﻿ 基于BP神经网络算法的路基工后沉降预测分析 Prediction Analysis of Subgrade Settlement after Construction Based on Neural Network Algorithm

Open Journal of Transportation Technologies
Vol.06 No.05(2017), Article ID:21894,6 pages
10.12677/OJTT.2017.65024

Prediction Analysis of Subgrade Settlement after Construction Based on Neural Network Algorithm

Rongchun Gao, Chen Chen

Received: Aug. 9th, 2017; accepted: Aug. 23rd, 2017; published: Aug. 31st, 2017

ABSTRACT

Subgrade settlement after construction plays an important role in operation and maintenance of road. As powerful nonlinear mapping ability of BP neural network algorithm, time and settlement are taken as input and output of neural network algorithm, respectively. Function relationship between settlement and time is established after neural network training. The engineering case analysis shows that the BP neural network algorithm has a certain accuracy to predict the post construction settlement of the subgrade and can meet the engineering requirements.

Keywords:Road Engineering, Post Construction Settlement, BP Neural Network Algorithm, Transfer Function, Connection Weight

1. 引言

2. BP神经网络算法

BP神经网络是一种多层前馈神经网络，该网络的主要特点是信号向前传递，误差反向传播。BP神经网络的拓扑结构如图1所示。

3. 路基工后沉降预测

1) BP神经网络初始化：以时间(d)位为输入，沉降为输出。由于输入输出节点均为1，可采用2个隐含层节点，输入层、隐含层和输出层之间的初始权值和偏置值随机确定。传递函数分别为logsig (式1)和purelin (式2)。

(1)

(2)

2) 隐含层输出计算：将时间t(d)乘以权值然后加上偏置之后代入输入层和隐含层之间的传递函数(式1)，可得隐含层的输出如式(3)。

Figure 1. The BP neural network topology

(3)

3) 输出层输出计算：将隐含层输出值Hj乘以权值然后加上偏置之后代入隐含层和输出层之间的传递函数(式2)，可得输出层的输出如式(4)。

(4)

4) 误差计算：根据式(4)计算的网络输出Ok和实测的沉降值Yk，计算网络的预测误差ek

(5)

5) 权值更新：根据网络预测误差e，按照式(6)和式(7)，更新网络连接的权值。

(6)

(7)

6) 偏置值更新：根据网络预测误差e，按照式(8)和式(9)，更新网络节点的偏置值。

(8)

(9)

7) 判断算法迭代是否满足结束条件(误差满足要求)，若不满足，返回步骤2，以下一节点数据进行计算。

4. 工程案例

Figure 2. Measured data of subgrade settlement

Figure 3. BP neural network training

(10)

Figure 4. Fitting precision of neural network

Figure 5. Fitting curve with BP neural network

Table 1. The predicted value compared with the measured value

5. 结论

Prediction Analysis of Subgrade Settlement after Construction Based on Neural Network Algorithm[J]. 交通技术, 2017, 06(05): 179-184. http://dx.doi.org/10.12677/OJTT.2017.65024

1. 1. 冯怀平, 耿会岭, 韩博文, 等. 非饱和土地区高速铁路路基沉降预测模型[J]. 岩土工程学报, 2017, 39(6): 1089-1095.

2. 2. 曹文贵, 印鹏, 贺敏, 等. 基于数据新旧程度和预测取值区间调整的沉降组合预测方法[J]. 岩土力学, 2017, 38(2): 534-540.

3. 3. 王海英, 常肖, 阮祺, 等. 建筑垃圾填埋路基沉降预测的三点-星野法[J]. 铁道科学与工程学报, 2017, 14(3): 473-478.

4. 4. 李景林, 董淑海, 杨杰, 等. 双屈服面模型在高填方路基工后沉降分析中的应用[J]. 公路, 2017(3): 1-6.

5. 5. 姜屏. 基于位移监测信息的边坡稳定性实时评判分析方法[D]: [博士学位论文]. 长春: 吉林大学, 2012.

6. 6. 冯震, 祝海涛, 熊仲华. 泊松模型预测路基沉降[J]. 华东地质学院学报, 2003, 20(4): 364-366.