交通运输系统工程与信息 ›› 2013, Vol. 13 ›› Issue (4): 94-99.

• 系统工程理论与方法 • 上一篇    下一篇

一种可用于高速公路投资风险评估的神经网络模型

王作功1,李慧洋1,贾元华*2   

  1. 1. 河南大学 风险管理研究所,河南 开封 475004;2. 北京交通大学 系统工程研究所,北京 100044
  • 收稿日期:2012-11-07 修回日期:2012-12-26 出版日期:2013-08-26 发布日期:2013-09-05
  • 作者简介:王作功(1967-),男,河南太康人,教授,博士.
  • 基金资助:

    国家自然科学基金项目(U0970116).

A Neural Network Model for Expressway Investment Risk Evaluation and its Application

WANG Zuo-gong1, LI Hui-yang1,JIA Yuan-hua2   

  1. 1. Institute of Risk Management, Henan University, Kaifeng 475004, Henan, China; 2. Institute of System Engineering & Control, Beijing Jiaotong University, Beijing 100044, China
  • Received:2012-11-07 Revised:2012-12-26 Online:2013-08-26 Published:2013-09-05

摘要:

高速公路投资风险评估对高速公路可持续发展具有重要作用.首先,本文针对传统的准三层BP神经网络评价模型运算精度和运算效率之间的矛盾,从其结构优化的角度出发,提出并设计了输入层神经元到隐含层神经元、隐含层神经元到输出层神经元的随机重连过程的变结构神经网络模型;其次,针对中国高速公路的投资特点,建立了高速公路项目投资风险的评价指标体系,设计了基于随机重连过程的变结构神经网络的高速公路项目投资风险评价模型,运用中国10条高速公路项目对模型进行了训练并对4条高速公路投资风险进行评估.研究结果表明,该模型预测的平均相对误差为1.91%,最大相对误差为2.63%,具有良好的预测效果.

关键词: 公路运输, 高速路, 神经网络, 变结构, 风险评价, 系统工程

Abstract:

Investment risk evaluation of expressways plays a significant role in the sustainable development of expressways. A consideration of the inconsistency between operation accuracy and operation efficiency of the traditional quasi3layer BP neural network evaluation model is first established. This paper proposes and designs a variablestructure neural network model in the random relinking process from the input layer neurons to the hidden layer neurons. Then, the model acquires the hidden layer to the output layer from the angle of structural optimization. Secondly, in view of the characteristics of the Chinese expressway investment, this paper develops an expressway project investment risk evaluation index system. Furthermore, a design of the expressway project investment risk evaluation model is completed based on the variablestructure neural network of the relinking random process. In addition, the model has been verified with ten Chinese expressway projects. The risk evaluations have been conducted for four of the ten expressway projects. The research result shows that the average relative error predicted with such model is 1.91% and the maximum relative error is 2.63%.Therefore, the prediction result is deemed suitable.

Key words: highway transportation, highway, neural network, variable structure, risk evaluation, system engineering

中图分类号: