Study on the Identification of Opportunistic Behavior of Subway Project Construction Enterprises

Abstract

With the rapid development of urban rail transportation, people’s demand for subways has gradually manifested itself. The inherent complex attributes of subway project construction determine that subway project construction has a relatively high risk, resulting in huge losses. This paper takes the opportunistic behavior of the subway project as the research object, proposes the opportunistic behavior identification process, and constructs the opportunistic behavior identification model based on the BP neural network. Firstly, through the collection and analysis of subway accident cases, the main forms of opportunistic behavior are summarized, and the primary characteristic indicators for opportunistic behavior recognition are extracted using cluster analysis. Secondly, a recognition model based on a BP neural network is designed. The number of neurons in the input layer, hidden layer, and output layer of the model is determined, and the recognition model is subsequently trained and tested to validate its feasibility. Finally, the constructed opportunistic behavior recognition model is applied to an actual subway construction project, revealing that the construction enterprise of the project in question exhibits a high level of opportunistic behavior risk. Overall, the research results of this paper have important theoretical significance and practical value for the management level of subway project construction enterprises under the new situation and the identification and governance of opportunistic behavior of subway project construction enterprises.

1. Introduction

With the continuous urban development, subway construction has encountered unprecedented opportunities and is progressing and expanding at an accelerated pace. According to data released by the China Urban Rail Transit Association, as of 2021, 56 cities have unveiled plans for rail transit construction, encompassing a total length of 6988.3 km with the completion of 3828 stations. These data fully illustrate the rapid development and growing trend of urban rail transit, providing solid support for its future development. With the rapid development of subway construction, the complexity of its underground construction technology, the huge size of the construction scale and the multiplicity of risk factors have given rise to a series of negative impacts, such as opportunistic behavior of subway project construction enterprises. These opportunistic behaviors not only cause great losses to the national economy but also lay hidden dangers to the operation safety of China’s urban rail transit. The so-called construction party opportunistic behavior, that is, for the use of regulatory loopholes, system deficiencies to achieve the purpose of distortion or concealment of information, the use of information asymmetry to avoid the construction responsibilities that should be performed, the choice of companions or groups of people to sacrifice the interests of others in the cooperation, in order to make their own interests at the expense of the project’s overall safety and quality of production value and the adoption of devious speculative behavior. Practice shows that the existence of opportunistic behavior in construction projects not only leads to huge economic losses but also causes safety and quality accidents.

In recent years, academics have conducted in-depth research on the mechanism of opportunistic behavior, identification methods, prevention strategies, etc. Doree and Johan both used game methods to explore the problem of collusion of construction companies in bid-rigging in the Dutch construction market [1]. Due to many reasons, the construction unit will take improper means to seek its own interests; thus, there are many kinds of opportunistic behaviors, among which jerry-building, slack delay, and collusion are the three most common forms [2]. Liu et al. proposed an optimal loss-bearing mechanism according to the loss caused by “lazy” behavior so as to protect the interests of the company and effectively prevent the occurrence of opportunistic behaviors [3]. Guo et al. put forward a new point of view through the analysis of engineering reality: that is, the construction unit seeks rent from the owner’s agent, which is the root cause of the frequent occurrence of engineering accidents [4]. Khaled et al.’s research shows that due to job changes and the lack of owner management, the contractor’s ‘opportunistic’ bidding behavior will be affected. Therefore, they proposed a series of bidding decision strategies to help contractors reduce the resulting injustice and also help the project proceed smoothly [5]. Hua et al. used the web crawler software GooSeeker (V10) to obtain the relevant data to carry out an empirical study, which confirmed the feasibility of the identification of opportunistic behaviors [6]. Meng conducted a study on the identification of hazard sources in construction, proposing to enhance dynamic identification and hierarchical management of hazard sources, establish a safety management platform, and optimize the construction environment to mitigate safety incidents and ensure efficient and secure construction [7].

In summary, opportunistic behavior shows strong context dependence in different research fields and problem contexts. Its psychological characteristics, behavioral patterns, and manifestations also have significant differences. Taking the construction side of subway projects as the research object, this paper aims to analyze the application of the identification of opportunistic behavior in practical situations. In terms of safety management on the construction side, this study focuses on typical opportunistic behaviors such as risky operations, speculation, and inexperience on the part of the participants. By deeply exploring the characteristics and manifestations of these opportunistic behaviors, this study aims to provide useful references and suggestions for the management of metro projects, thus making its due contribution.

2. Literature Review

Because there are many negative impacts due to opportunistic behavior in subway construction, many fields have begun to focus on the study of opportunistic behavior identification, and depending on the seriousness of opportunism, appropriate management measures should be taken to inhibit its development.

In recent years, opportunistic behavior identification studies have begun to receive attention in domestic and international studies, especially in the field of subway engineering, such as subway construction. These studies not only focus on concept establishment and qualitative research, but also on reliability calculations of structures and rock-on-rock medium materials, with a view to better identifying opportunistic behaviors. The Copenhagen metro collapse triggered global attention, which led to a focus on risk analysis in large-scale projects, especially in relation to underground engineering. Over time, risk management has become an important part of many underground engineering projects. The concept of risk analysis in tunneling was first introduced by Professor Einstein, who emphasized its importance and provided important guidelines for the safety of tunneling projects, as well as the principles that should be followed. He also proposed a risk management model for tunnels based on computer simulations, ushering in a whole new era [8]. Since then, more scholars have conducted more in-depth discussions in this area.

Chung et al. conducted research for the study of protection and reduction of construction cycle time with the aim of identifying the most widely used tools as well as tools related to successful project management, especially those related to effective project risk identification management [9]. Yu brought the concept of risk analysis. The application of this method in the field of opportunistic behavior identification focuses on its basic ideas, principles, and detailed steps, including the analysis of major risk elements, risk assessment system, planning, and decision-making for major risk solutions, using which potential risks can be effectively predicted and the cost of completion can be accurately controlled so that the cost of the project can be controlled at an acceptable level, with a focus on risk-based identification, assessment, monitoring, control, and decision analysis [10].

Zhou et al. conducted a study on emergency risk identification for subway construction collapse accidents, described the necessity of establishing emergency preparedness for subway engineering and construction in response to subway construction collapse accidents, identified potential risks and their results during subway construction of tunnel boring machines, and established emergency preparedness for the outbreak of collapse accidents by using FMEA [11]. Jodl et al. provided an overview of various risk identification and defined risk as a description of an event that unimpeded, will most likely lead to unintended consequences and resulting damage. The magnitude of the risk depends on the extent of the damage and the likelihood of its occurrence [12]. Ge et al. accurately identified the risk factors, cleared the possible impacts of any risk factors on the construction project, and used a dynamic approach to establish a dynamic risk assessment index system based on the construction planning of the earth and rock dams and the possible means by which any hazardous factors could interfere with the construction objectives [13].

Zhao introduced the theory and method of risk analysis, took the highway bridge project as an example to discuss the possible risks in the construction, and put forward some effective identification methods from the three aspects of engineering measures, non-engineering measures, and biological measures [14]. Yu focused on the identification of construction risks in large airport terminals. Through the collection and analysis of cases, the construction risks associated with non-stop renovation of several large airport terminals were studied, and the characteristics of non-stop renovation construction were summarized, which played a guiding role in controlling construction risks [15]. Su concluded that the construction technology of subway stations has matured, but the construction quality of interval tunnels is crucial, and it discussed the risk identification analysis and control analysis of subway shield interval tunnel construction [16]. Li et al. used the BP neural network’s intelligent algorithm to accurately identify the various risks that may exist in the subway construction project and, combined with the actual situation, crafted a set of perfect risk management strategies to ensure the smooth implementation of the project, and at the same time, it can also better control and optimize the safety of the construction in order to ensure the construction safety [17]. Due to the special nature of the subway transportation environment and the high demand for construction requirements, it is necessary to clearly recognize the possible sources of hazards in construction to ensure the safe construction and operation of subway construction [18].

When studying the impact of project governance on project success in the field of PPP projects, Maqbool used quantitative survey and qualitative case study methods to find that opportunism and project uncertainty have a marginal effect on sustainable development and project success in the context of countries such as the United Kingdom [19]. Through the survey data of 323 general contractors in China’s construction industry, Tang found that both result control and clan control can inhibit the opportunistic behavior of subcontractors, and the dispersion of subcontractors has a significant moderating effect [20]. Using the partial least squares structural equation model (PLS-SEM), Zhang proved that contract flexibility can inhibit the opportunistic behavior of contractors, and this relationship is mediated by the perception of fairness among contractors’ organizations. In addition, the study also reveals the important role of communication quality in strengthening the impact of contract flexibility on inter-organizational fairness perception and opportunistic behavior [21].

3. System Construction

3.1. Main Manifestations

In the subway project, the construction party’s opportunistic behavior is a prevalent issue. Through the collection and analysis of subway accident cases, the primary manifestations of the construction party’s opportunistic behavior in subway projects can be summarized as follows.

(1) Awareness level: The awareness level of the constructor of a metro project refers to the extent to which the staff pays attention to safety and understands safety management in the process of metro construction. This includes the identification and control of hazardous factors on the construction site, conscious compliance with safety norms and operating procedures, and timely response to accident risk assessment and emergency treatment. Improvement of subway construction safety awareness can reduce accidents in the construction process, to protect the normal progress of construction work and staff safety. Workers’ awareness and responsibility: Workers’ awareness and responsibility have a great impact on the safety of metro construction. If workers do not have enough awareness of the risks of construction or have a weak sense of responsibility, they may ignore the safety risks and lead to accidents.

(2) Psychological factors: The psychological factors of subway construction safety refer to the influence of the psychological state of workers on the construction in the process of subway construction. This factor includes the following aspects: workers’ mental health and stability: Workers’ mental health and stability have an important impact on construction safety. If workers are not psychologically stable, negative emotions such as anxiety and nervousness may occur, leading to construction errors and accidents. Workers’ ability to communicate and cooperate: In subway construction, communication and cooperation among workers are needed to accomplish tasks. If there is poor communication or cooperation between workers, misunderstandings or construction accidents may easily occur.

(3) Unsafe physical state: The unsafe physical state of subway construction refers to the physical state that may cause safety hazards in the process of subway construction, such as construction tools, mechanical equipment, building materials, dangerous goods, etc. These objects may cause serious threats and hazards to the workers at the construction site and the surrounding social environment, so it is necessary to take corresponding safety measures to prevent and avoid potential risks.

(4) Unsafe workspace: Unsafe workspace for metro construction refers to a risky work environment that may lead to injuries, waste formation, fire, and other accidents in the process of metro construction or maintenance where construction workers or staff must operate in a narrow, restricted or hazardous elemental space. For example, subway road tunnels, pipelines, cable erection, blind holes, and mine shafts may become unsafe workspaces. Therefore, attention should be paid to the relevant safety regulations and procedures in subway construction to ensure the safety, life, and health of the staff.

(5) Poor hydrogeology: Poor hydrogeology of subway construction refers to the decline in construction quality and lag in project progress during subway construction due to factors such as water table, stratigraphic structure, underground water flow, and geological structure. For example, hydrogeological problems encountered during construction may include foundation water surges, groundwater infiltration, earth collapses, rock cracking, etc., which will affect the stability and safety of subway tunnels. In order to effectively deal with these problems, the subway construction project must carry out adequate hydrogeological exploration and technical research and take appropriate measures to ensure construction safety and quality.

(6) Underground integrated pipeline complex: Subway integrated pipeline complex subway construction refers to the subway construction process. The ground has a variety of pipelines and facilities that are entangled with each other, dense, and intricate. These pipelines and facilities include water supply, drainage, electricity, communications, gas, and other infrastructure, as well as underground roads, underground shopping centers, and other construction facilities. As these pipelines and facilities are conflicting or even overlapping, they may lead to great risks and challenges during the subway construction process, increasing the difficulty and cost of subway construction. Therefore, detailed surveys, planning, and coordination are necessary to avoid damage to underground pipelines and facilities, etc., and to ensure construction safety and quality.

(7) Opportunistic management: The opportunistic management of metro construction refers to the construction process in which the managers, based on their own profit motives, are centered on the pursuit of personal interests, ignoring the long-term interests and planning of the whole project, and even the phenomenon of malicious destruction and bad competition. This kind of management will cause problems such as reduced project quality, delayed schedule, and increased costs, which will have an adverse effect on the whole subway project.

(8) Lack of safety culture: The lack of safety culture of subway construction managers refers to the lack of attention to safety management in the construction process of subway construction managers, the lack of safety awareness and culture, easy to ignore the safety risks, and lax management, resulting in the construction of the probability of safety accidents increases. This situation is manifested in the following ways: Lack of understanding of safety regulations and standards, lack of full awareness of safety risks and countermeasures in the construction process, failure to conduct the necessary safety training and education, and inability to effectively transmit safety awareness and the culture of attaching importance to safety, insufficiently stringent control of technical quality, non-standardized project management, and lack of effective safety management measures. The lack of safety culture of subway construction managers will seriously affect the safety and quality of subway construction and should be given enough attention and improvement.

3.2. Cluster Analysis

In order to further study the level of opportunistic behavior in various regions, based on the data of opportunistic behavior cases in various regions of China as of 2022, cluster analysis was conducted on the subway construction status of 43 cities under construction in the ‘China Statistical Yearbook’ of the National Bureau of Statistics based on SPSS software (19.0). A total of 477 subway construction cases were collected.

Table 1. National statistics on opportunistic cases of urban rail transit.

No.	Cities	Mileage (km) X1	Human Factor Accident Cases X2	Management-Related Factor Accident Cases X3	Cases of Physical Factor Accidents X4	Cases of Accidents Involving Factors of the Environment X5
1	Shanghai	825	24	16	6	2
2	Beijing	797.3	18	12	5	2
3	Guangzhou	609.8	11	7	3	1
4	Shenzhen	558.6	15	10	4	2
5	Chengdu	557.8	14	10	4	1
6	Hangzhou	516.0	10	8	3	1
7	Wuhan	504.3	10	8	3	1
8	Nanjing	448.8	7	4	1	0
9	Chongqing	434.6	10	7	2	0
10	Qingdao	323.8	7	6	2	1
11	Tianjin	286.0	4	3	1	0
12	Xi’an	272.4	6	4	1	0
13	Suzhou	254.2	6	5	2	1
14	Zhengzhou	233.0	3	2	0	0
15	Shenyang	216.7	5	3	1	0
16	Dalian	212.6	7	5	2	1
17	Changsha	209.1	6	4	1	0
18	Ningbo	186.0	7	5	2	1
19	Hefei	168.8	7	5	2	0
20	Kunming	165.9	4	3	1	0
21	Nanchang	128.5	7	4	1	0
22	Nanning	128.2	4	3	1	0
23	Foshan	127.3	3	2	0	0
24	Wuxi	110.8	4	3	1	0
25	Fuzhou	110.7	3	2	1	0
26	Changchun	106.7	4	3	1	0
27	Xiamen	98.4	3	2	0	0
28	Jinan	84.1	5	3	1	0
29	Harbin	78.1	5	4	1	0
30	Guiyang	74.4	4	3	1	0
31	Shijiazhuang	74.3	5	4	1	0
32	Xuzhou	64.1	3	2	0	0
33	Changzhou	54.0	4	3	1	0
34	Wenzhou	52.5	0	1	0	0
35	Hohhot	49.0	1	1	0	0
36	Shaoxing	47.1	2	1	0	0
37	Wuhu	46.2	1	1	0	0
38	Luoyang	43.5	3	2	0	0
39	Nantong	38.5	1	1	0	0
40	Dongguan	37.8	2	1	0	0
41	Urumqi	26.8	1	1	0	0
42	Lanzhou	25.5	0	0	1	0
43	Taiyuan	23.3	1	1	0	0

selects the values of five indicators of urban rail transit opportunistic behavior data of Chinese provinces, cities and autonomous regions up to 2022 and summarizes eight relevant variables used to describe opportunistic behavior in urban rail transit projects in the cases, including the number of subway construction mileage, the number of human factor accident cases, the number of physical factor accident cases, the number of management factor accident cases, and the number of environmental factor accident cases. The number of subway construction mileage was obtained from the National Rail Transit Association statistics.

Each sample is described by five uncorrelated vectors, and the entire database is represented by a 43-row, 5-column matrix. The form of describing the opportunistic behavior of each region in a matrix facilitates the continuous improvement of the vector of opportunistic behavior characteristics by adding or removing variables.

Step 1: the min-max standardization (normalization) process.

Clustering will enable similar behaviors to be grouped together. The similarity of opportunistic behaviors across regions is assessed by a prescribed distance metric. In order to quantify the distance between opportunistic behaviors in the database, the data min-max standardization (normalization) process is first performed to make the data metrics comparable to each other, which is calculated by the formula:

$$X_{ij}^{\ast }={\displaystyle \frac{X_{ij}-{\mathit{min}}_{1\le i\le n}\left\{X_{ij}\right\}}{R_{ij}}}$$

(1)

$$R_{ij}={\mathit{max}}_{1\le i\le n}\left\{X_{ij}\right\}-{\mathit{min}}_{1\le i\le n}\left\{X_{ij}\right\}$$

(2)

Step 2: calculate the class-to-class distance.

The Euclidean distance Equation (3) is used to define the distance between two accident vectors.

$$d=\sqrt{{\sum _{j=1}^5\left(x_{ij}-y_{ij}\right)}^2},i=1,2,\cdots ,n;j=1,2,3,\cdots ,5$$

(3)

In Equation (3), $i$ is the number of rail transit operating cities in the database, $i=43$; $j$ is the $j$th constituent element of each opportunistic behavior; $x_{ij}$ and $y_{ij}$ are vectors taken as values; and $d$ is the value of the matrix element (i.e., the distance between behaviors). The distance between any pair of behaviors is represented by an $n\times n$ matrix.

Table 2. Summary of case processing ^a,b.

Validity		Deficiencies		Total
Cases	Percentage	Cases	Percentage	Cases	Percentage
43	100%	0	0	43	100.0%

* ^a Squared Euclidean distance in use. ^b Average linkage (between groups).

indicates that out of 43 samples, there are 43 valid cases and 0 invalid cases, the sample validity rate is 100%, and the cluster analysis results are reliable.

Step 3: data clustering analysis based on SPSS.

Enter the SPSS program, select the systematic clustering function, carry out the systematic cluster analysis (Hierarchical Cluster Analysis), import X1 to X4 as variables, and carry out the systematic cluster analysis of the samples.

The significance of each item listed in

Table 3. Cluster analysis of cases.

Stage	Cluster Combination		Coefficients	Stage at Which Clustering First Occurs		Next Stage
	Cluster 1	Cluster 2		Clustering 1	Clustering 2
1	39	40	1.490	0	0	21
2	36	37	1.810	0	0	8
3	30	31	2.010	0	0	10
4	24	25	2.010	0	0	13
5	4	5	2.640	0	0	34
6	41	42	3.690	0	0	9
7	22	23	3.810	0	0	12
8	35	36	6.225	0	2	14
9	41	43	9.545	6	0	28
10	29	30	15.065	0	3	19
11	16	17	16.250	0	0	17
12	21	22	16.265	0	7	29
13	24	26	17.405	4	0	20
14	35	38	20.833	8	0	18
15	19	20	22.410	0	0	26
16	33	34	28.250	0	0	18
17	15	16	43.285	0	11	27
18	33	35	58.475	16	14	21
19	28	29	76.377	0	10	24
20	24	27	126.980	13	0	29
21	33	39	128.188	18	1	28
22	6	7	136.890	0	0	34
23	11	12	189.960	0	0	30
24	28	32	208.283	19	0	32
25	8	9	220.640	0	0	39
26	18	19	357.925	0	15	33
27	14	15	434.353	0	17	33
28	33	41	472.835	21	9	32
29	21	24	487.588	12	20	36
30	11	13	677.740	23	0	35
31	1	2	820.290	0	0	41
32	28	33	1367.254	24	28	36
33	14	18	2132.618	27	26	38
34	4	6	2369.185	5	22	37
35	10	11	2981.320	0	30	38
36	21	28	4656.259	29	32	40
37	3	4	6326.032	0	34	39
38	10	14	8486.066	35	33	40
39	3	8	13,056.646	37	25	41
40	10	21	28,741.802	38	36	42
41	1	3	89,345.239	31	39	42
42	1	10	237,474.788	41	40	0

is as follows: stage—clustering step number; cluster combination—cases merged in a particular step; coefficient—sample Euclidean distance (the larger the value, the more dissimilar the two are); stage at which clustering occurs for the first time—newly generated clusters; and, in the next step, the new class will be merged with the other classes or cases that fulfill a specific need.

As shown in Table 3, in stage 31, for example, cluster 1 Shanghai and cluster 2 Beijing are merged into one category, and their coefficients are 820.290. The mileage of rail transit in Shanghai is 825 km, and the number of accidents caused by human factors is 24, the number of accidents caused by managerial factors is 16, the number of accidents caused by physical factors is 6, and the number of accidents caused by environmental factors is 2. The mileage of Beijing’s rail transit is 797.3 km, and there are 18 cases of accidents caused by human factors, 12 cases of accidents caused by management factors, 5 cases of accidents caused by physical factors, and 2 cases of accidents caused by environmental factors. Cluster analysis classified them into one category, indicating that both are very similar in all indicators. After 42 steps, all cluster clustering is completed.

3.3. Identification Feature Analysis

It can be found through the analysis that the clustering results of the cases of opportunistic behavior of the construction side of the subway project in China, although jointly affected by a variety of indicators, its clustering still presents a certain logic from the specific classification type to get the quantitative indicators that may affect the generation of opportunistic behavior, which can be specifically divided into the following categories.

(1) Schedule delay: There are strict schedule requirements for subway projects, which may induce the emergence of opportunistic behavior when the constructor is delayed in the schedule. Therefore, the schedule delay can be regarded as a quantitative indicator to assess the possibility of opportunistic behavior.

(2) Degree of development of construction cities: Developed cities tend to have more mature and standardized systems in construction management, including project management, supervision systems, and technical standards. These factors can reduce the likelihood of opportunistic behavior and increase the extent to which the constructor performs in accordance with the law. According to the ‘2020 China Census County Data’, cities can be divided into five categories and seven grades, with cities having a population of over 10 million classified as supercities, cities with a population ranging from 5 million to 10 million considered megacities, cities with 3 million to 5 million inhabitants designated as type I large cities, and cities with 1 million to 3 million inhabitants labeled as type II large cities.

(3) Size of station area and station burial depth: Through the comparative analysis of the five clustering results, it is found that the excavation area of the station is large, which makes the construction time prolonged, and the shape of the foundation pit is not regular enough, and the support structure is complex, which brings great challenges to the stability of the foundation pit, and there will also be the problem of deep and shallow pits excavation. The subway station is divided into island platforms, side platforms, mixed platforms, etc. Different station structure behaviors may lead to different probabilities of subway construction accidents.

(4) Geological conditions: In recent years, due to the increasing scale of urban rail transit, they have faced increasingly complex construction conditions, including high-rise buildings, water systems, mountain ranges, rocks, and geological formations, making the construction process very difficult. This poses a great challenge to the survey, design, construction, risk management, operation, and maintenance of urban rail transit.

(5) Construction indices: The construction of building projects is affected by many factors. In addition to the traditional influence of meteorological conditions, in different construction segments, different humidity, wind, air pollution, and rainfall will also have different degrees of potential impact on construction conditions, which are often not easily detected but have a significant impact on the consumption of bulk building materials such as cement and rebar. According to the national construction index report issued by the Ministry of Industry and Information Technology of China, the comprehensive data suitable for construction in various regions of the country can be viewed in real-time, which can be used as one of the reference indicators for opportunity identification indicators.

(6) Construction methods: With the changes in the regional construction environment, the methods used in the construction of subway lines in China are also changing. At present, the main construction methods are the open-cut method, mine method, and shield method. Among them, the open-cut method has the advantages of being a simple process, safe and reliable, easy to control, parallel or cross operation can be used, and the construction can be flexibly arranged according to the schedule, and the comprehensive cost is also relatively low. In the construction of subway stations, the surrounding environmental conditions should be considered, such as spacious sites, underground pipelines that can be safely relocated, road traffic convenience, etc., to ensure the construction quality and safety; the mine method construction can be borrowed from the construction experience of mountain tunnels, but due to the ground settlement being too large, it makes the construction cost high, and the risk is also higher, so it is less often used. In contrast, the shield method of construction is more advanced, with mechanical digging, rapid construction, and ease of control, so the conditions of the section are usually preferred.

4. Research Methodology

The BP neural network is a multi-layer feedforward network trained by an error backpropagation algorithm. According to Kolmogorov’s theorem, BP neural networks have excellent multi-as-function mapping ability, and a three-layer BP neural network can realize the approximation of arbitrary nonlinear functions. In a simple BP neural network structure, each neuron takes the output of the previous layer of neurons as input, outputs its own computation and passes it as input to the next layer. For the opportunistic behavior recognition model of the subway construction party in this study, the input variables include many factors such as environmental factors and construction methods, and the BP neural network can learn and store a large number of input-output mode-mapping relationships, which is very important for capturing and analyzing various possible trigger factors of opportunistic behavior. BP neural networks can be single-layer or multi-layer, and the neural network has an input layer, an output layer, and a hidden layer, as shown in Figure 1.

Firstly, data preparation is carried out, i.e., input data and corresponding output result data are prepared, which can be obtained through field experiments or historical data organization. Then, on the basis of data preparation, parameter initialization is carried out, i.e., the weights and biases in the network are adjusted so that the network can better fit the input data and output results in the subsequent training process.

After parameter initialization, the output results of the network are calculated by forward propagation, i.e., the output values of the network are calculated based on the input data and parameters. Next, the gradient of each parameter in the network is calculated by error backpropagation, and the parameters are updated using an optimization algorithm. In this process, the forward propagation and backpropagation steps need to be repeatedly performed until a preset number of training iterations is reached or the error reaches a small threshold. This allows the network to gradually converge on the training data and learn the mapping relationship between the input data and the output result. The forward propagation process of the signal is as follows:

$$o_i=\phi \left(ne{t}_i\right)=\phi \left({\sum }_{j=1}^M{w}_{ij}{x}_{ij}+{\theta }_i\right)$$

(4)

$$o_k=\psi \left(ne{t}_k\right)=\psi \left({\sum }_{i=1}^q{w}_{ki}{o}_i+a_k\right)=\psi \left[{\sum }_{i=1}^q{w}_{ki}\varphi \left({\sum }_{j=1}^M{w}_{ij}{x}_j+{\theta }_i\right)+a_k\right]$$

(5)

where $x_j$ denotes the input of the $j$th node of the input layer, $j=1,\cdots ,M$; $w_{ij}$ is the weight of the $i$th node of the implicit layer to the $j$th node of the output layer; ${\theta }_i$ is the threshold of the $i$th node of the implicit layer; $\phi$ is the incentive function of the implicit layer; $w_{ki}$ denotes the weight of the $k$th node of the output layer to the $i$th node of the implicit layer, $i=1,\cdots ,q$; $a_k$ is the output layer of the $k$th node’s threshold, $k=1,\cdots ,L$; and $\psi$ denotes the excitation function of the output layer. The backpropagation process of the signal is as follows:

$$\Delta w_{ij}=\eta \sum _{p=1}^p\sum _{k=1}^L\left(T_k^p-o_k^p\right){\psi }^{\prime }\left(ne{t}_k\right)w_{ki}{\phi }^{\prime }\left(ne{t}_i\right)x_j$$

(6)

$$\Delta {\theta }_i=\eta {\sum }_{p=1}^p{\sum }_{k=1}^L\left(T_k^p-o_k^p\right){\psi }^{\prime }\left(ne{t}_k\right)w_{ki}{\phi }^{\prime }\left(ne{t}_i\right)$$

(7)

where $\eta$ is the learning rate; $\Delta w_{ki}$ is the output layer weight correction; $\Delta a$ is the output layer threshold correction; $\Delta w_{ij}$ is the implied layer weight correction; and $\Delta {\theta }_i$ is the implied layer threshold correction.

After the training is completed, the trained model needs to be evaluated using a portion of the data to verify the generalization ability and performance of the model, and the recognition ability of the model on new input data can be evaluated by these metrics. Finally, the trained model is applied to the new input data for the task of identifying opportunistic behaviors of the constructor of a metro project in order to improve construction safety and efficiency.

4.1. Determination of the Number of Neurons in Each Layer of the BP Neural Network Recognition Model

4.1.1. Number of Neurons in the Input Layer

The advantage of the artificial neural network lies in avoiding a large number of tedious calculations, making the identification work simpler and easier, and weakening the human factor in determining the weights of factors in the evaluation process. The previous section organizes some of the indicators that may affect the emergence of opportunistic behavior, laying the foundation for the establishment of the model. The quantized value of the feature only represents the attribute and does not represent the weight, and the size of the weight is learned by the neural network autonomously. The meanings of the eigenvalues of specific features are shown in

Table 4. Quantification of eigenvalues for identifying opportunistic behaviors of subway project construction enterprises.

Attribute Characterization Quantitative Values	1	2	3	4	5
Total mileage U1	<10 KM	10~20 KM	20~30 KM	30~40 KM	>40 KM
Total investment U2	CNY < 10 billion	CNY 10~20 billion	CNY 20~30 billion	CNY 30~40 billion	CNY > 40 billion
Station area U3	<2500 m2	2500~10,000 m2	10,000~25,000 m2	25,000~50,000 m2	>50,000 m2
Station buried depth U4	<13 m	13~18 m	18~23 m	23~30 m	>30 m
Station structural form U5	Island platform	Side platform	Mixed platform
Enclosure structure form U6	Retaining wall	Deep soil mixing piles	Pile enclosure system	Diaphragm wall	Else
Excavation method U7	Open excavation	Mining method	Shield method	Sunken tube method	Else
Lining method U8	Integral molded concrete lining	Modular lining	Cave pile method lining	Composite lining	Double spectacle method lining
Geological condition U9	General	Relatively complicated	Complicated
Degree of development of construction cities U10	Type II large city	Type I large city	Megacity	Supercity
Construction index U11	0~20 points	20~40 points	40~60 points	60~80 points	80~100 points

.

The constructor identification model of opportunistic behavior studied in this paper involves 11 indicators, and the corresponding input to the neural network is an 11-dimensional vector, including the total mileage, total investment, station area, station depth, station structure form, enclosure structure form, excavation method, lining method, geological conditions, the degree of development of the construction city, and the construction index, which together constitute the input vector of the constructor identification model of opportunistic behavior.

4.1.2. Number of Neurons in the Hidden Layer

Determining the number of neurons in the hidden layer of a neural network has been an important challenge in the design process of neural networks. The choice of the number of hidden layer neurons directly affects the training efficiency and accuracy of the network. To solve this problem, empirical formulas are usually used to estimate the approximate range of the number of hidden layer neurons. However, the applicability of such empirical formulas is somewhat limited due to the fact that the characteristics of each problem are different. Therefore, it is usually necessary to determine the appropriate number of neurons through experiments in practical applications. This is done by adding different numbers of neurons to the network and observing their effects on the network training effect. If the number of neurons in the hidden layer is too small, the network may not be able to capture the complex features of the problem, leading to underfitting; if the number of neurons in the hidden layer is too large, the network may overfit the training data, leading to poor generalization. Therefore, an appropriate number of neurons needs to be found experimentally to ensure that the network can learn and generalize effectively.

In conclusion, choosing an appropriate number of neurons in the hidden layer is an important part of the neural network design process. Through a combination of empirical formulas and experiments, the optimal number of neurons can be found to improve the training efficiency and accuracy of the network and provide better performance for practical applications. The reference formula for determining the number of neuron nodes in the hidden layer is as follows:

$$n_H=\sqrt{n+m}+a$$

(8)

$$n_H=\sqrt{n+m}$$

(9)

$$n_H={\displaystyle \frac{n+m}{2}}$$

(10)

$$n_H\le \sqrt{m(n+3)}+1$$

(11)

$$n_H={\mathit{log}}_{2}m$$

(12)

In the above formula, $n_H$ denotes the number of hidden layer neurons, m and n denote the number of output layer neurons and the number of input layer neurons, respectively, and a denotes a constant between 1 and 10. After actual testing, it was found that the number of hidden layer neurons was finally determined to be 10, taking into account the error and network performance.

4.1.3. Number of Neurons in the Output Layer

According to the “Technical Specification for Risk Management of Railway Tunnel Engineering” (Q/CR 9247-2016) [22],the loss of the consequences of accidents is divided into I, II, III, IV, and V, see

Table 5. Quantification of output values for identifying opportunistic behaviors on the construction side of metro projects.

Risk Level	Level 5	Level 4	Level 3	Level 2	Level 1
	Minor	Major	Serious	Very Serious	Catastrophic
Economic loss (million CNY)	<500	500~1000	1000~5000	5000~10,000	>100,000
Personnel casualties	B ≤ 5	A ≤ 3 or 5 ≤ B ≤ 10	3 ≤ A ≤ 10 or 10 ≤ B ≤ 50	10 ≤ A ≤ 30 or 50 ≤ B ≤ 100	A > 30 or B > 100
Construction delay	C ≤ 0.01	0.01 ≤ C ≤ 0.1	0.1 < C ≤ 1	1 < C ≤ 10	C > 10
Output vector	5	4	3	2	1

* A denotes fatalities (including missing); B denotes serious injuries; C denotes delays (control construction period project, months/single incident).

for details, and according to the grades classified are transformed into the output value of BP neural network to validate the feasibility of the identification model.

This paper evaluates the impact of the opportunistic behaviors of the constructor on the safety level in a specific case. In order to identify the presence of opportunistic behaviors of the constructor in a certain set of input level indicator links and to develop corresponding prevention strategies according to different safety levels, a BP neural network-based opportunistic behavior identification model was developed. In this model, the level of danger caused by opportunistic behavior is classified as level 1, level 2, level 3, level 4, and level 5, corresponding to 1, 2, 3, 4, and 5 in the output vector.

The corresponding output vectors for different danger levels are given in Table 5. In order to determine the number of neurons in the output layer of the neural network, it is necessary to consider the results of the problem study that one wants to obtain. The aim of the research in this paper is to identify opportunistic behaviors based on the size of the danger level, so the number of neurons in the output layer is chosen to be 1 in order to achieve the goal of binary classification.

Determining the number of neurons in the output layer is an important decision in the process of neural network design, which directly affects the classification effect of the network. In this paper, a suitable number of neurons in the output layer is selected according to the research purpose and data characteristics. Through the establishment of the BP neural network model, the identification and classification of the opportunistic behavior of the constructor is realized, and the corresponding prevention strategy is provided for practical application, which has a certain application value. As a result, the BP neural network recognition model established in this paper is configured as 11 × 10 × 1 (i.e., 11 input layer neurons, 10 hidden layer neurons, and 1 output neuron). The recognition model is implemented using the BP neural network tool of MATLAB software (R2018a) for calculation.

4.2. Training and Detection of BP Neural Network Recognition Models

The training process of the BP neural network recognition model refers to the use of existing metro construction datasets for training so as to obtain a more accurate recognition model. The training process needs to take into account the quantity, quality, diversity, and other factors of the dataset; through the preprocessing, feature extraction, normalization, and other processing of the dataset, the original dataset is transformed into a format suitable for BP neural network training, and the dataset is randomly divided into a training set, a validation set, and a test set for training and verification.

The detection process of the BP neural network recognition model refers to the application of the model to the subway construction site, and through the collection and processing of real-time data, it is judged whether there is any opportunistic behavior of the current construction party. The detection process needs to take real-time accuracy, stability, and other factors into account, and the recognition ability and robustness of the model are continuously improved through feedback and correction of the detection results.

4.2.1. BP Neural Network Recognition Model Training Process

Learning, training, and testing are the three key steps in building neural networks, and they are inextricably linked. Only through continuous practice can the neural network reach the optimal state, and through training and testing, the calculation results can be continuously improved to reduce the error. According to the number of neurons in the input layer, the number of neurons in the hidden layer, and the number of neurons in the output layer, which have been established in the previous section, combined with the specific cases of opportunistic behavior of the construction side of the subway project that have been collected and arrived based on python tools (this paper will give 25 groups of related training and testing sample data in the case), and quantified their eleven index numbers, we obtained 25 training samples of recognition model values, as shown in

Table 6. Datasheet of cases of opportunistic behavior on the construction side of metro projects.

Cases Indicators	Input Value											Output Value
	U1	U2	U3	U4	U5	U6	U7	U8	U9	U10	U11
1	1	2	4	3	1	2	2	1	1	4	3	4
2	2	3	3	3	4	1	2	5	2	3	3	5
3	2	2	2	3	1	1	3	5	2	4	2	4
4	3	3	3	3	1	2	1	5	2	3	2	5
5	2	2	2	3	2	2	2	4	2	4	4	3
6	3	3	2	2	5	4	3	3	2	3	3	3
7	2	3	1	2	3	4	1	5	3	4	2	3
8	3	4	5	1	4	4	2	2	3	2	1	2
9	1	2	3	2	3	2	3	1	1	4	3	5
10	2	3	4	3	4	1	2	2	2	3	2	5
11	2	2	3	3	3	2	1	3	2	4	3	5
12	5	5	5	2	2	4	1	1	2	3	2	4
13	2	3	3	3	3	2	2	2	1	3	5	3
14	2	2	2	2	4	1	3	3	2	3	5	5
15	3	3	3	2	3	2	4	2	2	3	2	3
16	5	5	5	3	4	1	5	1	2	2	4	5
17	3	3	4	1	3	4	5	1	1	3	3	5
18	2	3	2	1	4	4	5	2	2	2	2	2
19	3	4	3	3	3	2	4	3	1	3	3	5
20	5	5	5	1	4	4	3	4	2	3	5	4
21	2	2	2	2	2	1	5	2	2	4	2	2
22	3	3	1	1	2	4	2	1	2	3	3	5
23	2	3	2	2	3	3	1	1	1	4	2	5
24	3	5	3	1	4	1	2	1	2	3	2	5
25	4	5	4	3	5	1	2	1	1	5	3	2

.

By importing the original data into the BP neural network, the convergence speed can be effectively improved, and the accuracy of the index can also be ensured. In addition, by normalizing the raw data to the interval [0~1], the S-type function can be effectively utilized, thus avoiding the saturation of the function and thus enhancing the sensitivity of the BP neural network. Equation (13) is usually utilized to normalize the raw data to the interval [0~1].

$$\overline{x_i}={\displaystyle \frac{x_i-x_{\mathrm{m}\mathrm{i}\mathrm{n}}}{x_{\mathrm{m}\mathrm{a}\mathrm{x}}-x_{\mathrm{m}\mathrm{i}\mathrm{n}}}}$$

(13)

where $x_i$ denotes the original data of the input indicator, $x_{\mathrm{m}\mathrm{i}\mathrm{n}}$ denotes the minimum value of these data, and $x_{\mathrm{m}\mathrm{a}\mathrm{x}}$ denotes the maximum value of these data. The normalized dataset is brought into MATLAB software (R2018a) for the training of the neural network model, and the specific training process is shown in Figure 2.

In order to identify the opportunistic behavior of the constructor of the metro project, this paper proposes an identification model based on the BP neural network, as shown in Figure 2. In the recognition model, a three-layer BP neural network model is used for recognition. The input layer of the model includes 11 identification indicators, and the weights and thresholds in the network are adjusted by the backpropagation algorithm to realize the identification and prediction of the opportunistic behavior of the constructor.

The output layer of the model is the loss of the consequences of the accident, and the analysis and determination of the output results are used to determine whether the construction party has opportunistic behavior. The model has been analyzed after many experimental simulation results, and it is found that the convergence speed and training accuracy of the model reach its best when the number of implied layers is set to 10. Before training the model, the dataset needs to be prepared and processed. Preprocessing, feature extraction, and normalization are performed on the dataset to transform the raw data into a format suitable for neural network training, and it is also necessary to group the dataset and randomly divide it into a training set, a validation set, and a test set for training and validation. During the training process, the BP neural network continuously learns and adjusts the data, and through the continuous adjustment and optimization of the weights and thresholds in the network, the prediction accuracy and robustness of the model are continuously improved. At the same time, the model also needs to be evaluated and validated, and the performance and generalization ability of the model are evaluated through cross-validation, error analysis, and other methods.

During the detection process, the model collects and processes real-time data from the construction site to determine whether there is opportunistic behavior on the part of the current constructor. The recognition ability and robustness of the model are continuously improved through feedback and correction of the detection results. In addition, the detection results need to be analyzed and interpreted, so as to assess and guide the behavior of the constructor.

4.2.2. BP Neural Network Recognition Model Training Results Detection

As shown in Figure 3, the number of recognition failures is the least when the number of training generations reaches the 12th time, at which time the precision of recognition meets the recognition requirements, and the specific training output results are shown in

Table 7. Recognition model training results.

	Real Number	Accurate Identification	Accuracy
Training set	17	13	76.4%
Validation set	4	4	100.0%
Test set	4	3	75.0%
Total	25	20	80.0%

.

The analysis in Figure 4 reveals that the recognition model achieves 76.4% accuracy on the training set. For both the validation set and the test set, the recognition accuracy of the model is higher, 100% and 75%, respectively, while the overall recognition success rate reaches 80%, which indicates that this BP neural network recognition model has high recognition accuracy and robustness, and can be used to identify construction enterprises with opportunistic behaviors in subway construction projects.

Although the recognition accuracy of the validation set is 100%, this result does not mean that the model’s performance in real environments will necessarily reach 100% accuracy. Therefore, when using the model for practical applications, the performance and generalization ability of the model need to be considered comprehensively, and the model needs to be evaluated and validated to ensure the accuracy and reliability of the model.

By analyzing the experimental results of the BP neural network recognition model, it can be concluded that the model has high recognition accuracy and robustness and can be used to identify the construction enterprises with opportunistic behaviors in subway construction projects so as to improve the quality and efficiency of subway construction.

5. Case Study

Combined with the actual construction project, the opportunistic behavior identification model of the subway project construction party is used to obtain the risk level of the opportunistic behavior of the case.

5.1. Project Profile

This case is analyzed for a city’s subway Line 8. The total length of the main route, in this case, is 49.896 km, all of which are underground lines. The total length of the auxiliary lines is 7.682 km, of which the length of the entrance and exit line for the vehicle base is 3.131 km, and the length of the entrance and exit line for the parking lot is 4.551 km. The length of this underground tunnel reaches 49.896 km, and the buried depth of the tunnel is between 13 and 18 m. It is constructed by the mining method. A total of 37 underground stations are set up, including 18 transfer stations. The demolition area for the Line 8 project has reached 435,700 square meters. The demolition areas for the stations and the field sections reached 143,400 m² and 209,300 m², respectively.

5.2. Analysis of Project Characteristics

Through systematic research, analysis and synthesis, the characteristics of the Metro Line 8 project are determined, including total mileage, total investment, station area, station buried depth, station structure, enclosure structure, excavation method, lining method, geological conditions, development level of construction city, and construction index. All the information is summarized to ensure that each feature can accurately reflect the actual situation of the project. By constructing the attribute feature set, accurate quantitative results can be obtained, as shown in

Table 8. Metro Line 8 project characteristics.

Characteristics	Characteristic Value	Quantized Value
Total mileage U1	49.896 km	5
Total investment U2	CNY 38,286 million	4
Station area U3	143,400 m2	5
Station buried depth U4	13~18 m	2
Station structural form U5	Mixed platform	3
Enclosure structure form U6	Pile enclosure system	3
Excavation method U7	Mining method	2
Lining method U8	Composite lining	4
Geological condition U9	Relatively complicated	2
Degree of development of construction cities U10	Megacity	3
Construction index U11	60~80 points	4

.

Because the BP neural network identification method is dynamic in identifying the indicators of the construction side of the subway project, that is, different time periods, different construction stages, and different construction methods all change, the number of case indicators in this paper is selected from 1 November 2022 to 26 December 2022. The eight-week construction process was used for research, and the scores of each factor of the case were finally selected as U1 = 5, U2 = 4, U3 = 5, U4 = 2, U5 = 3, U6 = 3, U7 = 2, U8 = 4, U9 = 2, U10 = 3, and U11 = 4.

5.3. Results

The desired output of the construction party’s opportunistic behavior identification model in this paper is to identify the presence of opportunistic behavior of the construction party in a certain set of input layer indicator links based on the size of the danger level and, if there is an opportunistic danger level, the focus is on examining and monitoring the presence of opportunistic behavior of that construction party.

Input the Metro Line 8 eigenvalues, normalize the eigenvalues, and bring the normalized dataset into MATLAB software (R2018a) for neural network model training. As shown in

Table 9. Output value level description.

Output Value	Corresponding Output Vector	Level Description
3.998	4	Major risk

, an output value of 3.998 is obtained after inputting the eigenvalues, which is closer to the output vector of 4. Consequently, the BP neural network is utilized to identify the existence of a hazard class of 4 for this builder, indicating that there is a significant hazard class associated with this project.

According to the output results of the BP neural network identification model, the risk level of opportunistic behavior occurring at the construction site of the Metro Line 8 project is considered to be higher. Upon examination of the construction site of the Metro Line 8 project, various issues were identified at this stage, including fires occurring at the worksite, irregularities in temporary power supply, chaotic management of the construction site, and injuries to workers. These findings indicate that opportunistic behavior within the Metro Line 8 project is more prevalent, thus confirming the accuracy of the identification results.

6. Conclusions

Based on the literature review, this study thoroughly explores various possible types of opportunistic behaviors and their manifestations, as well as the potential causes associated with these behaviors. Combined with the construction characteristics of the subway project, a set of opportunistic behavior identification systems in line with the construction side of the subway project is constructed, and the BP neural network algorithm is used to analyze the identification model and, through the comprehensive analysis of the results of the study, the construction side of the under-construction subway project that exists in the presence of opportunistic behavior can be clearly identified. Thus, it can improve the control ability and early warning ability of subway construction, improve the safety level of the construction site, and reduce accidents in the process of subway construction in China. It can avoid or reduce the loss as much as possible, provide a safety guarantee for promoting the construction of urban rail transit in China, and provide useful reference and guidance for the field of subway construction to ensure the efficient, safe and smooth implementation of subway projects.

This study has the following limitations. First, the opportunistic behaviors occurring in metro projects may involve the whole construction process, while this paper only focuses on the metro construction stage and examines the identification of opportunistic behaviors of the construction side of metro projects. Future research can analyze the opportunistic behaviors of subway project construction enterprises from the perspective of the whole lifecycle of subway construction. Secondly, the main research focus of this paper is to identify the opportunistic behavior of the subway project constructor, but not involved in the governance strategy of the opportunistic behavior of the subway project constructor due to the subway construction having a complex environment, a large scale of the project, construction requirements of high standards, strict control standards, safety risks, and other characteristics. The governance of opportunistic behavior is still a complex research topic, which needs to be constantly developed and improved to thus realize the good development of urban rail transit construction. Finally, we do realize that the data of this study are mainly concentrated in China, which limits the global universality of the research results to a certain extent. In future studies, we will strive to expand the geographical scope of data sources, including the collection and analysis of data from different countries and regions, in order to improve the universality and reliability of research results.