Forecast of Ground Deformation Caused by Tunnel Excavation Based on Intelligent Neural Network Model

An, Puzhen; Jia, Baoxin; Meng, Fanli; Wang, Zhixin; Wei, Hailiang; Zhang, Yinghui

doi:https://doi.org/10.1155/2022/2924093

Mobile Information Systems

On this page

Abstract Introduction Discussion Conclusions Data Availability Conflicts of Interest Acknowledgments References Copyright Related Articles

Special Issue

Intelligent Indoor Mobile Communications: Architecture, Protocols and Management

View this Special Issue

Research Article | Open Access

Volume 2022 | Article ID 2924093 | https://doi.org/10.1155/2022/2924093

Forecast of Ground Deformation Caused by Tunnel Excavation Based on Intelligent Neural Network Model

Puzhen An,^1,2Baoxin Jia ,¹Fanli Meng,²Zhixin Wang,²Hailiang Wei,²and Yinghui Zhang²

Academic Editor: Fusheng Zhu

Received11 Mar 2022

Revised09 Apr 2022

Accepted28 Apr 2022

Published21 May 2022

Abstract

With the development of science and technology, the demand for traffic has increased, and the requirements for tunnel excavation have become more and more stringent. Tunnel excavation is an important traffic construction engineering technology. Due to the influence of many factors in the excavation process, surface settlement or deformation will inevitably occur, so its deformation must be predicted in real time to prevent safety accidents and property losses. The previous numerical methods and neural network methods cannot accurately predict in real time, and the intelligent neural network model can more accurately predict the deformation of the ground because of the characteristics of adaptive organizational learning according to different situations and different environments. This article aims to study and design an intelligent neural network model to predict and calculate the amount of ground deformation caused by the tunnel excavation process. An intelligent neural network model with more accurate prediction is proposed, and simulation experiments are carried out on tunnel excavation of different terrains, and the accuracy of the model for predicting the deformation amount is calculated. The experimental results show that the prediction error range of the model is 10 times smaller than that of the traditional neural network. The prediction accuracy of this model is above 95%, and the volatility rate of prediction accuracy is lower than 11%, while the volatility rate of traditional prediction accuracy is even more than 365%. The intelligent neural network model can effectively predict the deformation of tunnel excavation.

1. Introduction

1.1. Background and Significance

With the continuous development of world science and technology and the continuous improvement of people’s living standards, the demand for transportation is increasing. As an important transportation construction technology, tunnel engineering plays an increasingly important role in transportation. In the process of tunnel excavation, it is inevitable that there will be settlement and deformation problems on the surface of the tunnel. In order to avoid safety accidents and property losses caused by such settlement and deformation, it is necessary to predict the amount of settlement or deformation more accurately before tunnel excavation so as to timely understand the potential safety hazards during the excavation process, and take corresponding preventive measures to ensure the safety of the tunnel excavation process. There are many factors in the tunnel excavation process, which may produce different settlement and deformation, such as noise, signal interference, and air environment. Therefore, it is necessary to predict the possible settlement and deformation before excavation and in real time during the excavation. If it can accurately predict the amount of settlement or deformation that may occur during excavation, this will be a major technological breakthrough. Neural network is an increasingly widely used data analysis tool. Neural network has been developed for many years. At present, intelligent neural network will be the key research area of neural network. Intelligent neural network is more efficient and can adapt to various complex situations, self-organized learning, and other characteristics. Using intelligent neural network to predict the deformation caused by tunnel excavation will be a huge challenge (mainly refers to the establishment of intelligent neural network model) and a great project.

1.2. Related Work

With regard to the prediction of surface deformation caused by tunnel excavation, more and more scholars have conducted in-depth research. Among them, Lin et al. studied the deformation behavior of the existing tunnel based on numerical simulation (rely on electronic computer, combined with the concept of finite element or finite volume, through numerical calculation and image display methods, to solve engineering problems, physical problems, and analysis methods of various problems in nature), studied the undercrossing oblique crossing of the newly built double tunnel, and analyzed the change law of the foundation settlement trough and the Earth pressure acting on the existing tunnel during the construction of the shield tunnel, obtained the lateral deformation, internal force, and torsion behavior of the existing tunnel caused by the excavation of the new tunnel. In addition, he also conducted parameter studies. The results show that due to the oblique crossing, the lateral deformation and internal force of the existing tunnel show obvious asymmetric characteristics, and the existing tunnels on both sides of the new tunnel show irreversible partial torsional deformation after shielding [1]. In order to reveal the stratum displacement and stress distribution caused by the excavation of shallow shield tunnels in composite strata, the equivalent layer method (method of equivalent multilayer complex problem to one-layer or two-layer simple problem, the essence of equivalent layer thinking is to transform a more complex practical problem into a simple familiar problem under the same effect so as to highlight the main factors, grasp its essence, and find out the rules. Therefore, when applying the equivalent method, it often substitutes simpler factors for more complex factors to simplify the problem and make it easier to solve and is used to simplify the problem to stratum displacement and stress distribution. Helin et al. used the modified Ronaldson formula to derive the expressions of formation displacement, strain component, and stress component, and established a displacement field distribution prediction model. Based on the theoretical prediction model, the predicted value and the measured value are compared. The research results show that during the tunnel construction process, the upper hard and lower soft stratum has displacement and stress diffusion, and the upper soft and lower hard stratum has displacement and stress concentration [2]. In geological engineering, mechanical problems such as overloading on the ground surface and shallow tunnel excavation are often encountered. In order to eliminate the infinite singularity, the far-field displacement obtained by the modified displacement solution of the complex variable method is logically consistent with the reality. Lin et al. proposed two basic cases of a single overload load acting on the ground surface and a single tunnel excavation to verify the modified displacement solution, and a good agreement was observed. By comparing with the original solution, the main advantage of the proposed solution is that it eliminates the singularity of the infinite displacement and makes the far-field displacement logically conform to the field observation [3]. Many new excavations are planned and constructed near or parallel to existing tunnels. Adjacent excavations may significantly affect the stress and deformation of existing tunnels. In order to evaluate the deformation and stress transfer mechanism, Sun et al. use an advanced nonlinear constitutive model (called a reduced plasticity model, refers to the constitutive model where the stress-strain relationship curve is parallel to the abscissa axis) to conduct a reverse analysis of the fully-instrumented centrifuge model test. The research results show that the tunnel is elongated vertically and compressed horizontally. The tunnel on the side of the basement is elongated along the junction between the right shoulder and the left knee, and compressed between the left shoulder and the right knee [4]. In recent years, with the development of underground construction, multitunnel engineering has become a concern of people. Because the close interaction between the tunnels will cause additional deformation of the ground and surrounding structures, and even cause serious damage to the buildings on the surface. Taking into account the influence of various factors such as the anisotropy of soft clay, Jin et al. uses numerical methods to effectively predict these tunnel displacement and soil deformation problems. The researchers analyzed the results including ground settlement, lining force and bending moment, and tunnel convergence. This research can provide references for the design and construction of multiple tunnels [5]. The Himalayan fold zone is full of geological wonders, and “hybrid” is one of them, which brings difficulties to tunnel construction. The mélange is characterized by a chaotic and heterogeneous geological mixture of strong blocks (scale independent) and weaker shear fine-grained matrices. The heterogeneity of tectonic mixing leads to stress concentration in the rock block, and there is also relatively high deformation in the matrix. The new Austrian Tunneling Method (NATM) adopted by Dhang proved to be a useful tunneling tool [6]. Due to the massive excavation and unloading of the ground surface, a large number of cracks and fractures appeared in the segments of the underlying shield tunnel. Dai uses a finite element simulation method, from the aspects of stratum conditions, Earth excavation, construction, etc., and the changes in the displacement and internal force of the underlying tunnel during the process of large-scale excavation and unloading on the surface, the sequence, the assembly quality of the segments, and the locations of segment cracks and fractures are analyzed. The results show that the overall uplift deformation of the underlying shield tunnel shows a law of decreasing horizontal diameter and increasing vertical diameter. The bending moment of the tunnel vault and the arch bottom is reduced, the axial force remains unchanged, and the internal tension is replaced by the lateral tension of the local area of the tunnel vault. The vault joint changed from an inner opening to an outer opening and internal extrusion [7]. However, these studies have certain shortcomings. Either it is difficult to implement, or the calculation is not accurate enough, or the cost is too high, so it is difficult to pass. Compared with the intelligent neural network, its cost is relatively low, the calculation accuracy is high, and it is easier to implement. Therefore, the research of this paper has practical value and significance.

1.3. Innovation

(1) Utilizing the self-organizing learning characteristics of intelligent neural network, it can adapt to a variety of different complex environments, and it can solve the complex and changeable problems of the tunnel excavation environment. (2) The tunnel excavation process requires real-time acquisition of surrounding environment information for calculation to predict the resulting settlement or deformation. The intelligent neural network model can detect and calculate this in real time. (3) Carrying out tunnel excavation simulation experiment to test the intelligent neural network model, and use data to prove the feasibility and accuracy of the model. (4) Comparing the traditional tunnel excavation deformation prediction method with the intelligent neural model prediction method in this paper can better reflect the advantages of the model’s predictions, such as more accurate, higher stability, and reliability.

2. Tunnel Excavation Deformation and Intelligent Neural Network

2.1. Deformation Caused by Tunnel Excavation

2.1.1. Tunnel excavation process

The tunnel excavation process must be rigorous and orderly. The topography is surveyed before excavation, and after the survey is completed and all preparations and necessary equipment are gathered, the drilling can be carried out. According to different terrains and topography, the drilling machines or drilling methods used are different. After drilling, fill the medicine and use the detonator to detonate. After the blasting is completed, the ventilation treatment must be carried out first, and the ventilation process must be continued, and then, the tunnel excavator is installed, the model is shown in Figure 1. Figure 2 shows the model of the cutter head of the tunnel excavator. After the installation is completed, excavation and ballasting are carried out, then concrete spraying for the first time, followed by excavation, ballasting (the operation of transporting the blasted ballast out of the tunnel and discarding it) and concrete spraying (construction method of spraying and pouring fine stone concrete with pressure spray gun), and so on repeatedly. During this period, it must continue to detect the terrain to prevent accidents [8], the process can be shown in Figure 3.

2.1.2. Surface Deformation

Tunnel excavation destroyed the original equilibrium state of the soil. In addition, the change of the water level caused the original stress state in the soil to change, resulting in surface deformation. The main reasons for the deformation of the tunnel surface include (1) the influence of the characteristics of the stratum soil, (2) the influence of groundwater, (3) release of rock stress, (4) the superimposed influence of the tunnel effect, (5) construction method, (6) influence of excavation footage, and (7) initial support stiffness [9].

The excavation process of a shield excavator is taken as an example. The main torsion force comes from the advancement and friction of the cutter head. The cutter head model is shown in Figure 2. To analyze the torque of the cutter head, the friction torque of the cutter face can be shown bywhere r represents the radius of the excavator blade, α₁ is the adjustment parameter, α₂ is the pressure coefficient, λ is the soil volume, ƒ is the friction coefficient, u is the cutting rate of the cutter head, and S is the tunnel depth, P₁ is the initial lateral pressure, and P₂ is the additional squeezing force generated by excavation [10].

The friction torque of the tool surface attachment can be expressed bywhere W is the width of the cutter head.

The torque T_c required for cutter head cutting can be shown bywhere θ_c is the shear strength of the soil, ₁ is the forward speed of the excavator, and ₂ is the rotation speed of the cutter head.

The rotating torque of the cutter head can be expressed bywhere d₁ is the diameter of the roller, ƒ_g is the coefficient of rolling friction.

The torque of the cutter head weight can be expressed bywhere m₀ is the weight of the cutter head and d₀ is the contact radius.

Then, the total torque formula of the cutter head in the operating system is as follows:

Then, the ground deformation can be calculated by the differential method, that is, the arbitrary microelement stress of the shield excavation surface is

The amount of deformation generated by the shield excavator in the three-dimensional direction during the excavation process is represented by I₁, J₁, and K₁, and the amount of deformation I in the horizontal and transverse direction is

Deformation in horizontal and longitudinal direction J is as follows:

Vertical deformation K is as follows:

In the process of pouring cement, its pouring amount and pouring pressure will affect the amount of terrain deformation. Similarly, use the differential method to express any microelement:

The amount of deformation produced in the horizontal and vertical directions is represented by I₂, J₂, and K₂, and the total horizontal and lateral deformation produced during the construction process is as shown in

The total horizontal and vertical deformation is

The total deformation in the vertical direction is

2.1.3. The Surface Subsidence or Uplift

The excavation of the tunnel causes the surrounding rock of the excavation surface to squeeze out, or the surrounding rock to relax due to the collapse. After excavation, the surrounding rock and the support, the gap between the surrounding rock and the lining were consolidated due to the drop of the groundwater level, which caused the support in the weak surrounding rock to sink, causing the ground to sink. When the surface deformation reaches a certain level, it will cause the surface to settle or uplift [11].

2.2. Common Methods for Predicting Deformation

2.2.1. Empirical Method

Using previous construction experience, the measurement data are replaced with a formula, and the necessary mathematical operations are performed on the formula to study the relationship between the respective variables and functions. For example, the Peck prediction formula (15) for the distribution of surface subsidence is as follows:where Y(d) is the surface settlement value at the position d away from the central axis of the tunnel, is the introduction parameter of ground settlement width, Y_max is the maximum settlement value of the center line of the tunnel, and Y_max satisfies the following formula:

2.2.2. Actual Measurement Method

Hence the name implies, it is to compare the actual measurement data with the requirements of construction specifications and quality standards and allowable deviation values to determine whether the construction safety and quality meet the requirements [12].

2.2.3. Numerical Analysis Method

The method of numerical simulation to study the mechanical properties of rock and soil is mainly based on the two aspects of continuous medium mechanics and discontinuous medium mechanics. The basic principle is based on the most basic Newton’s second law of motion. Based on the rigid body motion equation of each element, an explicit equation set describing the motion of the entire system is established, and then, iterative calculations are performed by the dynamic relaxation method (i.e., dynamic analysis method), analysis over time, and the dynamic relaxation method is also introduced in LS-DYNA3D to solve the implicit problem. It adopts the calculation idea of backward time step iteration and adds artificial damping to approximately solve the static problem. The larger the artificial damping is, the faster the convergence will be. However, the size of the artificial damping cannot exceed the critical damping; otherwise, the calculation time will be too long (or very small). It is mainly used to explicitly solve the static load of the previous model. The basic motion equation is shown as follows:where Y represents the load function, m represents the displacement, e represents the elastic parameter, f represents the friction coefficient, and represents the mass.

Using this equation to solve, combined with data analysis, can understand the changes of various mechanical parameters of terrain movement.

2.2.4. Neural Network Method

The deformation of the rock and soil around the tunnel and the settlement and deformation of the ground above the tunnel caused by tunnel excavation is a very complex nonlinear dynamic system, and there are many factors involved in ground settlement. Neural network has the advantages of parallel distributed processing, nonlinear processing, self-learning, self-organization and self-adaptive ability, fast information processing speed, strong calculation ability, and high fault tolerance, so it can solve this problem well. Figure 4 is a model diagram of a neural network. The neural network is divided into an input layer, a competition layer, and an output layer. The input information is clustered and output to the output layer under the self-learning process of neurons in the competition layer. The process of using neural networks to solve problems is (1) determining the problem and the way of data expression; (2) establishing the neural network model; (3) setting the network parameters; (4) determining the training mode; (5) test samples; (6) test results [13].

Among the above methods, the concept of the empirical formula method is simple. Although the calculation is convenient, the empirical formula is an empirical method and lacks theoretical foundation. The physical meaning of the parameters involved is not clear, and it is not convenient for popularization and application. Although the actual measurement method can comprehensively and monitor in real time the construction process, it has strong pertinence, the monitoring results can be directly fed back to the engineering practice, the content of on-site monitoring is often scattered, the workload is relatively large, and the cost is relatively high. Numerical simulation analysis methods have high flexibility, convenient modeling, and can better reflect the actual geotechnical conditions, but the results are some qualitative conclusions, which are difficult to meet the quantitative requirements in specific engineering applications. The neural network method has a powerful nonlinear mapping function. It can build a network “model” based on the existing data to predict the future development trend and provide a reference for the next step of decision-making and control. However, the traditional neural network method requires too much sample data and too many calculations, and the credibility of the obtained results still needs a lot of verification. The intelligent neural network method proposed in this paper will improve the stability and reliability of the results, and the amount of calculation is much less than that of the traditional neural network [14].

2.3. Theory of Intelligent Neural Network

2.3.1. The Origin of Neural Networks

The Kohonen Network proposed by Professor Teuvo Kohonen of the University of Helsinki in Finland is a self-organizing competitive neural network. Self-organization means that the network is unsupervised and can learn independently, and adjust the element value through the self-organizing function relationship according to the characteristics of the environment so that the neural network can automatically distinguish and aggregate and classify. In this form of expression, neurons will match a specific input form and enhance the impression, resulting in sensitivity. Therefore, the input form can be divided into different clusters through self-organizing training and learning, and each cluster has different response characteristics to the input form, so the neuron can become a detector in a certain environment [15].

2.3.2. Types of Neural Networks

After years of research and development, the current neural network can be roughly divided into (1) Feedforward neural network: the most common type of neural network in practical applications. The first layer is the input, and the last layer is the output. If there are multiple hidden layers, we call it a “deep” neural network. Calculating a series of fusions that change the similarity of samples, and the activity of neurons in each layer is a nonlinear function of the activity of the previous layer. (2) Circulating neural network: the current output of a sequence is also related to the previous output. The specific manifestation is that the network will memorize the previous information and apply it to the calculation of the current output. That is, the nodes between the hidden layers are no longer unconnected but connected, and the input of the hidden layer includes not only the output of the input layer but also the output of the hidden layer at the previous moment. (3) Symmetrically connected network: symmetrically connected network is a bit like a cyclic network, but the connections between units are symmetrical (they have the same weight in both directions). Compared to cyclic networks, symmetrically connected networks are easier to analyze. There are more restrictions in this network because they obey the law of energy function. A symmetrically connected network without hidden units is called a “Hopfield network,” and a symmetrically connected network with hidden units is called a Boltzmann machine. However, when dealing with some big problems, a single neural network is often not enough. Therefore, a combination of multiple neural networks is currently combined into a group of neural networks. According to the structure and combination model of neural network, it can be divided into single neural network group and combined neural network group. The intelligent neural network group studied in this paper is a combined neural network group. Figure 5 shows the structure of two types of intelligent neural network models. Figure 5(a) is an intelligent neural network composed of DN neural network (a deconvolutional neural network, also called an inverse graph network, is the inverse process of a convolutional neural network) and HN neural network (i.e., the Hopfield network mentioned earlier), and Figure 5(b) is an intelligent neural network composed of the RNN neural network and GAN neural network [16].

(a)

(b)

The intelligent neural network model can be expressed by the function formula:where F is the output of the intelligent neural network model and i_n is the output value of the nth neural network.

Setting the neural network prediction deviation value to D_n, then,where h is the predicted trend parameter value, L is the predicted logical value, and is the neural network feature dimension value.

The improved output accuracy A_n of the intelligent neural network model is as follows:where , .

3. Tunnel Mining Simulation Experiment Based on Intelligent Neural Network

3.1. Design of Intelligent Neural Network Model

3.1.1. Selection of Model

Compared with the other two common neural networks, the feedforward neural network has a wider application range, higher flexibility, and can better meet the calculation requirements of this experiment. Therefore, the intelligent neural network model studied in this paper is based on a CNN neural network (a convolutional neural network is usually used to deal with highly nonlinear abstract classification problems). A CNN (convolutional neural network) is a kind of feedforward neural network with convolution calculation and deep structure, which is one of the representative algorithms of deep learning. Convolutional neural networks have the ability to learn representations and can perform translation-invariant classification of input information according to its hierarchical structure and RNN neural network as prototypes for improved combination. The improved neural network has more detailed calculations and higher accuracy. The model diagram is shown in Figure 6 [17].

3.1.2. Performance Analysis of the Model

The neural network model has self-organized learning and self-adaptive ability can store a large amount of information and perform large-scale processing and has strong fault tolerance. The learning rate is faster than the traditional neural network, the network structure is layered clearly, and the experimental prediction effect is more [18].

3.1.3. Acquisition and Training Methods of Data Samples

The data in this article are derived from the actual values of the simulation experiment, and the actual values of the simulation experiment fluctuate randomly within a certain range according to the actual situation. The training method is based on the continuous monitoring of one of the deformations of the tunnel to obtain the deformation values i₁, i₂,..., i_n that change over time, then according to its adaptive learning, predict the deformation values I₁, I₂,..., I_n for a period of time later, and compare the predicted value with the actual value, and adjust the error adaptively, after many studies until the required high accuracy is achieved. Each neural network consists of three layers. The first neural network that has been trained with high accuracy is used to calculate the predicted value, and then, the predicted result is compared with the actual result and fed back to the second neural network for training and recalculation, and finally, the obtained prediction result is used as the test result. The training method can be shown in Table 1 [19]:

3.1.4. Test Method

According to the actual parameters of the simulation experiment as the input value, the calculation is performed by using the trained super intelligent neural network algorithm to obtain the accurate prediction, and the calculation result is the predicted deformation value. According to the predicted deformation value, the experimental parameters are modified to verify the accuracy of the predicted result, and calculate the error and accuracy between the predicted result and the actual result [20].

3.2. Design of Simulated Tunnel Excavation Experiment

This paper simulates different terrains and environments, taking real-time data of settlement monitoring points as samples, and the obtained deformation value is the output value. The relevant data of several main factors affecting settlement are obtained below. Among them, the rock grade is C, the temperature is T, and the groundwater condition is W, the construction quality is Q, the monitored settlement value is I, and the tunnel excavation depth to diameter ratio is R. The training data obtained are shown in Table 2.

According to the adaptive and self-organizing learning capabilities of the intelligent neural network, the training data in Table 2 are used for multiple training so that it can generate sensitive feature relationships on the data and accurately predict the output value. The training result of the training sample is the final value obtained through the training of the multilayer neural network. In order to obtain a more accurate prediction effect, this experiment carried out two sets of training on the samples. The first set of training results is represented by I1, the second set of training results is represented by I2, and the actual value of the sample is represented by I. Fitting the two sets of training results with the actual values of the sample can see whether the efficiency and prediction accuracy of the neural network model’s self-organized learning can meet the requirements. The two sets of fitting results are shown in Figure 7 [21].

(a)

(b)

Among them, the second set of fitting is performed after the first set of fitting. As can be seen from Figure 7(a), the fitting result is relatively acceptable accuracy, but the range error is still not small. After the second set of fitting is performed, it can be seen from Figure 7(b) that the fitting effect is significantly better, and it fits almost perfectly. This also shows that the intelligent neural network has a strong self-learning ability and a deep sensitivity to data. The calculated second set of fitting results has sufficient accuracy for sample testing [22].

Using the second set of fitted intelligent neural networks to test the actual measured data of the simulated tunnel excavation, and predict the output value, that is, the amount of deformation I, according to the aforementioned parameters, as shown in Table 3.

3.3. Predicted Results

The intelligent neural network selected according to the above test samples is tested and compared with the traditional neural network, and the intelligent neural network prediction value is I_a, the traditional neural network prediction result is I_b, and the actual result is I_c. The comparison is shown in Table 4 [23].

From this, the prediction error values of the two neural networks can be obtained, as shown in Figure 8.

(a)

(b)

It can be seen from the comparison of Figure 8 that the graphical prediction value of the intelligent neural network is almost consistent with the actual value, while the prediction error value of the traditional neural network is still relatively large, and some are even larger than the predicted value, such as 4 test values. The prediction result of the intelligent neural network is always a little smaller than the actual value, which is very close to the actual result. From Table 4, the prediction error (absolute value) and output accuracy A_n of the two neural networks can be obtained, as shown in Figure 9.

(a)

(b)

It can be clearly seen from Figure 9(a) that the prediction error value of the intelligent neural network is significantly lower than that of the traditional neural network, and its prediction error is stable at about 0.1. The error value of the traditional neural network is between 0.1 and 1, and its error range is about 10 times that of the intelligent neural network. It can be seen from Figure 9(b) that the accuracy of the intelligent neural network model is always above 90%, while the accuracy of the traditional neural network model fluctuates in a relatively large range. Although there is also relatively high accuracy, its accuracy is always lower than the prediction accuracy of the intelligent neural network [24].

In order to understand the actual reliability of the prediction results of the intelligent neural network system, this paper calculates the accuracy of the two prediction results before and after the two types of neural networks and obtains the fluctuation rate of the calculation accuracy, accuracy fluctuation rate = (precision difference between the previous two times/precision after one time) × 100%, and take the absolute value. The volatility of accuracy here is calculated under the premise that the accuracy meets the requirements. The greater the volatility of accuracy, the greater the forecast error and the lower the forecast reliability. That is, the reliability of the prediction result is inversely proportional to its volatility, and the final calculation result is shown in Figure 10 [25].

It can be seen from Figure 10 that the test accuracy of the intelligent neural network has very little volatility, no more than 11%, and its prediction results are very stable. However, the prediction accuracy of traditional neural networks is relatively volatile, with the highest volatility exceeding 365%, and its stability is obviously not as good as that of intelligent neural networks. It can be seen that, whether it is a comparison of the accuracy of the prediction results or the comparison of the stability of the prediction, the intelligent neural network has obvious advantages. Using intelligent neural network to predict the deformation of tunnel excavation will effectively improve the accuracy of the prediction results [26].

4. Discussion

The intelligent neural network studied in this paper can solve the shortcomings of traditional neural network algorithms to a certain extent and can meet the accuracy requirements (required accuracy cannot be lower than 90%, and most construction projects use cement as the main building material. The total allowable error in the construction of concrete columns, beams, and walls is about 10 to 30 mm. The inclination of the axis of high-rise buildings is required to be 1/2000 to 1/1000. The total error of steel structure construction varies with the construction method, and the allowable error is 1–8 mm. The construction error of the upper stone is allowed to reach 10 cm of tunnel deformation monitoring, predictive analysis and safety evaluation. However, because the article involves a lot of content and author’s level is limited, there are still many areas that need to be improved [27].(1)The factors of surface deformation caused by tunnel excavation are complex, which involve many qualitative factors. In terms of quantification of qualitative influencing factors, only rare data and experience cannot accurately express the magnitude of its influence. There is a need for clearer guidance and in-depth research in the quantification of qualitative factors.(2)The tunnel excavation process must ensure the stability of the excavation surface, and what kind of engineering measures should be taken to reduce the impact of shield construction on the surrounding environment is not deep enough.(3)When selecting the single model of the combined model, not every single model will improve its output after the combination. How to select a reasonable single model to achieve the best combination effect will require further research [28, 29].(4)There are certain limitations to relying only on real-time monitoring data to test the predictive deformation evaluation parameters of tunnel excavation, and there is no comprehensive use of knowledge of multiple disciplines.

The actual process of tunnel excavation is more complicated than expected in this article, and the terrain will change with the progress of the excavation process. In addition, the construction process has the influence of natural weather, such as drought and rain, and the stress of the terrain structure will change. Due to the limited knowledge of the author, further study and exploration are needed [30, 31].

5. Conclusions

This article first investigates the current research status of many scholars on tunnel excavation prediction deformation in the related work part and understands the commonly used prediction methods among them. Combined with the research of intelligent neural network, it is proposed to use the characteristics of intelligent neural network to predict the deformation of tunnel excavation. Also, it is proposed to further study the process and principle of tunnel excavation, understand the reason of deformation during excavation, and use mechanical analysis to calculate the deformation amount. Then, it analyzes the commonly used methods of predicting deformation, including empirical method, actual measurement method, numerical analysis method, neural network method, and compares the characteristics of these commonly used methods. It is concluded that the intelligent neural network model obtained by using multiple neural network combinations can obtain more accurate and efficient predictions. Then, in-depth theoretical research on the intelligent neural network is conducted and the intelligent neural network model is designed, and then the simulation experiment of tunnel excavation is carried out, and high self-learning and self-organization ability of the intelligent neural network is used to conduct data sample training. The trained intelligent neural network has a high degree of data sensitivity, and its prediction accuracy is high enough. Then, the trained model is used for sample testing, and the prediction results obtained are compared with the actual results. The results show that the prediction error range of the model is below 0.1, while the maximum error of the traditional neural network is above 1, and its error range is reduced by 10 times. The prediction accuracy of this model is above 95%, and the volatility rate of prediction accuracy is less than 11%, while the volatility rate of traditional prediction accuracy is even as high as 3.65 times. It can be clearly seen that the intelligent neural network model can more effectively predict the deformation of tunnel excavation.

Data Availability

No data were used to support this study.

Conflicts of Interest

The authors declare that there are no conflicts of interest with any financial organizations regarding the material reported in this manuscript.

Acknowledgments

This study was supported by the Liaoning Revitalization Talents Program (XLYC2007163): Research on classification method and occurrence criteria of microseismic signals of compound dynamic disaster; discipline innovation team of Liaoning Technical University (LNTU20TD08): Underground engineering disaster mechanism and control innovation team.

References

X. T. Lin, R. P. Chen, H. N. Wu, and H. Z. Cheng, “Deformation behaviors of existing tunnels caused by shield tunneling undercrossing with oblique angle,” Tunnelling and Underground Space Technology, vol. 89, pp. 78–90, 2019.
View at: Publisher Site | Google Scholar
F. U. Helin, J. Zhang, and Z. Huang, “Prediction and analysis of ground displacement and stress caused by the excavation of a shallow-buried shield tunnel in a compound stratum,” Modern Tunnelling Technology, vol. 54, no. 4, pp. 97–106, 2017.
View at: Google Scholar
L.-b. Lin, F.-q. Chen, and D.-y. Li, “Modified complex variable method for displacement induced by surcharge loads and shallow tunnel excavation,” Journal of Engineering Mathematics, vol. 123, no. 1, pp. 1–18, 2020.
View at: Publisher Site | Google Scholar
H. S. Sun, L. W. Wang, Y. D. Chen, Y. Dong, J. H. Zhang, and W. B. Sun, “Nonlinear analysis of tunnel deformation and mechanisms of stress transfer caused by basement excavation,” Soil Mechanics and Foundation Engineering, vol. 55, no. 5, pp. 298–304, 2018.
View at: Publisher Site | Google Scholar
Y. F. Jin, B. Q. Zhu, Z. Y. Yin, and D.-M. Zhang, “Three-dimensional numerical analysis of the interaction of two crossing tunnels in soft clay,” Underground Space, vol. 4, no. 4, pp. 310–327, 2019.
View at: Publisher Site | Google Scholar
P. C. Dhang, “Tunneling in lesser Himalaya, Jammu and Kashmir, India with special emphasis on tectonic mélange,” Journal of the Geological Society of India, vol. 88, no. 5, pp. 593–602, 2016.
View at: Publisher Site | Google Scholar
Z. Dai, “Study on mechanism of segment crack and fracture of underlying shield tunnel caused by extensive excavation and unloading of ground surface,” China Railway Science, vol. 38, no. 4, pp. 62–69, 2017.
View at: Google Scholar
M.-l. Hou, W.-t. Ding, X.-g. Wang, K.-q. Liu, and L. Chen, “Analysis of the soil loss and settlement of silty clay with different subclassifications,” Geotechnical & Geological Engineering, vol. 38, no. 6, pp. 5819–5838, 2020.
View at: Publisher Site | Google Scholar
Z. C. Li, G. H. Wang, J. W. Hao et al., “The influence of construction scheme of asymmetric three-cabin utility tunnelling on the surface settlement behaviour,” KSCE Journal of Civil Engineering, vol. 25, no. 9, pp. 3568–3582, 2021.
View at: Publisher Site | Google Scholar
Z. Liu and Y. Wang, “Deformation control technology for tunnel excavation of interaction of the phyllite and slate after earthquake,” Journal of Railway Engineering Society, vol. 36, no. 1, pp. 70–73, 2019.
View at: Google Scholar
Z. Zhang, M. Zhang, Y. Jiang, Q. Bai, and Q. Zhao, “Analytical prediction for ground movements and liner internal forces induced by shallow tunnels considering non-uniform convergence pattern and ground-liner interaction mechanism-ScienceDirect,” Soils and Foundations, vol. 57, no. 2, pp. 211–226, 2017.
View at: Publisher Site | Google Scholar
J. Zhang, L. Huang, T. Peng, H. Wang, Y. Zhang, and L. Guo, “Model testing on failure mechanism of tunnel face in sandy cobble stratum,” Arabian Journal for Science and Engineering, vol. 45, no. 5, pp. 4077–4089, 2020.
View at: Publisher Site | Google Scholar
H. M. Shahin, T. Nakai, K. Ishii, T. Iwata, and S. Kuroi, “Investigation of influence of tunneling on existing building and tunnel: model tests and numerical simulations,” Acta Geotechnica, vol. 11, no. 3, pp. 679–692, 2016.
View at: Publisher Site | Google Scholar
Z. Wang, W. Yao, Y. Cai, B. Xu, Y. Fu, and G. Wei, “Analysis of ground surface settlement induced by the construction of a large-diameter shallow-buried twin-tunnel in soft ground,” Tunnelling and Underground Space Technology, vol. 83, pp. 520–532, 2019.
View at: Publisher Site | Google Scholar
H. Jiang, M. Yin, and Y. Jiang, “Structure deformation caused by shield tunnel excavation above existing tunnels of shenzhen metro,” Modern Tunnelling Technology, vol. 55, no. 1, pp. 194–202, 2018.
View at: Google Scholar
O. N. Melnikova and K. V. Pokazeev, “Estimation of water-surface deformation by vortices in a viscous horizontally decelerating air flow,” Moscow University Physics Bulletin, vol. 72, no. 3, pp. 326–328, 2017.
View at: Publisher Site | Google Scholar
N. Zhang, Y. Ma, and Q. Zhang, “Prediction of sea ice evolution in Liaodong Bay based on a back-propagation neural network model,” Cold Regions Science and Technology, vol. 145, pp. 65–75, 2018.
View at: Publisher Site | Google Scholar
Z. Huang and Y. Chen, “An improving SRL model with word sense information using an improved synergetic neural network model,” Journal of Intelligent and Fuzzy Systems, vol. 31, no. 3, pp. 1469–1480, 2016.
View at: Publisher Site | Google Scholar
H. Inokuchi and T. Akiyama, “Verification of car following model using fuzzy neural network,” Journal of Japan Society for Fuzzy Theory and Intelligent Informatics, vol. 28, no. 5, pp. 791–800, 2016.
View at: Publisher Site | Google Scholar
M. Zhou, Y. Wang, R. Xu et al., “Feed-forward control for magnetic shape memory alloy actuators based on the radial basis function neural network model,” Journal of Applied Biomaterials & Functional Materials, vol. 15, pp. 25–30, 2017.
View at: Publisher Site | Google Scholar
M. S. Alkhasawneh and L. T. Tay, “A hybrid intelligent system integrating the cascade forward neural network with elman neural network,” Arabian Journal for Science and Engineering, vol. 43, no. 12, pp. 6737–6749, 2018.
View at: Publisher Site | Google Scholar
M. B. Ahmed and A. A Boudhir, [Lecture Notes in Networks and Systems]Innovations in Smart Cities and Applications Volume 37||An Intelligent Model for Enterprise Resource Planning Selection Based on BP Neural Network, Springer International Publishing, New York, NY, USA, 2018.
C. Y. Lin, N. Xue, and D. Zhao, A Convolution BiLSTM Neural Network Model for Chinese Event Extraction, Springer International Publishing, New York, NY, USA, 2016.
M. Fakhri and R. Shahni Dezfoulian, “Pavement structural evaluation based on roughness and surface distress survey using neural network model,” Construction and Building Materials, vol. 204, pp. 768–780, 2019.
View at: Publisher Site | Google Scholar
L. Haibo, L. Zhenglin, and L. Guoliang, “Neural network prediction model to achieve intelligent control of unbalanced drilling’s underpressure value,” Multimedia Tools and Applications, vol. 78, no. 21, pp. 29823–29851, 2019.
View at: Publisher Site | Google Scholar
J.-H. Shin, H.-B. Jun, and J.-G. Kim, “Dynamic control of intelligent parking guidance using neural network predictive control,” Computers & Industrial Engineering, vol. 120, pp. 15–30, 2018.
View at: Publisher Site | Google Scholar
M. Stoffel, F. Bamer, and B. Markert, “Artificial neural networks and intelligent finite elements in non-linear structural mechanics,” Thin-Walled Structures, vol. 131, pp. 102–106, 2018.
View at: Publisher Site | Google Scholar
G. A. Kilin, B. V. Kavalerov, and A. I. Suslov, “Set of neural network models for intelligent control of low- and medium-capacity gas-turbine power plants,” Russian Electrical Engineering, vol. 91, no. 11, pp. 659–664, 2020.
View at: Publisher Site | Google Scholar
A. Ahmed and M. Khalid, “An intelligent framework for short-term multi-step wind speed forecasting based on Functional Networks,” Applied Energy, vol. 225, pp. 902–911, 2018.
View at: Publisher Site | Google Scholar
B. Liao and Q. Xiang, “Robustness analyses and optimal sampling gap of recurrent neural network for dynamic matrix pseudoinversion,” Journal of Advanced Computational Intelligence and Intelligent Informatics, vol. 21, no. 5, pp. 778–784, 2017.
View at: Publisher Site | Google Scholar
H. Yin, H. Zhang, and R. Zhang, “Measurement and simulation analysis of vertical cross longitudinal settlement between turning shield and foundation pit,” Advances in Civil Engineering, vol. 2021, Article ID 6698268, 14 pages, 2021.
View at: Publisher Site | Google Scholar

Copyright

Copyright © 2022 Puzhen An 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.

PDF Download Citation

Download other formats

Order printed copies