Shield Tunneling Parameter Matching Model and UI Interface

In order to improve the accuracy of shield tunneling parameter matching under the limited data, the matching model based on support vector machine (SVM) and exponential adjustment inertia weight immune particle swarm optimization (EAIW-IPSO) is proposed. The nonlinear relationship model between the tunneling parameters and the ground settlement is constructed by SVM and trained with the actual engineering sample data. Based on the trained model, EAIW-IPSO is used to optimize the tunneling parameters. At the same time, UI interface was developed based on the tunneling parameter matching model. The matching model based on BP neural network and PSO algorithm is compared in simulation experiments and engineering case. It is veriﬁed that the matching model based on SVM and EAIW-IPSO still maintains great accuracy and stability as the number of samples continues to decrease. The paper provides a better solution for the matching of tunneling parameters in actual engineering.


Introduction
e shield construction technology is widely used in tunnel construction in China due to its low construction cost, wide application range, and low disturbance to the surrounding environment [1][2][3]. Since the whole construction process is located in the underground space, the safety is very important. In addition, with the improvement of construction quality requirements, more and more factors need to be considered [4]. As for shield tunnel construction, the ground settlement control is the key among all the influencing factors. If the ground settlement is too large during the construction process, the soil will collapse and cause a safety accident, affecting the construction period [5]. Many factors affect ground subsidence [6], such as the tunnel shape and diameter, the engineering geological conditions at the location, the hydrological conditions, the level of construction technology, and the level of construction management. Some of these influencing factors are uncontrollable. Some influencing factors have little impact on the construction phase once the plan is determined. Among the controllable factors, tunneling parameters have the greatest impact on ground settlement during construction [7]. erefore, how to effectively control the ground settlement by selecting shield tunneling parameters needs to be further studied.
In practical engineering, it is difficult to establish mathematical expressions for the nonlinear relationship between tunneling parameters and ground settlement accurately. erefore, shield tunneling parameter matching belongs to a class of nonlinear single-objective optimization problems that are difficult to establish objective functions accurately. At present, many scholars combined machine learning and intelligence optimization algorithms to solve such problems. In order to control the earth pressure balance of the shield enclosure in real time, Liu et al. [8] established a relationship model about the process parameters, rolling power, and austenite grain diameter by the neural network. Based on the trained model, the ideal point method was applied to convert it into a single-objective optimization problem. And then genetic algorithm performs parameter optimization. Aiming to match the optimal construction parameters of super-large diameter earth pressure balance shield, Qiang [9] established the relationship model between the construction parameters and the ground settlement above the tunnel axis based on BP neural network. And then shield tunneling parameters are optimized by genetic algorithm. Based on the neural network and PSO algorithm, Yan et al. [10] improved the accuracy in measuring the elastic constant of carbon/carbon composite materials successfully. Xu and Li [11] sought to find the optimal production tank voltage and corresponding production conditions with the aim of reducing the production cost of electrolytic aluminum. e neural network was used to establish a prediction model for the relationship between the voltage and the influencing factors of the electrolytic aluminum tank. Based on the prediction model, Cell voltage optimization was performed using genetic algorithm. e abovementioned solutions are adopted to deal with the nonlinear single-objective programming problems which are difficult to establish the objective function accurately. It is known that the performance of the nonlinear relationship prediction model and the optimization algorithm will affect the final results. e training of the nonlinear relationship prediction model is based on actual sample data. Because the economics of obtaining samples needs to be considered in engineering, the number of the obtained samples is also limited. erefore, it is critical to obtain a better training model under the limited sample data.
To solve the shield tunneling parameter matching with the limited sample data, this paper proposes a model based on SVM and EAIW-IPSO. e SVM is used to construct the nonlinear relationship model between the tunneling parameters and the ground settlement. EAIW-IPSO is adopted to optimize the tunneling parameters. e paper studies the performance of the model when the number of training samples is decreasing. Meanwhile, the corresponding UI interface is developed. e study not only provides a better solution for shield tunneling parameter matching under the limited sample data in actual engineering but also realizes the interface operation to achieve the tunneling parameter matching conveniently.

Analysis of SVM and Its Parameter Selection.
In the 1990s, Professor Vapnik first proposed the SVM algorithm, which is based on the development of statistical learning theory. Both SVM and neural networks are machine learning algorithms [12,13]. Compared to the neural network, SVM has better improvements in local minimum values, network structure, convergence speed, and training sample size requirements [14]. e parameters involved in the SVM mainly include loss function parameters ε, penalty factor C, and parameters in the kernel function [15].
(1) e loss function parameter ε affects the number of vectors and the generalization ability of the training, usually ranging from 0.001 to 0.1. (2) e penalty factor C is the tolerance to the error. e bigger the value of C, the higher the accuracy of the training model. e smaller the value of C, the stronger the generalization ability.
(3) is study adopted the radial basis kernel function that has a parameter σ. e value of the parameter σ is generally in the range of 0.1 to 3.8.
In this study, genetic algorithms are used to select these parameter values in the SVM.

EAIW-IPSO Algorithm.
e PSO algorithm proposed by Eberhart and Kennedy in 1995 is one of the computational intelligent optimization algorithms [16]. In PSO algorithm, particle velocity and position are continuously adjusted according to the optimal particle in the search space [17]. e adjustment for particle velocity and position is mainly achieved by two extreme values [18]. One is the best solution that the particle itself obtained during the adjustment process, namely, the individual extremum. e other is the optimal solution that the entire particle population finds in all previous iterations, namely, the global extremum [19]. e adjustment equation for particle velocity and position is [20] where v is is the velocity of the ith particle in the sth direction (i � 1, 2, . . ., M) (s � 1, 2, . . ., N); ω is inertia weight; c 1 and c 2 are acceleration factors; r 1 and r 2 are random values at (0, 1); p is is the individual extremum; and p gs is the group extremum.
In PSO algorithm, the particle lacks mechanism of variation and only relies on the interaction between particles. In the iterative process, it is easy to fall into local extremum.
e trajectory of the whole iterative process is sinusoidal wave swing [21]. e early iteration speed is faster, the later speed is slowed down, and even the stagnation phenomenon occurs, which makes the algorithm have "premature" phenomenon.
us, the overall search ability is weak.
In order to improve the performance of the PSO algorithm, Hou et al. [22] improved the inertia weight adjustment equation and the population variation mechanism, proposing the EAIW-IPSO algorithm. EAIW-IPSO algorithm has higher accuracy in both single-peak and multipeak problems and has great stability. e new population variation mechanism in EAIW-IPSO algorithm reduces the possibility of the algorithm falling into local extremes. In addition, EAIW-IPSO algorithm can quickly converge near the optimal value in the early stage of iteration and perform the optimal value search in late period, which shows that EAIW-IPSO has better iterative characteristics. erefore, this paper applies EAIW-IPSO algorithm to optimize the shield tunneling parameters. e inertia weight adjustment equation in the EAIW-IPSO algorithm is where ω min is minimum inertia weight; ω max is maximum inertia weight; t is the current number of iterations; and T is maximum number of iterations. e population variation strategy in the EAIW-IPSO algorithm is before the next iteration and another particle population with size of M is regenerated. Among the original and new populations, M particles with larger fitness values are selected from the 2M particles as the group of the next iteration. e others are eliminated. 2 Advances in Civil Engineering EAIW-IPSO algorithm specific process: Step 1: set the value of all parameters, including M, T, c 1 , c 2 , ω min , and ω max .
Step 2: initialize the position and velocity values and calculate the fitness value of each particle.
Step 3: adjust the position and velocity values based on equations (1)-(3) and set the personal and global best position.
Step 4: initialize another particle group with size of M. M particles with larger fitness values are selected from the 2M particles as the group of the next iteration, updating the global best position.
Step 5: the next iteration is entered. If the termination condition is satisfied, the iterations stop.

Matching Model Based on SVM and EAIW-IPSO.
In order to better predict the nonlinear relationship between the tunneling parameters and ground settlement in actual engineering, formation and geometric condition parameters are considered in the study. Set the number including tunneling parameters, formation, and geometric condition parameters to m {x 1 , x 2 , . . ., x m }. e objective function value is the ground settlement value D. n groups of training samples (X(j), d(j)) (j � 1, 2, . . ., n) are obtained through the tunneling test section. Map the above sets of training samples to a high-dimensional (ldimensional) feature space and use the following function for linear regression [15]: where u is an l-dimensional weight vector, X is a set with m parameter variables, φ(X) is a mapping function, and b is an offset term. e abovementioned regression problem is transformed into the optimal value by using the ε insensitive loss function. e converted optimal value problem is as follows: subject to where ξ j and ξ * j are relaxation factors, C is a penalty factor, and ε is a loss function parameter.
Introducing the nonnegative Lagrangian multipliers α j , α * j , η j , and η * j to transform equation (5) into the following form: Equation (7) derivation of ω, b, ξ j , and ξ * j : Replace the corresponding variables in equation (7) with equations (8)- (11) and get the dual problem as follows: e appropriate kernel function is selected. In addition, the genetic algorithm is used to select the values including ε, C, and kernel function parameters. e optimal solution is obtained: Using the Karush-Kuhn-Tucker conditions (KKT) condition to get value of b, and the result is e value of b is averaged, and the SVM model, that is, ultimately applied to predict the nonlinear relationship between selected parameters (tunneling parameters, formation, and geometric condition parameters) and ground settlement is e EAIW-IPSO algorithm is used to optimize the tunneling parameters. Since the number of the selected parameter variables is m, each particle spatial position consists of these m parameter variables and the particle space dimension is m. e formation and geometric condition parameter values are artificially unadjustable.
us, the maximum and minimum parameter values are set equal to each other. e particle population size is set to G. Initialize the parameter variable values for each group.
. In addition, each particle speed needs to initialize.
Using the SVM model obtained from training to calculate the ground settlement value corresponding to each parameter variable set d(X (s) ) (s ∈ {1, 2, . . ., G}), the reciprocal of the settlement value is taken as the fitness value of the parameter variable set: Take each parameter variable set as the individual extremum of the corresponding particle P (s) � (p (s) . e parameter variable set with the biggest fitness value is used as the group extremum P (g) �(p (g) (1)-(3). After the update, each parameter variable value needs to be checked within the value range. If it is out of range, it is reinitialized. e fitness value corresponding to the updated parameter variable set is calculated.
e individual and group extreme values are updated.
en, other G parameter variable sets are generated, calculating the corresponding fitness values and setting individual extreme values. G parameter variable sets with higher fitness values are selected and used as the particle population of the next iteration. Update individual and group extremum and loop iterative process until the number of iterations ends.

Model Performance Verification.
Two examples were selected to study the performance of the model based on SVM and EAIW-IPSO. In the experiment, the performance of the model was observed by continuously reducing the number of training samples. At the same time, the results were compared with model based on BP and PSO.
Case 1.   rough simulation experiments, the total error of the prediction system (MSE) and the relative error of the objective function value are obtained and shown in Table 1.
e above mentioned data is plotted in Figures 1(a) and 1(b). From Figure 1(a), the values of MSE based on SVM are all lower than 0.01. e values of MSE based on BP neural network are all above 0.04. erefore, when the sample size is gradually reduced from 200 to 100, the prediction model based on SVM still keeps high accuracy. In addition, according to variation of curves from SVM and BP, it can be found that the curve based on SVM is significantly smaller as the number of samples continues to decrease, which shows that the SVM has better stability in the case of relatively small sample size.
us, the prediction model based on SVM has better comprehensive performance.
According to Figure 1(b), the relative error curve of the optimized objective function value based on SVM and EAIW-IPSO algorithm has small fluctuations of about 0.005. e relative error curve based on BP and PSO algorithm are over the curve based on SVM and EAIW-IPSO algorithm.
e maximum relative error based on BP and PSO algorithm is close to 0.04. erefore, the accuracy of the matching model based on SVM and EAIW-IPSO algorithm is better. In addition, it can be seen from the curve fluctuation that the curve variation based on SVM and EAIW-IPSO algorithm is significantly smaller.
ese results show that the model based on SVM and EAIW-IPSO algorithm still maintains   Advances in Civil Engineering high accuracy and stability when the number of training sample is small.
subject to  Table 2.
e abovementioned data is plotted in Figures 2(a) and 2(b). It is known that the values of MSE based on SVM are always lower than that based on BP neural network in Figure 2(a). Meanwhile, the maximum value of MSE based on BP is close to 0.08. e values of MSE based on SVM are all lower than 0.01. e difference between the maximum and minimum values of the MSE is 0.0006 based on SVM. Another is 0.051 based on BP neural network. ese data also show that the SVM has better accuracy and stability in the case of relatively small sample size.
In Figure 2(b), it shows that the variation of curve based on SVM and EAIW-IPSO algorithm is significantly smaller than that based on BP and PSO algorithm. In addition, the curve of the former is below the curve of the latter. ese results also verify the performance of the model based on SVM and EAIW-IPSO. erefore, the results from these two experiments all show that the matching model based on SVM and EAIW-IPSO not only has high accuracy but also has better stability.
is model can solve the parameter matching in the case of relatively small sample size.

UI Interface of Shield Tunneling Parameter Matching
Module. Based on the proposed shield parameter matching model, the UI interface of the tunneling parameter matching module was developed. e development language is C# and the development environment is Visual Studio 2012. e window UI interface is designed by Windows Forms technology.
e UI interface of the module is mainly composed of four tabPage pages, and the corresponding functions are reading and processing of training sample data, training of the SVM predictive model, testing of the SVM predictive model, and optimization of tunneling parameters.
Reading and processing of the training sample: the training sample data including the selected parameters and the surface settlement values in the .txt file are read by the data stream. Meanwhile, the data are normalized and displayed on the interface for the personnel to view. It should be noted that the data needs to be arranged in a certain format.
Training of the SVM prediction model: based on the genetic algorithm and support vector machine theory knowledge, the button from the UI interface is used to select the corresponding training parameters, train the model, and analyze the results. e genetic algorithm iterative process and the fit to the training sample are displayed by the chart control.
Test of the SVM prediction model: the test sample is read through the data stream and displayed in the interface. e ground settlement values corresponding to the selected parameter in the test sample is calculated based on the prediction model and compared with the actual ground settlement values. erefore, the model generalization ability can be analyzed.
Optimization of tunneling parameters: based on the trained prediction model, the proposed EAIW-IPSO algorithm theory knowledge is used to optimize the shied tunneling parameters. Each variable range is added through the ListBox control in the interface. e optimization result is displayed in the TextBox, and the optimization process is displayed through the chart control.

Case Study
e shield section of Shenwan Metro Line 2 between the Shenwan town to Wuyi Square is selected. e data in this section is used to test the model performance in practical application and feasibility of the UI interface. erefore, based on the data obtained in the literature [23], 34 groups of samples in which the maximum settlement values are all less than 10 mm were randomly acquired. e selected parameter sample values are shown in Table 3. e first 30 groups were applied to train the model and the last 4 groups were used as test samples. e selected geometric condition parameters is the ratio of buried depth H and diameter excavation D. Formation condition parameters are groundwater level, cohesion, internal friction angle, earthwork heavy, and side-pressure coefficient.
e tunneling parameters are synchronous grouting amount, shield thrust, cutter head torque, the ratio of tunneling speed and cutter speed R (the cutter speed is usually 1.5 rad/min), and earth pressure [24,25].

Implementation of Tunneling Parameter Matching by UI Interface Based on SVM and EAIW-IPSO Model.
In the SVM, the radial basis function is selected [26]. e loss function parameter ε, the penalty factor C, and the σ in the kernel function are optimized by genetic algorithm [27,28]. e reading interface of the training sample is shown in Figure 3.
e specific values including the iteration number, population size, crossover probability, and mutation probability in the genetic algorithm are shown in Figure 4. e SVM model is trained based on the training samples. e MSE is 0.0295. e fitting curves of training samples are shown in Figure 4 too. e predicted value curve and the actual value curve are basically merged. e model fits the training sample better. e test results for the trained SVM model are as follows: the relative error values are all lower than 10% except for one group. e average value of the relative error is 5.98%. erefore, the generalization ability of the model that is applied to build the relationship between the selected parameters and ground settlement is also better.
Based on the obtained SVM model, the tunneling parameters from the tunnel section in Table 4 are optimized.
e related parameter values in EAIW-IPSO are set [29]. e optimization results are shown in Figure 5. e obtained optimization tunneling parameter values based on SVM and EAIW-IPSO are shown in Table 5. e corresponding maximum ground settlement is 5.3 mm.

Implementation of Tunneling Parameter Matching by UI Interface Based on BP and PSO Model.
e initial weight of BP neural network is selected by genetic algorithm. e number of nodes in the input layer and output layer are 11 and 1, respectively. e number of nodes in the hidden layer is finally determined to be 3 by multiple experiments. e number of iterations is 1000 [30]. Based on the training samples, the BP neural network model is trained. e MSE is 0.4652. e fitting curve based on BP neural network is shown in Figure 6. It can be seen from the figure that the predictive effect is poorly around the 5th sample point. Moreover, it is also not well between the 10th sample point and the 20th sample point. erefore, the effect of the BP neural network is relatively poor in the case of the limited number of samples. e generalization ability test of the BP neural network model was carried out. e maximum and average relative errors are 36.55% and 12.72%, respectively.     8 Advances in Civil Engineering     Based on the BP neural network model, the tunneling parameters of the section in Table 4 are also optimized. e optimization results are shown in Figure 7 and Table 6. Figures 4 and 6, the SVM model has better fitting effect in the case of relatively small sample size. e predicted value curve and the actual value curve can be basically merged. But the BP neural network model has poor fitting effect in some of the sample points, and the overall predicted effect is not well. In addition, the MSE based on SVM is smaller than that based on BP neural network.

Model Generalization Performance.
e model prediction accuracy provides an important basis for subsequent parameter optimization. erefore, it is very important to improve the generalization ability of the model. e maximum predicted relative error based on SVM for the test sample is 13.87%, and the average relative prediction error is 5.98%. While the maximum predicted relative error based on the BP neural network is 36.55% and the average value is 12.72%. It can be seen that the generalization ability of the SVM is better than that of the BP neural network when the number of training samples is limited.

Optimization Results.
e optimal target value based on SVM and EAIW-IPSO model is 5.3 mm, and the optimal target value based on BP neural network and PSO model is 6.52 mm. e optimization results shows that the tunneling parameter values obtained by the model based on SVM and EAIW-IPSO can minimize the ground settlement value within the range of factors considered.
In actual construction, there are many factors in the construction site. It is difficult to quantify and consider all the factors at present. erefore, the obtained tunneling parameter values need further adjustment in engineering application. e performance requirements for the matching model are relatively high. us, the proposed model based   on SVM and EAIW-IPSO provides a more effective way for the selection of actual tunneling parameters, avoids the blindness of the selection, and provides a more reliable guarantee for the control of ground subsidence risk.

Conclusions
(1) In order to improve the reliability and accuracy for the tunneling parameter selection under the limited data, the model based on SVM and EAIW-IPSO algorithm is proposed and constructed. (2) In the simulation examples, it is verified that the parameter matching model based on SVM and EAIW-IPSO can maintain high accuracy and stability as the number of samples continues to decrease. Meanwhile, based on the parameter matching model of SVM and EAIW-IPSO, the corresponding tunneling parameter matching UI interface was developed to realize the interface operation. (3) e application in actual engineering case also verified that the model based on SVM and EAIW-IPSO is superior in sample fitting, model generalization ability, and optimization result. erefore, the proposed model provides a more effective solution for tunneling parameter matching in actual shield construction.

Data Availability
e data used to support the findings of this study are included within the article.

Conflicts of Interest
e authors declare that they have no conflicts of interest.