Method and device for deformation prediction and optimization of tunneling based on hybrid intelligent method

Abstract

The application belongs to the technical field of tunnel deformation control, and particularly discloses a mixed intelligent method-based excavation tunnel deformation prediction and optimization method and equipment. The method comprises the following steps: based on the monitoring data of small-interval excavation tunnel construction parameters, determining influence parameters and performing data preprocessing; using a Bayesian (BO) optimization random forest (RF) prediction model parameter to perform hyperparameter optimization and influence parameter importance sorting, and obtaining a prediction result; based on the nonlinear regression mapping relationship between the optimization target obtained by prediction and the influence parameters, establishing an objective optimization function in an NSGA-III model, obtaining a Pareto frontier solution set, and realizing multi-objective optimization of small-interval excavation tunnel construction. The application combines BO-RF-NSGA-III to construct a small-interval excavation tunnel deformation prediction and optimization control model, realizes advanced prediction of ground surface settlement, tunnel vault settlement and tunnel hance settlement, and provides a basis and guidance for small-interval excavation tunnel construction operation.

Description

TECHNICAL FIELD

[0001] The present application belongs to the technical field of tunnel deformation control, and in particular relates to a method and device for predicting and optimizing deformation of a tunnel excavated by a mixed intelligent method, and more particularly relates to a method and system for predicting and optimizing deformation of a small-spacing tunnel excavated by a BO (Bayesian-Optimization)-RF (Random Forest)-NSGA-Ⅲ method. BACKGROUND

[0002] In the process of small-spacing tunnel excavation construction, the advancement of the tunnel face causes ground settlement and adjacent tunnel settlement deformation, and there are unavoidable safety risk problems. By adjusting and optimizing the key construction influencing factors that cause ground settlement and adjacent tunnel deformation, the deformation of the tunnel excavated can be predicted, which can effectively control the construction risk of the small-spacing tunnel excavated. Accurate small-spacing tunnel deformation prediction plays an important role in improving tunnel safety. The ground settlement and tunnel settlement deformation caused by small-spacing tunnel construction can be predicted by using construction parameters with direct physical meaning as input variables. In combination with the NSGA-III method, multi-objective optimization is realized, the key control factors of small-spacing tunnel excavation are optimized and designed, and a BO-RF-NSGA-Ⅲ multi-objective intelligent optimization prediction model is established. SUMMARY

[0003] In view of the above defects or improvement needs of the prior art, the present application provides a method and device for predicting and optimizing deformation of a tunnel excavated by a mixed intelligent method, which introduces a BO-RF-NSGA-Ⅲ method for predicting and optimizing deformation of a small-spacing tunnel excavated, and provides a new idea and solution for the application of machine learning in the field of engineering to predict and optimize deformation of a small-spacing tunnel excavated. The present application provides guidance and reference for small-spacing tunnel excavation construction operation. The present application uses Bayesian (Bayesian-Optimization, BO) optimization and random forest (Random forest, RF) prediction model parameters, and in combination with NSGA-Ⅲ, proposes a framework for predicting and optimizing deformation of a small-spacing tunnel excavated, and constructs a system for predicting deformation of a small-spacing tunnel excavated. The intelligent prediction method and system for small-spacing tunnel excavation first screen and determine the influencing parameters in combination with the actual small-spacing tunnel construction conditions, mobilize the relevant data in the construction monitoring database for preprocessing, then perform hyperparameter optimization and importance sorting based on Bayesian optimization and random forest, establish a prediction model, and finally establish a target optimization function based on the prediction model, perform multi-objective optimization based on NSGA-Ⅲ, and obtain the optimal Pareto solution, which provides timely and useful information and guidance for the tunneling construction of a small-spacing tunnel excavated.

[0004] To achieve the above object, according to one aspect of the present application, a mixed intelligent method-based tunneling deformation prediction and optimization method is provided, comprising the following steps:

[0005] Step one: based on the monitoring data of small-interval tunneling construction parameters, determine the influence parameters and perform data preprocessing;

[0006] Step two: use the Bayesian optimization RF prediction model parameters to perform hyperparameter optimization and influence parameter importance sorting to obtain the prediction result;

[0007] Step three: based on the nonlinear regression mapping relationship between the predicted optimization target and the influence parameters, establish the objective optimization function in the NSGA-III model, obtain the Pareto frontier solution set, and realize the multi-objective optimization of small-interval tunneling construction.

[0008] Further, the step one based on the monitoring data of small-interval tunneling construction parameters to determine the influence parameters and perform data preprocessing comprises the following steps:

[0009] (1) Determine the influence parameters

[0010] By analyzing the small-interval tunneling construction process and referring to a large amount of practical experience and related literature, four main influence parameters can be determined, which are geometric parameters, surrounding rock parameters, blasting parameters and tunneling parameters. An input index system for influencing the ground settlement and adjacent tunnel settlement deformation is proposed: tunnel buried depth (x1), span ratio (x2), tunnel centerline spacing (x3), interlayer thickness (x4), surrounding rock strength (x5), soil elastic modulus (x6), soil cohesion (x7), soil internal friction angle (x8), soil density (x9), peripheral hole spacing (x 10 ), blast hole density coefficient (x 11 ), smooth layer thickness (x 12 ), tunneling speed (x 13 ), longitudinal spacing of working face (x 14 ), soil reinforcement strength (x 15 ).

[0011] (2) Data preprocessing

[0012] Before training the model, the data of the sample set needs to be normalized to eliminate the influence of the characteristic value dimension of different samples on the prediction efficiency and accuracy. This specification will normalize the data samples of different targets to the interval [-1, 1], so that the input variables can be directly compared, avoiding the influence of different magnitudes of different targets on subsequent optimization. The normalization formula is as follows:

[0013]

[0014] where y is the normalized standard value, y max and y min is usually 1 and -1, x is the sample value, x max and x min is the maximum and minimum value of the sample value.

[0015] As a further preferred, the step two of using Bayesian optimization RF prediction model parameters, hyperparameter optimization and influence parameter importance ranking, to obtain the prediction results, including the following steps:

[0016] (1) Hyperparameter optimization

[0017] Different values of RF algorithm parameters will directly affect the regression prediction effect of the model, therefore, when training the sample using RF prediction algorithm, first, the important parameters in the prediction model need to be tuned. The RF prediction model needs to tune the parameters mainly have three, respectively, the maximum depth of the decision tree in the random forest model (max_depth), the maximum number of random features extracted at each node (max_features) and the maximum number of weak learners iterations (n_estimators), the value of the three parameters has a direct impact on the prediction performance of the RF model, the larger the parameter value, the better the regression fitting performance of the model, however, when the parameter value is too large, the calculation amount increases, and the prediction performance of the model is negligible. Therefore, the hyperparameters need to be set reasonably to achieve the best prediction performance of the model.

[0018] (2) Feature importance

[0019] The random forest algorithm is an integrated learning algorithm, which can not only predict variables, but also calculate the importance of feature parameters using the training model. By calculating the importance of construction parameters, index screening, and considering the actual situation of the project and control requirements, the construction parameters that need to be focused on in the tunneling process are selected for focused management, so as to reduce the ground subsidence.

[0020] In order to enhance the diversity of the decision tree, the training set is randomly sampled with replacement, and the number of data extracted is the data amount of the training data. If the training data has n samples, the probability of the same sample being drawn each time is The probability of not being drawn is The probability of not being drawn for m times is These data are called out-of-bag data (OOB).

[0021] The RF algorithm evaluates the importance of input variables based on out-of-bag data error, and evaluates the importance of feature variables by measuring the reduction of mean square residual error (%Inc MSE) and the reduction of model accuracy (Inc Node Purity) after random permutation. The process is as follows:

[0022] Step 1: Establish a regression decision tree. Assuming that there are m decision trees in the random forest, the out-of-bag data can be predicted by the random forest model, and the mean square residual error of b out-of-bag data can be obtained, which are MSE1, MSE2, …, MSE b The calculation formula is as follows:

[0023]

[0024] Where y i represents the true value of the dependent variable in the out-of-bag data; represents the predicted value of the regression model, the value of the jth feature variable X j of the out-of-bag data is randomly changed, and the new out-of-bag error accuracy MSE j is calculated.

[0025] Step 2: Generate error matrix. When constructing a regression decision tree for splitting, the feature variable is randomly selected. The variable X j is randomly permuted in b out-of-bag data samples, then a new OOB test set is formed, and the established random forest regression model is used to predict the new test set again, and the new OOB residual mean square MSE ij is obtained. The error matrix A is generated as follows:

[0026]

[0027] Where P is the number of influencing factor variables; b is the number of training sample set.

[0028] Step 3: Importance score. Subtract MSE1, MSE2, …, MSE b from the corresponding row of matrix A, take the average, and then divide by the standard error to obtain the average reduction of mean square residual error of variable X i , i.e. the importance score VIM i of the feature variable (MSE), which can be expressed as:

[0029]

[0030] Where MSE j is the mean square residual error of the jth sample; S E is the standard error. The higher the importance score of the feature variable, the more important the variable is to the model result.

[0031] (3) Prediction model precision evaluation

[0032] The prediction model obtained by the random forest algorithm also needs to evaluate the prediction precision of the model, so the root mean square error RMSE and the goodness of fit R 2 are introduced to test the precision of the model. Among them, RMSE is a measure of the deviation between the true value and the predicted value, and the better the model fitting effect, the closer its value is to 0; R 2 is used to measure the dispersion degree of the sample, and its value range is 0~1, and the better the model fitting effect, the closer its value is to 1. Each index is calculated through the following equation:

[0033]

[0034]

[0035] In the formula, n is the sample number, y i is the actual observation value; is the average value of the actual observation value; f i is the predicted value.

[0036] In order to further verify the reliability of the Bayesian optimization RF prediction model established in the present application, the prediction performance of the Bayesian optimization RF prediction model and other commonly used models is compared. Back propagation neural network (BPNN), support vector machine (SVM) and gradient boosting decision tree (GBDT) are commonly used methods for surface deformation and tunnel settlement deformation, so back propagation neural network (BPNN), support vector machine (SVM) and gradient boosting decision tree (GBDT) three models are selected for comparison. In order to make the results of the model comparison more fair, the three prediction models of back propagation neural network (BPNN), support vector machine (SVM) and gradient boosting decision tree (GBDT) and the Bayesian optimization RF prediction model use the same programming environment, performance evaluation index and data sample.

[0037] As a further preferred, the nonlinear regression mapping relationship between the predicted optimization target and the influence parameter in step three is used to establish the objective optimization function in the NSGA-III model, obtain the Pareto frontier solution set, and realize the multi-objective optimization and prediction of small-spacing tunnel construction, including the following steps:

[0038] (1) Design scenario analysis

[0039] Safety, cost and duration are the key concerns in the decision-making of construction parameter control during tunneling, but it is difficult to achieve the optimal reduction of safety risk and cost and the optimal shortening of duration simultaneously. In order to consider multiple decision objectives simultaneously, the geometric parameters, surrounding rock parameters, blasting parameters and tunneling parameters during the construction of small-spacing underground tunnels are selected as decision variables, and the minimization of surface settlement and adjacent tunnel settlement deformation is selected as the objective to construct a multi-objective optimization problem.

[0040] In actual engineering, only a few construction operation parameters can be adjusted, and it is not cost-effective to adjust all parameters simultaneously, so it is necessary to determine the constraint parameters of multi-objective optimization based on the importance and influence law of construction parameters, and design different scenarios by adjusting parameters for scenario analysis.

[0041] (2) Establish the objective optimization function

[0042] Before multi-objective optimization, the objective optimization function between construction parameters and three optimization objectives needs to be established. The nonlinear function relationship between surface settlement, adjacent tunnel settlement deformation, soil reinforcement strength, tunneling speed and longitudinal distance of working face is obtained by using the random forest prediction model, and the function relationship of the three is used as the fitness function of NSGA-Ⅲ algorithm according to the empirical formula.

[0043] (3) Set the parameter constraint range

[0044] In the optimization design of parameters, in order to ensure that the generated optimization scheme can be reasonably implemented, it is necessary to set the constraint range of each parameter. In order to avoid the large deviation between the parameters optimized by NSGA-Ⅲ algorithm and the actual parameters, which cannot be realized in actual engineering, the construction parameters need to be constrained. The constraint condition is generally expressed as:

[0045] b bl <x i <b ul (7)

[0046] Where x i represents the i-th input parameter, b bl and b ul represent the lower limit and upper limit of the i-th design parameter value, respectively.

[0047] (4) NSGA-Ⅲ multi-objective optimization

[0048] The steps of optimizing the set multi-objective based on NSGA-Ⅲ algorithm are as follows:

[0049] 1) Set an initial population represented as P tAccording to the given variable boundary limit, a decision variable is randomly generated, thereby generating an initial population with N individuals.

[0050] 2) Selection, crossover and mutation operations are performed on the first generation population to obtain a first batch of offspring population with the same size as the first generation population. The two populations are merged to obtain a population with double size;

[0051] 3) Non-dominated sorting is performed on the double-size population to determine the non-dominated level of the individuals. The individuals are retained in descending order of level until the next generation offspring set is greater than the size of the initial population. The non-dominated level at this time is recorded.

[0052] 4) The minimum value of each dimension of the individual in the current population is selected to form the ideal point of the current population:

[0053]

[0054] In the formula, is the minimum value of each objective function.

[0055] The population is translated to move the ideal point to the origin. Scalarization is performed on each objective function to find the extreme point, calculate the intercept on the corresponding coordinate axis, and perform normalization operation.

[0056] 5) Calculate the reference points on multiple objective functions. The number of reference points depends on the dimension M of the objective vector and another positive integer H, as shown in the following formula:

[0057]

[0058] The number of solutions to the equation can be calculated as follows:

[0059]

[0060] Let (x j,1 ,x j,2 ,...,x j,m ) T be the jth solution to the equation, and the reference point λ j is:

[0061]

[0062] The line connecting the ideal point and the reference point is the reference line L. The distance of the individual in the population to each reference line is calculated. The individual with the closest distance to the reference line is associated with the corresponding reference line. Assuming u is the projection of f(x) on the reference line L, d j,1 (x) is the distance between the origin and u, and d j,2 (x) is the perpendicular distance from f(x) to line L. The distance calculation formula is:

[0063]

[0064] 6)Screening offspring and deleting reference points, each reference point is traversed to find the reference point with the least number of references, that is, the reference point associated with the least number of population individuals. If the number of population individuals associated with this reference point is zero, find the point with the smallest distance from the individuals associated with this reference point vector and extract it to be added to the selected next generation population and the number of references is +1; if no individual is associated with the reference point, delete the reference point vector, if the number of references is > 0, select the nearest reference point from it until the population size is N, and obtain the Pareto optimal solution set.

[0065] 7) The ideal point method is used to select the optimal solution from the Pareto solution set.

[0066] Taking three-objective optimization as an example, the distance between each solution point in the Pareto front and the ideal point is calculated, and the calculation formula is as follows:

[0067]

[0068] In the formula, D i is the distance between the average of all points and the ideal point, (x i ,y i ,z i ) is the coordinate corresponding to the optimal Pareto front point, (x Epoint ,y Epoini ,z Epoint ) is the coordinate corresponding to the ideal point, and the point with the smallest distance is the optimal point, that is, the distance between each point in the optimized Pareto front and the ideal point is calculated as follows:

[0069] D opt = min(D i ) (14)

[0070] Therefore, an optimal solution can be determined from the Pareto solution set using the ideal point method.

[0071] According to another aspect of the present application, there is also provided a small-spacing tunneling deformation prediction and optimization control system based on BO-RF-NSGA-III, comprising: a first main module: based on the monitoring data of the small-spacing tunneling construction parameters, data preprocessing is performed; a second main module: RF prediction model parameters are optimized by using Bayesian optimization, hyperparameter optimization and influence parameter importance sorting are performed, and a prediction result is obtained; and a third main module: based on the nonlinear regression mapping relationship between the optimization target and the influence parameters obtained by prediction, an objective optimization function in the NSGA-III model is established, a Pareto frontier solution set is obtained, and multi-objective optimization and prediction of small-spacing tunneling construction are realized.

[0072] According to another aspect of the present application, there is also provided an electronic device, comprising:

[0073] at least one processor, at least one memory and a communication interface; wherein,

[0074] The processor, the memory and the communication interface communicate with each other;

[0075] The memory stores program instructions executable by the processor, and the processor invokes the program instructions to execute the above method.

[0076] According to another aspect of the present application, a non-transitory computer readable storage medium stores computer instructions, and the computer instructions make the computer execute the above method.

[0077] Overall, compared with the prior art, the above technical solutions conceived by the present application mainly have the following technical advantages:

[0078] 1. The present application pre-processes the initial data collected, and on this basis, important sorting and feature selection are performed by using RF, and the optimal parameter set for surface subsidence, tunnel vault subsidence and tunnel haunch subsidence prediction is obtained, thereby providing conditions for small-spacing tunneling deformation prediction and optimization control based on BO-RF-NSGA-III.

[0079] 2. The present application determines the optimal hyperparameters of the RF prediction model by using Bayesian optimization, and based on the selected hyperparameters, the surface subsidence, tunnel vault subsidence and tunnel haunch subsidence prediction model is constructed, thereby realizing accurate small-spacing tunneling risk prediction and providing a reference for small-spacing tunneling construction operation. BRIEF DESCRIPTION OF DRAWINGS

[0080] Figure 1 The flowchart of the small-spacing tunneling deformation prediction and optimization control method and system based on BO-RF-NSGA-III provided by the embodiments of the present application is provided;

[0081] Figure 2 (a), (b), (c) in FIG. 1 are schematic diagrams of random forest algorithm hyperparameter optimization results provided by embodiments of the present application;

[0082] Figure 3 (a), (b), (c) in FIG. 2 are schematic diagrams of influence factor importance distribution provided by embodiments of the present application;

[0083] Figure 4 (a), (b), (c) in FIG. 3 are small-interval tunnel risk prediction results of Bayesian optimization random forest provided by embodiments of the present application;

[0084] Figure 5 (a), (b), (c), (d), (e), (f), (g), (h) in FIG. 4 are NSGA-III three-objective optimization Pareto solution result graphs provided by embodiments of the present application. DETAILED DESCRIPTION

[0085] In order to make the objects, technical solutions and advantages of the present application clearer, the present application will be further described in detail below in combination with the drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and do not limit the present application. In addition, the technical features involved in each embodiment of the present application described below can be combined with each other as long as they do not conflict with each other.

[0086] As Figure 1 , the small-interval cut-and-cover tunnel deformation prediction and optimization control and system based on BO-RF-NSGA-III proposed by the present application mainly includes the following steps:

[0087] (1) Parameter selection

[0088] The analysis of the construction process of small-interval cut-and-cover tunnel and the reference of a large amount of practical experience and related literature can determine four main influence parameters, which are geometric parameters, surrounding rock parameters, blasting parameters and tunneling parameters. This paper proposes an input index system for influencing the ground settlement and adjacent tunnel settlement deformation, as shown in Table 1: tunnel burial depth (x1), span ratio (x2), tunnel centerline spacing (x3), interlayer thickness (x4), surrounding rock strength (x5), soil elastic modulus (x6), soil cohesion (x7), soil internal friction angle (x8), soil density (x9), peripheral eye spacing (x 10 ), blast hole density coefficient (x 11 ), smooth layer thickness (x 12 ), tunneling speed (x 13 ), longitudinal spacing of working face (x 14 ), soil reinforcement strength (x 15 ).

[0089] Table 1 15 input indicators affecting ground settlement and tunnel settlement deformation and their definitions

[0090]

[0091] The tunnel depth has a direct impact on the ground settlement of small-interval underground excavation tunnel construction. With the increase of tunnel depth, the ground transverse settlement gradually decreases, and the settlement groove width gradually increases. The cover-span ratio is one of the key variables for estimating ground settlement and may have an adverse impact on tunnel settlement. The tunnel centerline spacing is an important basis for reflecting the construction trend of parallel tunnels and has an important impact on the interaction of the tunnel face excavation. The impact on the intercalated rock in tunnel construction is the most direct, which disturbs and destroys the stability of the intercalated rock, and then endangers the adjacent tunnel. The increase of the thickness of the intercalated rock will weaken some negative effects. The greater the strength, elastic modulus, internal friction angle, cohesion and density of the soil, the greater the friction between the soil and the equipment, resulting in larger tunnel face thrust, slower advancing speed, more difficult soil removal and smaller ground settlement. The peripheral hole in the blasting construction is related to the shaping of the tunnel, and the spacing should not be too large. The blast hole density coefficient and smooth layer thickness have a greater impact on the blasting effect. Too fast excavation speed will increase the disturbance to the soil and affect the stability of the tunnel, and too slow speed will affect the construction period and reduce economic benefits, so it needs to be reasonably valued. The longitudinal spacing of the tunnel face will have an important impact on the plastic zone of the intercalated rock column, and is an important factor that cannot be ignored in the construction process of small-interval parallel tunnels.

[0092] (2) Data preprocessing

[0093] Through real-time data recording and monitoring on site, some construction data were obtained, and Table 1 shows the detailed information of the obtained data. The distribution of input and output parameters is shown.

[0094] Table 1 Detailed information of underground excavation tunnel construction data

[0095]

[0096] (3) Hyperparameter optimization

[0097] Before applying machine learning algorithms to predict ground settlement and tunnel settlement deformation, the hyperparameters of each machine learning model need to be optimized, which can improve the accuracy of model prediction. In order to obtain the best prediction results, the maximum number of features max_features is set to default auto, and max_depth and n_estimators are optimized and adjusted. The performance evaluation index for training is mean square error MSE. The initial range of max_depth and n_estimators is set to [7, 10] and [0, 100], respectively.

[0098] The optimal hyperparameters obtained by the Bayesian optimization algorithm, in Figure 2 (a) When n_estimators = 69 and max_depth = 8, the model reaches the optimal, and the MSE is 0.0145; in Figure 2 (b) When n_estimators = 77 and max_depth = 9, the model reaches the optimal, and the MSE is 0.0042; in Figure 2 (c) When n_estimators = 65 and max_depth = 7, the model reaches the optimal, and the MSE is 0.0152.

[0099] (4) Importance ranking

[0100] There is a certain correlation between each parameter index, so it cannot be qualitatively analyzed that a single parameter has an impact on the ground settlement. Feature importance analysis can find the relationship between different features and the prediction target, which helps to understand and find the sensitive factors affecting the ground settlement, and then strictly control it. Therefore, the random forest algorithm can be used to calculate the importance of each feature and rank the importance of these features, as shown in Figure 3 .

[0101] The primary and secondary factors of the influence of construction parameters on ground settlement are: longitudinal distance of working face > driving speed > peripheral eye distance > smooth surface thickness > blast hole density coefficient > soil reinforcement strength > surrounding rock strength > soil elastic modulus > soil cohesion > soil density > thickness of interlayer > tunnel centerline distance > tunnel burial depth > coverage ratio > internal friction angle. The primary and secondary factors of the influence of construction parameters on tunnel vault settlement are: longitudinal distance of working face > driving speed > soil reinforcement strength > peripheral eye distance > smooth surface thickness > blast hole density coefficient > thickness of interlayer > tunnel centerline distance > surrounding rock strength > soil elastic modulus > soil cohesion > tunnel burial depth > coverage ratio > soil density > internal friction angle. The primary and secondary factors of the influence of construction parameters on tunnel haunch settlement are: longitudinal distance of working face > driving speed > soil reinforcement strength > thickness of interlayer > peripheral eye distance > smooth surface thickness > blast hole density coefficient > tunnel centerline distance > surrounding rock strength > soil elastic modulus > soil cohesion > tunnel burial depth > coverage ratio > soil density > internal friction angle. The research results show that:

[0102] 1) The longitudinal distance of working face is the main factor that affects the ground settlement and tunnel settlement deformation in small interval excavation construction parameters. The change of the longitudinal distance of working face changes the distribution of the surrounding rock pressure of the cavern, so that the construction mechanics of the cavern is different. Increasing the longitudinal distance of working face can avoid the superposition of the disturbance effect of the two caverns on the surrounding rock, and can fully utilize the reinforcement effect of the support structure of the leading cavern to reduce the construction influence of the following cavern.

[0103] 2) The edge-to-eye distance, the smooth layer thickness and the blast hole density coefficient are all related factors of blasting construction, and the disturbance to the ground settlement is similar. Reasonable control of blasting construction technology can effectively control the overbreak, underbreak of surrounding rock soil, and reduce the disturbance to the soil. The excavation speed directly affects the disturbance time of the excavation construction to the surrounding rock structure, and appropriate improvement of the excavation speed can effectively reduce the settlement deformation of the soil and the tunnel caused by the time-space effect.

[0104] 3) In China's tunnel engineering, tunnels are divided into deep-buried tunnels and shallow-buried tunnels according to the depth, and the critical depth is determined according to whether the tunnel top cover layer can form a pressure arch natural arch. According to the different depths, the calculation method of the surrounding rock pressure is adjusted. When the tunnel depth is shallow, the ground movement caused by strata loss during construction is obvious, and the influence on the surrounding environment is larger. At the same time, the surrounding rock pressure is closely related to the tunnel depth, and the change of the depth and the span ratio directly determines the change of the surrounding rock pressure, which is an important factor affecting the stability of the surrounding rock.

[0105] 4) The thickness of the interlayer directly affects the deformation of the arch waist of the subsequent tunnel, and the greater the thickness, the smaller the stress on the arch waist of the subsequent tunnel. The influence of construction parameters and geometric parameters on ground settlement is greater than that of geological parameters. Figure 3 It is shown that the influence of soil cohesion and internal friction angle is relatively small, which is caused by the small change of soil parameters and aggregate parameters during the excavation process.

[0106] (5) Analysis of prediction results

[0107] Among the 300 groups of data obtained by monitoring, 260 groups of data are selected as the training set to train the model, and the remaining 40 groups of data are used as the test set to verify the model. According to the random forest prediction model, the training model and model fitting prediction are established for the ground settlement, cutter wear and excavation speed. The root mean square error (RMSE) and the goodness of fit (R 2 ) of the predicted results are calculated by equations (16) and (17), and the results are shown in Figure 4 .

[0108] From (a) in Figure 4 , it can be seen that the random forest algorithm fully learns the relationship between each influencing index and the ground settlement by training the training samples, and the error between the predicted value of the training model fitting and the actual value of the ground settlement is small. The model obtained by training the random forest algorithm is used to predict the test set data, and from (b) in Figure 4 , it can be more intuitively seen that the established random forest prediction model of the ground settlement has good fitting effect on the predicted value and the actual value of the test set data.

[0109] The mean square error (RMSE) between the predicted and actual values ​​in the surface subsidence training set is 0.239, and the goodness of fit R0 is [value missing]. 2 The mean squared error (RMSE) of the test set is 0.943, and the goodness of fit (R²) is 0.233. 2 The two metrics show that the random forest algorithm has a good predictive effect on the surface subsidence prediction model. The mean square error (RMSE) between the predicted and actual values ​​in the existing tunnel arch subsidence training set is 0.246, and the goodness of fit R0 is [value missing]. 2 The mean squared error (RMSE) of the test set is 0.951, and the goodness of fit (R²) is 0.205. 2 The two metrics show that the Random Forest algorithm has a good predictive effect on the model used for predicting the settlement deformation of existing tunnel arches. The mean square error (RMSE) between the predicted and actual values ​​of the settlement deformation of existing tunnel arches in the training set is 0.211, and the goodness of fit R0 is 0.951. 2 The mean squared error (RMSE) of the test set is 0.943, the mean squared error (RMSE) of the test set is 0.243, and the goodness of fit (Rfit) is 0.943. 2 The value is 0.932. These two indicators show that the random forest algorithm has a good predictive effect in building models for predicting surface subsidence, subsidence deformation of existing tunnel arches, and subsidence deformation of existing tunnel waists.

[0110] To further verify the effectiveness of the RF prediction model, its accuracy was compared with that of backpropagation neural network (BPNN), support vector machine (SVM), and gradient boosting decision tree (GBDT) prediction models. The table shows the prediction accuracy of each model, indicating that the RF model achieved the highest accuracy in this study. Therefore, this paper uses the RF algorithm to establish a nonlinear functional relationship between the shield tunneling parameters and the control objective, serving as the fitness function for multi-objective optimization.

[0111] Table 3. Prediction accuracy of the four models

[0112]

[0113] (6) Design Scenario Analysis

[0114] In actual engineering, only a few construction operation parameters can be adjusted, and adjusting all parameters would require a large amount of resources, which is not cost-effective. Therefore, based on the analysis of the importance and influence of construction parameters, the tunneling speed x is selected as the key parameter. 13 Longitudinal spacing of the working face x 14 Strengthening soil strength x 15 These three construction operation parameters, which have the greatest impact on surface settlement and tunnel settlement deformation, were adjusted to create seven scenarios. In scenarios 1 to 3, only one construction operation parameter was adjusted, namely x. 13 x14 or x 15 ; in scenarios 4 to 6, two construction operation parameters are adjusted, respectively, x 13 and x 14 for scenario 4, x 14 and x 15 for scenario 5, and x 13 and x 15 for scenario 6; in scenario 7, all three construction operation parameters are adjusted.

[0115] Table 4. Adjusted parameters of each scenario

[0116]

[0117] (7) Establishing the target optimization function

[0118] Before multi-objective optimization, the target optimization function between the influence parameters and the optimization target needs to be established. The target functions of ground surface settlement, tunnel vault settlement, and tunnel haunch settlement are obtained from the fitting relationship between the ground surface settlement predicted by the random forest algorithm and the construction parameters, and the trained regression function is introduced as the fitness function of optimization. Taking scenario 7 as an example, the target functions of ground surface settlement, tunnel vault settlement, and tunnel haunch settlement based on RF algorithm are:

[0119]

[0120] In the formula, f1, f2, and f3 are the fitness functions of ground surface settlement, tunnel vault settlement, and tunnel haunch settlement, respectively, x 13 , x 14 , and x 15 are the driving speed, the longitudinal distance of the working face, and the strength of the reinforced soil, respectively.

[0121] (8) Setting the parameter constraint range

[0122] When optimizing the parameters, in order to make the generated parameter combination more reasonable and feasible, it is also necessary to set the limit range of each decision variable according to the actual situation of the project and relevant specifications. Considering that the difference between the optimized construction parameter results and the actual engineering parameters is too large, it will cause a large parameter adjustment span and safety hazards. In this paper, the range of the initial parameters is set by referring to the range of relevant literature specifications and measured data as the main reference, which is specifically represented as:

[0123] Table 5. Range table of input parameters

[0124] Input parameters Parameter range Excavation speed (x 13 )]]> 1.5m / d ≤ x 13 ≤ 3m / d Longitudinal spacing of the faces (x 14 )]]> 45 < x < 55 14 110 < x Strengthened soil strength (x 15 )]]> 10 MPa < x 15 ≤ 36.5 MPa

[0125] (9) Optimizing to obtain the Pareto front solution set

[0126] The ground settlement, existing tunnel vault settlement deformation and existing tunnel waist settlement deformation are minimized as the optimization objective, and the NSGA-III algorithm is used for global optimization to determine the optimal solution of the operating parameter combination. Before RF-NSGA-III multi-objective optimization, the values of the model parameters need to be determined. The number of objectives of the genetic algorithm is set to 3, the population size is set to 100, the crossover operator is 0.7, the mutation operator is 0.01, the maximum evolution generation and the stop generation are 60, and after the above algorithm parameter settings, the multi-objective optimization is carried out according to the steps of the NSGA-III algorithm, and the optimal Pareto front is obtained.

[0127] Figure 5 The optimization results of scenarios 1 to 7 and the improvement degree of the three objectives under different scenarios are shown respectively. The optimization results of ground settlement, existing tunnel vault settlement deformation and existing tunnel waist settlement deformation under different scenarios are shown in Table 6.

[0128] Table 6 Improvement results of three optimization objectives under different scenarios

[0129]

[0130]

[0131] Note: The values in parentheses are the optimization percentage of the average value of the optimized value compared with the average value of the original data

[0132] (10) Result analysis

[0133] According to the multi-objective optimization results, multiple parameter optimization schemes can be obtained based on the NSGA-III algorithm, and the ideal point method is used to determine the optimal solution, and the distance between each solution point and the ideal point in the Pareto front is calculated. The point with the smallest distance is the optimal solution. Taking scenario 1 as an example, the ideal point refers to the point composed of the optimal values of each objective on the Pareto front, which is Figure 5 It can be seen that the ideal point coordinates composed of the optimal values of ground settlement, existing tunnel vault settlement deformation and existing tunnel waist settlement deformation are E(16.51, 22.22, 44.89), and the point coordinates are substituted into the distance calculation formula (13) to calculate the distance between each Pareto solution and the ideal point. According to formula (14), the point with the smallest distance is selected as the optimal solution of the multi-objective optimization of construction parameters, and the result is that the ground settlement is 15.79 mm, the existing tunnel vault settlement deformation is 21.76 mm, and the existing tunnel waist settlement deformation is 43.63 mm. Compared with the average value of the original data sample, the average improvement degree is 6.74%, and the optimal solutions under seven scenarios are shown in Table 7.

[0134] 1) This method can effectively solve multi-objective problems. Surface subsidence, existing tunnel arch subsidence deformation, and existing tunnel waist subsidence deformation were all well optimized and controlled under seven scenarios. For example... Figure 5 As shown, the proposed method effectively constructs the Pareto front for all cases and generates the optimal solution for each case under different scenarios. Compared with the average of the original data sample, the average improvement across the seven scenarios is 17.51%. Case studies validate the applicability of the multi-objective optimization results. The results demonstrate that the solution based on the BO-RF-NSGA-Ⅲ framework can simultaneously achieve the triple objective optimization of reducing surface subsidence, existing tunnel arch subsidence deformation, and existing tunnel waist subsidence deformation.

[0135] 2) The more construction parameters for small-spacing cut-and-cover tunnels are adjusted, the better the optimization and control effect will be on surface settlement, existing tunnel arch settlement and deformation, and existing tunnel waist settlement and deformation. For example... Figure 5 As shown in (h), in scenarios 1-3, adjusting only one construction parameter resulted in an average improvement of 12.02%; in scenarios 4-6, adjusting two construction parameters simultaneously resulted in an average improvement of 20.41%; and in scenario 7, adjusting all three construction parameters simultaneously resulted in the highest average improvement of 25.31%. Figure 5 As can be seen from (h), the surface settlement, existing tunnel arch settlement deformation, and existing tunnel waist settlement deformation of the optimal Pareto solution generated by scenario 7 are smaller than those of the optimal Pareto solutions generated by scenarios 1-6. This indicates that adjusting the three construction parameters can achieve the best optimization effect. Therefore, identifying and adjusting more small-spacing tunneling operation parameters can enhance the optimization effect.

[0136] 3) There are conflicts among the three objectives, and a complex relationship exists between the operational parameters of small-spacing tunnel construction and the optimization objectives. As shown, the Pareto front solution is widely distributed along the axis, covering the two extremes of objectives f1, f2, and f3. Achieving the optimum of one objective requires sacrificing the other two. Therefore, there are conflicts among the three objectives, making it difficult to simultaneously optimize multiple objectives. Furthermore, different tunnel excavation operational parameters improve different optimization objectives to varying degrees. For objective f1, considering parameter x... 13 In this scenario, the average percentage improvement is 26.91%, considering parameter x. 14 In this scenario, the average percentage improvement is 38.33%, considering parameter x. 15 In the given scenario, the average improvement percentage is 27.23%; for objective f2, considering parameter x 13 In this scenario, the average percentage improvement is 15.55%, considering parameter x. 14 In this scenario, the average percentage improvement is 18.92%, considering parameter x.15 The average improvement percentage is 17.36% for the scenario of the target f1, considering the parameter x 13 The average improvement percentage is 9.85% for the scenario of the target f2, considering the parameter x 14 The average improvement percentage is 11.52% for the scenario of the target f3, considering the parameter x 15 The average improvement percentage is 10.14% for the scenario of the target f3, considering the parameter x 14 The optimization of the target f1, the target f2 and the target f3 has a greater influence on the improvement. Therefore, in actual engineering, if the control requirement of a certain target is higher, the construction operation parameter of the corresponding target can be adjusted preferentially to achieve higher improvement efficiency.

[0137] According to another aspect of the present application, a mixed intelligent method-based excavation tunnel deformation prediction and optimization system is also provided, comprising: a first main module: based on the monitoring data of small-interval excavation tunnel construction parameters, data preprocessing is performed; a second main module: using a Bayesian optimization RF prediction model parameter, hyperparameter optimization and influence parameter importance sorting are performed to obtain a prediction result; a third main module: based on the nonlinear regression mapping relationship between the optimization target and the influence parameters obtained by prediction, an objective optimization function in the NSGA-III model is established, a Pareto frontier solution set is obtained, and multi-objective optimization and prediction of small-interval excavation tunnel construction are realized.

[0138] The method of the embodiment of the present application is realized by relying on an electronic device, and therefore it is necessary to introduce the related electronic device. For this purpose, the embodiment of the present application provides an electronic device, which comprises at least one processor, a communications interface, at least one memory and a communications bus, wherein the at least one processor, the communications interface and the at least one memory complete mutual communication through the communications bus. The at least one processor can call logical instructions in the at least one memory to execute all or part of steps of the method provided by the foregoing various method embodiments.

[0139] In addition, the logic instructions in the at least one memory described above can be implemented in the form of a software functional unit and sold or used as an independent product, and can be stored in a computer readable storage medium. Based on such understanding, the technical solutions of the present application essentially or the parts that contribute to the prior art or parts of the technical solutions can be embodied in the form of a software product. The computer software product is stored in a storage medium, and includes a plurality of instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in the various method embodiments of the present application. The aforementioned storage medium includes: a U disk, a mobile hard disk, a read-only memory (ROM, Read-Only Memory), a random access memory (RAM, Random Access Memory), a magnetic disk or an optical disk, and various media that can store program codes.

[0140] The device embodiments described above are only schematic, wherein the units described as separate components can or can not be physically separated, and the components displayed as units can or can not be physical units, i.e., can be located in one place, or can be distributed on multiple network units. Part or all of the modules can be selected to achieve the purpose of the embodiment according to actual needs. Those skilled in the art can understand and implement without creative labor.

[0141] Through the description of the above embodiments, those skilled in the art can clearly understand that each embodiment can be realized by means of software and the necessary general hardware platform, and of course can also be realized by hardware. Based on such understanding, the above technical solutions essentially or the parts that contribute to the prior art can be embodied in the form of a software product. The computer software product can be stored in a computer readable storage medium, such as a ROM / RAM, a magnetic disk, an optical disk, etc., and includes a plurality of instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0142] Those skilled in the art can easily understand that the above description is only a preferred embodiment of the present application, and is not intended to limit the present application. Any modification, equivalent replacement and improvement made within the spirit and principle of the present application shall be included in the protection scope of the present application.

Claims

1. A method for deformation prediction and optimization of a tunneling based on a hybrid intelligent method, characterized in that, The method comprises the following steps: Step one: based on the monitoring data of small interval tunnel construction parameters, determine the influence parameters, and perform data preprocessing; Step two: use Bayesian optimization RF prediction model parameters, perform hyperparameter optimization and influence parameter importance sorting, and obtain the prediction result; Step three: based on the nonlinear regression mapping relationship between the predicted optimization target and the influence parameters, establish the objective optimization function in the NSGA-III model, obtain the Pareto frontier solution set, and realize the multi-objective optimization of small interval tunnel construction; In step one, the influence parameters include geometric parameters, surrounding rock parameters, blasting parameters and tunneling parameters; According to the above influence parameters, the input index system affecting the ground settlement and the adjacent tunnel settlement deformation is constructed: tunnel buried depth x1, span coverage ratio x2, tunnel center line spacing x3, intermediate stratum thickness x4, surrounding rock strength x5, soil elastic modulus x6, soil cohesion x7, soil internal friction angle x8, soil density x9, peripheral eye spacing x 10 , blast hole density coefficient x 11 , smooth surface layer thickness x 12 , driving speed x 13 , longitudinal spacing of the working face x 14 , soil reinforcement strength x 15 ; In step three, the nonlinear regression mapping relationship between the predicted optimization target and the influence parameters is established to obtain the objective optimization function in the NSGA-III model, and the Pareto frontier solution set is obtained to realize the multi-objective optimization and prediction of small interval tunnel construction, which comprises the following steps: (1) Design scenario analysis Determine the constraint parameters of multi-objective optimization based on the importance and influence law of construction parameters, and design different scenes by adjusting parameters for scenario analysis; (2) Establish the objective optimization function Use the random forest prediction model to obtain the nonlinear function relationship between the ground settlement, adjacent tunnel settlement deformation and soil reinforcement strength, tunneling speed and longitudinal distance of the working face, and obtain the objective function according to the empirical formula, and take the nonlinear function relationship as the fitness function of the NSGA-III algorithm; (3) Set the parameter constraint range Set the constraint range of each parameter, and take the constraint range as the constraint condition of optimization, which is expressed as: b bl <x i <b ul (5) where x i represents the ith input parameter, b bl and b ul represent the lower and upper bounds of the ith design parameter value, respectively. (4) NSGA-III multi-objective optimization Based on the NSGA-III algorithm, the ground settlement, adjacent tunnel settlement deformation, soil reinforcement strength, tunneling speed and longitudinal distance of the working face are optimized; The NSGA-III algorithm is used to optimize the ground settlement, adjacent tunnel settlement deformation, soil reinforcement strength, tunneling speed and longitudinal distance of the working face, which comprises: (1) Set an initial population denoted as P t , randomly generate decision variables according to given variable boundary limits, thereby generating an initial population with N individuals; (2) Select, cross and mutate the first generation population to obtain the first batch of offspring population with the same size as the first generation population, combine the two populations to obtain a double-size population; (3) Non-dominated sorting is performed on the double-size population to determine the non-dominated level of each individual, and the individuals are reserved in descending order until the next generation offspring set is larger than the size of the initial population, and the non-dominated level at this time is recorded; (4) Select the minimum value of each dimension target of the individual in the current population to form the ideal point of the current population: wherein is the minimum value of each objective function, Translate the population to make the ideal point the origin, perform scalarization operation on each objective function, find the extreme point, calculate the intercept on the corresponding coordinate axis, and perform normalization operation; (5) Calculate the reference point on multiple objective functions, and the number of reference points depends on the dimension M of the objective vector and another positive integer H, as shown in the following formula: The number of solutions of the equation can be calculated as follows: Let (x j,1 ,x j,2 ,...,x j,m ) T be the jth solution of the equation, with respect to the reference point λ j : The line connecting the ideal point and the reference point is taken as the reference line L, the distance of each individual in the population to each reference line is calculated, the individual closest to the reference line is linked to the corresponding reference line, assuming u is the projection of f(x) on the reference line L, d j,1 (x) is the distance between the origin and u, d j,2 (x) is the perpendicular distance from f(x) to the line L, and the distance calculation formula is: (6) screening offspring and deleting reference points, each reference point is traversed to find the reference point with the least number of references, i.e. the reference point associated with the least number of population individuals, if the number of population individuals associated with this reference point is zero, find the point with the smallest distance from the reference point vector where individuals are associated to and extract it to the selected next generation population and the reference number is +1; if no individual is referenced to the reference point, delete the reference point vector, if the reference number > 0, select the nearest reference point from it until the population size is N, and obtain the Pareto optimal solution set.

2. The method according to claim 1, wherein, The data preprocessing in step one specifically includes: The data is normalized to eliminate the influence of the characteristic value dimension of different samples on the prediction efficiency and accuracy, and the data samples of different targets are normalized to the interval [-1, 1], and the normalization formula is as follows: where y is a normalized standard value, y max and y min are 1 and -1, x is a sample value, x max and x min are the maximum and minimum values of the sample values. 3.The method of claim 1, wherein, The RF prediction model parameters are optimized by using Bayes in step two, the hyperparameter optimization and the importance ranking of the influence parameters are performed, and the prediction result is obtained, including the following steps: (1) Hyperparameter optimization When training samples by using the RF prediction algorithm, the important parameters in the prediction model need to be tuned first. The parameters that need to be tuned in the RF prediction model mainly include three aspects, i.e. the maximum depth of the decision tree in the random forest model, the maximum number of randomly extracted features at each node, and the maximum iteration number of the weak learner; (2) Feature importance The RF algorithm evaluates the importance of input variables based on the out-of-bag data error, and evaluates the importance of feature variables by measuring the reduction of mean square residual and the reduction of model accuracy after random permutation; (3) Prediction model accuracy evaluation The root mean square error RMSE and the goodness of fit R 2 were introduced to test the accuracy of the models.

4. The method according to claim 3, wherein, The RF algorithm evaluates the importance of input variables based on the out-of-bag data error, and evaluates the importance of feature variables by measuring the reduction of mean square residual and the reduction of model accuracy after random permutation, which specifically includes: (1) Establish a regression decision tree, assuming that the random forest has m decision trees, and use the random forest model to predict the out-of-bag data, which can obtain the mean square error of b out-of-bag data, MSE1, MSE2, …, MSE b The calculation formula is as follows: where y i represents the true value of the dependent variable in the out-of-bag data; represents the predicted value of the regression model, the value of the jth feature variable X j of the out-of-bag data is randomly changed, and the new out-of-bag error accuracy MSE j value is calculated; (2) Generate error matrix, when building regression decision tree for splitting, the feature variable is randomly selected, the variable X j In b bagging data samples, a new OOB test set can be formed, and the established random forest regression model is used to predict the new test set again, and the new OOB residual mean square MSE can be obtained ij The error matrix A is generated as follows: Wherein, P is the number of influence factor variables, and b is the number of training sample sets; (3) Importance score is calculated as the average reduction in MSE1, MSE2,..., MSE b of the corresponding rows of matrix A, divided by the standard error. This gives the importance score VIM of the feature variable X i in terms of the average reduction in the mean squared residual (MSE), which can be expressed as: i MSE where MSE j is the mean squared error for the jth sample; S E is the standard error.

5. A system for performing the mixed intelligent method based deformation prediction and optimization method of tunneling as claimed in any one of claims 1 to 4, characterized in that, It comprises: a first main module: based on the monitoring data of the small-spacing underground excavation tunnel construction parameters, data preprocessing is performed; A second main module, RF prediction model parameters are optimized by using Bayes, hyperparameter optimization and importance ranking of influence parameters are performed, and prediction results are obtained; a third main module, based on the nonlinear regression mapping relationship between the predicted optimization target and the influence parameters, an objective optimization function in the NSGA-III model is established, a Pareto frontier solution set is obtained, and multi-objective optimization and prediction of small-spacing underground excavation tunnel construction are realized.

6. An electronic device, comprising: It comprises: At least one processor, at least one memory and a communication interface; wherein, The processor, memory and communication interface communicate with each other; The memory stores program instructions executable by the processor, and the processor invokes the program instructions to execute the method of any one of claims 1 to 4.

7. A non-transitory computer-readable storage medium, comprising: The non-transitory computer readable storage medium stores computer instructions, and the computer instructions cause the computer to execute the method of any one of claims 1 to 4.

Images