Computational Intelligence and Neuroscience

Volume 2018, Article ID 8547313, 16 pages

https://doi.org/10.1155/2018/8547313

## Risk Evaluation Model of Highway Tunnel Portal Construction Based on BP Fuzzy Neural Network

School of Civil and Architecture Engineering, Xi’an Technological University, Xi’an 710032, China

Correspondence should be addressed to Tian Xu; moc.361@11527294251

Received 3 December 2017; Revised 19 May 2018; Accepted 19 June 2018; Published 28 August 2018

Academic Editor: Juan L. Pérez-Ordoñez

Copyright © 2018 Xianghui Deng et al. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

#### Abstract

Risk assessment for tunnel portals in the construction stage has been widely recognized as one of the most critical phases in tunnel construction as it easily causes accident than the overall length of a tunnel. However, the risk in tunnel portal construction is complicated and uncertain which has made such a neural network very attractive to the construction projects. This paper presents a risk evaluation model, which is obtained from historical data of 50 tunnels, by combining the fuzzy method and BP neural network. The proposed model is used for the risk assessment of the Tiefodian tunnel. The results show that the risk evaluation level is IV, slope instability is the greatest impact index among four risk events, and the major risk factors are confirmed. According to the evaluation results, corresponding risk control measures are suggested and implemented. Finally, numerical simulation is carried out before and after the implementation of risk measures, respectively. The rationality of the proposed risk evaluation model is proved by comparing the numerical simulation results.

#### 1. Introduction

Over the past few decades, construction of highways has been developing quickly in China. Tunnel construction has become the first choice for highway alignment because of its advantages of optimal alignment, reduced travelling time, and enhanced operation efficiency. Meanwhile, tunneling is a dangerous occupation owing to its complicated construction technology, uncertainties risk factors, and complicated geological conditions. There are numerous casualties and millions of economic losses caused by tunnel accidents. Compared with the overall length of a tunnel, portals usually have a limited area of influence. The weathering extent of surrounding rock is heavier, the buried depth is shallow, and it is vulnerable to the impact of rainfall. In the entrance section, the construction is difficult, and these unfavorable factors easily lead to engineering accidents, such as slope instability, large deformation, tunnel collapse, and other accidents [1]. In most instances, risk assessment may reduce the probability of accidents and decrease economic costs. Therefore, it is necessary to take the modern risk management method to evaluate the risk of the tunnel portal construction process.

Risk assessment for highway tunnel portals during the construction stage has been widely recognized as one of the most critical phases. However, risk factors of tunnel portal construction are complex and uncertain. At the same time, most of the information in risk evaluation comes from the expert’s subjective judgement which is usually imprecise and subjective in the decision-making process. Handling vagueness and subjectivity becomes a primary task in risk assessment.

The fuzzy system describes a class of extrapolated blurring without explicit boundaries and establishes a correspondence between uncertainties and membership functions so that favorable mathematical tools can be used to analyze many inaccurate vague phenomena in nature. Fuzzy theory has found in-depth research and application in the mining, nuclear, petrochemical, and construction industries. van Laarhoven and Pedrycz introduced the concept of fuzzy set theory into the traditional analytic hierarchy process (AHP) and originally proposed the fuzzy analytic hierarchy process (FAHP) [2, 3]. Fuzzy hierarchy evaluation is a synthetic risk evaluation method based on AHP and fuzzy comprehensive evaluation [4, 5]. It is well known as a useful tool to deal with imprecise, uncertain, or ambiguous data and the high nonlinearity and complexity [6–8]. The uncertain comparison judgement can be represented by the fuzzy number. To deal with the uncertainty or vagueness of data, the fuzzy analytic hierarchy process (FAHP) has found huge applications in recent years [9]. Compared with the overall length of a tunnel, portals usually have the complicated terrain and poor geological conditions. Thus, the construction of a tunnel portal is often difficult and easily leads to engineering accidents [10]. Wang et al. applied the logarithmic fuzzy preference programming (LFPP) method to analyze the data [11]. Although FAHP has good applicability in construction industries than traditional risk assessment methods such as AHP, it still has subjectivities to identify the weight and set a single-factor judgement matrix in risk evaluation.

Artificial neural network (ANN) is also known as a neutral network which is a mathematical model for finding patterns among datasets where there are complex relationships between the inputs and outputs. ANN attempts to simulate the structure and operation of the human neural network system. Since ANN acts like a “black-box” and cannot explain the reasoning process, it can well achieve the self-adaptation through the learning function and can acquire the fuzzy data expression knowledge accurately and automatically. The ability to learn from examples has made this technique a very useful tool in data modeling [12–14]. In the past few decades, researchers applied ANN in many aspects of construction management such as risk analysis, resource optimization, and productivity assessment [15]. ANN is capable of learning from the data; however, it cannot explain the quality of the input-output mapping process. On the contrary, the fuzzy system is a systematic reasoning method that is more compatible with human logic and intuition.

ANN and fuzzy theory are complementary technologies. Fuzzy theory tries to describe and deal with the ambiguity concept in human language and thinking. The artificial neural network is based on the human brain’s physiological structure and information-processing process. With the rapid development of fuzzy system and artificial neural network research, it has been found that the original independent field can be compensated and fused together, which leads to a new field—fuzzy neural network (FNN). FNN provides effective tools for addressing uncertainties in decision-making [16]. In the past few years, this system has been widely applied to develop risk management models in the engineering construction. Wang et al. combined AHP with the backpropagation (BP) neural network (which is a multilayer error-feedforward network), and they proposed a model of coal mine water disaster emergency logistics risk assessment [17]. Mirahadi and Zayed developed a modified neural-network-driven fuzzy reasoning (NNDFR) model with optimized parameters and improved the developed model so that it can simultaneously deal with crisp values and fuzzy numbers [18]. Li et al. proposed an analytic hierarchy process model for the transformer risk assessment built by a transformer risk assessment method based on FAHP and artificial neural network (ANN) [19].

For the purpose of handling the vagueness and subjectivity in risk evaluation of highway tunnel portal construction, this paper proposes a risk evaluation model by combining fuzzy theory with the neural network. There are many kinds of neural network types. This paper uses the BP network—a multilayer feedforward neural network—which can achieve any nonlinear mapping from the input to output. 80% to 90% of the neural network model uses the BP network or its change form. The feasibility and effectiveness of this model are proved by an engineering case.

The remainder of this research is organized as follows: Section 2 introduces the BP fuzzy neutral network model in detail. Section 3 shows the case study. Section 4 verifies the accuracy of the model. This paper is concluded in the last section.

#### 2. BP Fuzzy Neutral Network Model Development

##### 2.1. Design of the Topology Structure

The BP neural network generally has three or more layers of neurons. There are input layer, hidden layer, and output layer, respectively. According to the Kolmogorov theorem, this model uses a three-layer BP neural network with a single hidden layer [20]. There are some debates about the relevance of Kolmogorov’s theorem. Girosi and Poggio have criticized this interpretation of Kolmogorov’s theorem. They reviewed Kolmogorov’s theorem on the representation of functions of several variables in terms of functions of one variable and showed that it is irrelevant in the context of networks for learning. However, Kůrková supported the relevance of this theorem to neural nets. He gave a direct proof of the universal approximation capabilities of perceptron-type networks with two hidden layers by taking advantage of techniques developed by Kolmogorov. He proved the feasibility of the Kolmogorov theorem in the BP neural network [21, 22]. The topology was built with the BP neural network combined with the characteristics of the highway tunnel portal construction, which is shown in Figure 1.