般延续很长，涉及范围广，工作环境复杂，容易出现安全隐患，因此，加强海堤安全监测是十分必要且意义重大的工作。随着海堤建设管理工作的推进，海堤堤身的安全监测已逐步成为保障海堤安全和运行的重要手段，得到了越来越多的关注。渗压是影响海堤堤身安全的主要指标，所以渗压监测在海堤安全监测分析中占有重要地位。 BP神经网络是一个相对比较成熟的网络，具有很强的非线性映射能力。考虑到海堤工作环境的复杂性，以及渗压与其影响因素之间的不明确性，本文尝试将BP神经网络应用于海堤渗压监测预报中。基于梯度下降法的标准BP算法在应用时存在不足之处，通过分析该算法的缺陷，选用优化激活函数方法和附加动量算法分别对标准BP神经网络进行改进。优化激活函数方法是在激活函数的公式中加入可调参数，对函数的陡度及映射范围进行调节，从而达到改善网络性能的目的；附加动量算法则是在每一个权值变化量的基础上加上一项正比于前一次权值变化量的值，进而加快网络权值更新，对标准BP算法进行改进。前两种方法是从两个不同的角度对标准BP神经网络进行改进，在此基础上，本文提出一种组合改进算法，即将附加动量算法与优化激活函数方法结合起来应用。以浦东某海堤实测数据为基础，考虑到单测点建模时不仅工作量大，而且各渗压测点之间信息关联度不高，本文从多测点角度出发进行建模，整体分析潮位、降雨、时效等因素对海堤渗压的影响。输入层因子选择时，以BP神经网络为分析手段，分别对简化因子形式和组合因子形式进行计算，选择预测效果比较好的一组因子作为网络模型的输入。网络结构确定后，分别对三种改进BP算法编程建模，根据训练和预测结果分析它们在海堤渗压多测点监测方面的应用情况，并比较改进模型对海堤渗压的预测效果。结果表明，三种改进BP神经网络在速度和精度方面都有所提高，其中组合改进模型比单一的改进模型具有更好的预测精度，在海堤渗压监测模型的分析预报方面取得了很好的效果。
[Abstract]:In order to ensure the safety of people's life and property, the importance of seawall has been paid more and more attention to, and the relevant departments have also strengthened the measures to protect the seawall. The seawall is the levee project built along the coast. The levee body is very long, involving a wide range, the working environment is complex, and it is easy to appear the hidden trouble of safety. Therefore, it is very necessary and significant to strengthen the safety monitoring of the seawall. With the development of the construction and management of the seawall, the safety monitoring of the seawall body has gradually become an important means to ensure the safety and operation of the seawall, and has been paid more and more attention. Seepage pressure is the main index affecting the safety of seawall levees, so seepage pressure monitoring plays an important role in the safety monitoring and analysis of seawall. BP neural network is a relatively mature network with strong nonlinear mapping ability. Considering the complexity of seawall working environment and the uncertainty between seepage pressure and its influencing factors, this paper attempts to apply BP neural network to the monitoring and forecasting of seawall seepage pressure. The standard BP algorithm based on gradient descent method has some shortcomings in application. By analyzing the shortcomings of the algorithm, the optimization activation function method and the additional momentum algorithm are used to improve the standard BP neural network. The method of optimizing activation function is to add adjustable parameters to the formula of activation function to adjust the steepness and mapping range of function, so as to improve the network performance. The additional momentum algorithm is to add a value proportional to the change of the previous weight on the basis of each weight change, and then to speed up the network weight update and improve the standard BP algorithm. The first two methods are to improve the standard BP neural network from two different angles. On this basis, this paper proposes a combinatorial improved algorithm, which combines the additional momentum algorithm with the optimization activation function method. Based on the measured data of a certain seawall in Pudong, considering the heavy workload and the low correlation degree of information between each seepage pressure measuring point, this paper analyzes the tidal level and rainfall from the point of view of multiple measuring points. Effect of aging and other factors on seepage pressure of seawall. In the selection of input layer factors, the simplified factor form and the combination factor form are calculated by BP neural network, and a set of factors with good prediction effect is selected as the input of the network model. After the network structure is determined, three kinds of improved BP algorithm are programmed and modeled respectively. According to the results of training and prediction, their application in monitoring the seepage pressure of seawall is analyzed, and the prediction effect of the improved model on seepage pressure of seawall is compared. The results show that the speed and accuracy of the three improved BP neural networks are improved, and the combined improved model has better prediction accuracy than the single improved BP neural network. Good results have been obtained in the analysis and prediction of seawall seepage pressure monitoring model.
【学位授予单位】：合肥工业大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：U657;U656.314

【参考文献】