Ultra-small radius curve shallow-buried excavation tunnel construction method based on deep learning

Abstract

The invention discloses an ultra-small radius curve shallow-buried excavation tunnel construction method based on deep learning. The method comprises the steps of establishment of a deep neural network model, collection of training data, training of the deep neural network model, collection of current circulation construction technical parameters, generation of a subsequent circulation construction scheme and real-time continuous decision making of a construction scheme. According to the method, real-time continuous prediction is carried out on a follow-up cyclic construction scheme through current construction technical parameters, the algorithm weight of a model is dynamically adjusted according to actual measurement data of construction, the adaptability and pertinence to the ultra-small radius curve shallow-buried excavation working condition are improved, and the follow-up cyclic construction scheme is continuously generated in real time, so that a basis is provided for construction; by establishing an evaluation index system of asymmetric stress and displacement of the steel arch, a visual and reasonable basis is provided for generation of a steel arch layout and reinforcement scheme, dynamic adjustment of parameters of the steel arch is facilitated to adapt to the characteristics of an ultra-small radius curve shallow-buried excavation tunnel, and intelligence, high efficiency and safety of tunnel construction are improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of underground engineering construction, and relates to a construction method for a shallow-buried and mined tunnel with an ultra-small radius curve based on deep learning. Background Technique

[0002] Urban underground spaces have characteristics such as limited site area, sensitive environment, and complex geological conditions, which result in unconventional structural forms being often adopted for tunnels built in cities. A typical example is the ultra-small radius curve shallow-buried and mined tunnel applied to the cross-passage of subway stations.

[0003] The shallow-buried and mined method is mainly used for the construction of near-surface underground chambers, and its remarkable feature is that the thickness of the overlying rock and soil mass is relatively thin. When passing through complex environments such as densely built-up areas, important structures, or underground pipelines, construction disturbances are likely to cause significant ground surface settlement. Therefore, the control of tunnel deformation should be fully considered when formulating the construction plan.

[0004] Ultra-small radius curve tunnels are usually defined as special tunnel forms with a radius of curvature not greater than 10 times the tunnel excavation width. The large degree of bending of their axes and the characteristics of the non-symmetrical geometric shape of the cross-section will cause non-symmetrical stress redistribution and deformation response of the rock and soil mass around the tunnel after excavation. This difference in stress state further causes non-uniform ground surface settlement and non-symmetrical convergence deformation of the tunnel body, posing a double threat to the safety of ground buildings and structures and the tunnel structure.

[0005] When conventional tunnel construction methods are applied to ultra-small radius curve shallow-buried and mined tunnels, the following problems will be faced: 1) The selection of the tunneling method and the connection of construction processes need to be weighed from two aspects. On the one hand, the disturbance to the surrounding strata should be reduced, and on the other hand, the construction speed should be increased to avoid the overlong exposure time of the surrounding strata after excavation; 2) The optimization and reinforcement of the steel arch support layout. Considering that when the steel arch supports are arranged at equal intervals along the tunnel axis, there will be a spatial distribution characteristic that the spacing is small on the inner side of the curve section and large on the outer side of the curve section, that is, the support strength on the tunnel cross-section is uneven, which may further induce non-uniform settlement on the ground surface and inside the tunnel; 3) The ground traffic control above the tunnel should comprehensively consider the needs of ground traffic and tunnel construction safety; 4) Surface grouting reinforcement and grouting behind the primary support. The characteristics of shallow burial and cross-sectional asymmetry of the tunnel often make it difficult to guarantee the construction quality of the primary support, and problems such as cavities in the overlying rock and soil mass of the tunnel and the mutual separation of the primary support and the strata occur frequently.

[0006] Deep learning algorithms can achieve automatic learning of the features contained in the data and finding the optimal expression of the task, and are suitable for making real-time decisions according to real-time feedback information in various industrial scenarios, especially suitable for tunnel construction operations where practical experience is very rich but theoretical results are relatively scarce. At present, there is no precedent for applying deep learning algorithms to the decision-making of the construction method for ultra-small radius curve shallow-buried and mined tunnels. Summary of the Invention

[0007] Aiming at the deficiencies of the prior art, the present invention proposes a construction method for shallow-buried tunneling of ultra-small radius curves based on deep learning, which is beneficial to solving the deficiencies existing in the application of conventional tunneling construction methods to shallow-buried tunneling of ultra-small radius curves, and realizing the intelligence, scientification and high efficiency of the decision-making of the construction method for shallow-buried tunneling of ultra-small radius curves.

[0008] The present application discloses a construction method for shallow-buried tunneling of ultra-small radius curves based on deep learning, including the following steps:

[0009] S1. Establishment of a deep neural network model: Establish a deep neural network model, including an input layer, several hidden layers and an output layer;

[0010] S2. Collection of training data: Screen the historical construction technical parameters and historical construction plans of shallow-buried tunneling of ultra-small radius curves from the historical data accumulated and stored in past tunnel construction operations;

[0011] S3. Training of the deep neural network model: Use the normalized historical construction technical parameters as input values and the historical construction plans as output values to train the deep neural network model;

[0012] S4. Acquisition of current cycle construction technical parameters: In the current tunnel construction cycle, obtain the current cycle construction technical parameters in real time through the Internet of Things module; both the historical construction technical parameters and the current cycle construction technical parameters include tunnel geometric parameters, surrounding stratum mechanical parameters, tunnel deformation parameters, steel arch mechanical parameters, surrounding stratum cavity parameters and construction rate parameters;

[0013] S5. Generation of subsequent cycle construction plans: Use the normalized current cycle construction technical parameters as input values, and output subsequent cycle construction plans through the deep neural network model; both the historical construction technical parameters and the subsequent cycle construction plans include excavation method plans, steel arch layout plans, steel arch reinforcement plans, ground traffic control plans and grouting reinforcement plans; after completing the current tunnel construction cycle, carry out the operations of the subsequent tunnel construction cycle according to the subsequent cycle construction plan;

[0014] S6. Real-time and continuous decision-making of construction plans: Repeat steps S4 and S5 until the construction of the shallow-buried tunneling of the ultra-small radius curve is completed.

[0015] Preferably, in step S1, the deep neural network model hybridly adopts a fully connected network and a long short-term memory network, which are respectively used to process static parameters and time-series dynamic parameters; the input layer is the first layer of the deep neural network model and is used to receive input data; the output layer is the last layer of the deep neural network model and is used to generate prediction results; the hidden layer is the layer located between the input layer and the output layer and is used for data processing and feature extraction.

[0016] Preferably, for the training of the deep neural network model in step S3, specifically: the historical construction technical parameters are normalized by the Min-Max normalization method, and the data is scaled to the range of [0,1] through linear transformation, and then input into the deep neural network model to calculate the evaluation value of the loss function; the total loss L of the evaluation value of the loss function total is a weighted multi-task loss, as shown in the following expression:

[0017]

[0018] where L i is the loss of the i-th evaluation value, and λ i is the weight of the i-th loss; the weight adjustment adopts the backpropagation algorithm, and the coefficient λ is optimized according to the real-time evaluation value of the loss function i ;

[0019] The evaluation value of the loss function includes the evaluation value of the difference between the model output and the historical construction plan, the evaluation value of the construction rate, the evaluation value of the tunnel deformation parameters, the evaluation value of the mechanical parameters of the steel arch, and the evaluation value of the void parameters of the surrounding strata; the evaluation value of the tunnel deformation parameters includes the evaluation value of the maximum surface settlement, the evaluation value of the maximum surface settlement change rate, the evaluation value of the poor surface settlement on the inside and outside of the curve, the evaluation value of the tunnel crown settlement, and the evaluation value of the tunnel horizontal convergence deformation; the evaluation value of the mechanical parameters of the steel arch includes the evaluation value of the asymmetric stress of the steel arch, the evaluation value of the overall displacement of the steel arch, and the evaluation value of the poor vertical displacement of the steel arch foot; the parameters of the deep neural network model are optimized and updated according to the evaluation value of the loss function to guide the learning process of the model until the evaluation value of the loss function meets the threshold requirements.

[0020] Preferably, in step S4, the Internet of Things module includes displacement sensors, stress sensors, and ground-penetrating radars; the displacement sensors are installed on the surface above the tunnel, the tunnel crown, the tunnel shoulders, and the steel arch; the stress sensors are installed on the steel arch; the ground-penetrating radars are installed on the construction machinery and equipment; the Internet of Things module is connected to the control center system through a wireless network.

[0021] Preferably, in step S5, the excavation method plan is generated by the deep neural network model. The excavation method plan includes the excavation method, the footage of each excavation, the cross-sectional area of each excavation, and the time interval between each excavation. The excavation methods include the full-face excavation method, the bench method, the upper bench reserved core soil method, the middle diaphragm method, the cross middle diaphragm method, and the double-sided drift method. When the tunnel deformation parameters obtained in real time through the Internet of Things module exceed the threshold requirements, the tunnel construction is suspended, the weight of the construction rate evaluation value is reduced, and the weight of the tunnel deformation parameter evaluation value is increased. The deep neural network model is retrained and a new excavation method plan is generated according to steps S3 to S5. Then, the construction is adjusted according to the newly generated excavation method plan.

[0022] Preferably, in step S5, the steel arch support layout plan is generated by the deep neural network model. The steel arch support layout plan includes the steel arch support model, the layout position, and the spacing. The calculation of the asymmetric stress evaluation value I s of the steel arch support needs to satisfy the following expression:

[0023]

[0024] where R is the curvature radius of the tunnel bend, D is the tunnel excavation width, k is the amplification coefficient of the ultra-small radius curve, which is used to quantify the influence brought by the stress difference between the inner and outer sides of the bend; σ i is the maximum stress value of the steel arch support on the inner side of the bend obtained in real time through the Internet of Things module, and σ o is the maximum stress value of the steel arch support on the outer side of the bend obtained in real time through the Internet of Things module. [I s ] is the threshold value of the asymmetric stress evaluation value of the steel arch support. When the asymmetric stress evaluation value of the steel arch support does not meet the threshold requirements, an alarm signal is triggered through the control center system and the tunnel construction is suspended. The weight of the asymmetric stress evaluation value of the steel arch support is increased, and the deep neural network model is retrained and a new steel arch support layout plan is generated according to steps S3 to S5. Then, the construction is adjusted according to the newly generated steel arch support layout plan.

[0025] Preferably, in step S5, the steel arch support reinforcement plan is generated by the deep neural network model. The steel arch support reinforcement plan includes the specifications and quantities of the locking foot anchor pipes, the types and areas of the enlarged foundation at the arch feet, and the specifications and quantities of the longitudinal connecting steel bars of the steel arch support. When the overall displacement of the steel arch support and the vertical displacement difference at the arch feet of the steel arch support obtained in real time through the Internet of Things module do not meet the requirements, a new steel arch support reinforcement plan is generated according to steps S4 to S5. The calculation of the overall displacement evaluation value I x of the steel arch support and the vertical displacement evaluation value I f of the arch feet of the steel arch support satisfies the following expression:

[0026]

[0027] Among them, Δx is the overall displacement of the steel arch, S is the spacing of the steel arch, Δh is the vertical displacement difference at the arch feet of the steel arch, x is the threshold value of the evaluation value of the overall displacement of the steel arch, f is the threshold value of the poor evaluation value of the vertical displacement at the arch feet of the steel arch.

[0028] Preferably, in step S5, the ground traffic control plan and the grouting reinforcement plan are generated by the deep neural network model. The ground traffic control plan includes the control range and duration, and the grouting reinforcement plan includes the type of grouting fluid, grouting pressure, and grouting volume for drilling grouting in the void area of the overlying strata and grouting behind the initial support.

[0029] Compared with the prior art, the beneficial effects of the present invention are as follows: Aiming at the deficiencies of the conventional tunnel construction method when applied to the shallow tunneling method with super-small radius curves, a construction method for shallow tunneling with super-small radius curves based on deep learning is proposed, including the establishment of a deep neural network model, the collection of training data, the training of the deep neural network model, the acquisition of current cycle construction technical parameters, the generation of subsequent cycle construction plans, and the real-time continuous decision-making of construction plans; By using a deep neural network model with a hybrid neural network architecture and a weighted multi-task loss function, the collaborative processing of static parameters and dynamic parameters and multi-objective collaborative optimization are realized, improving the accuracy and rationality of tunnel construction plan decision-making; The subsequent cycle construction plan is predicted in real time and continuously based on the current construction technical parameters, and the algorithm weights of the model are dynamically adjusted according to the measured construction data, improving the adaptability and pertinence of the deep learning algorithm to the working conditions of shallow tunneling with super-small radius curves, and generating the construction plan of the subsequent cycle in real time and continuously, thereby providing a quantitative reference basis for construction technicians; By establishing an evaluation index system for the asymmetric stress and displacement of the steel arch, an intuitive and reasonable basis is provided for the generation of the steel arch layout and reinforcement plan, which is beneficial to dynamically adjusting the steel arch parameters to adapt to the characteristics of shallow tunneling with super-small radius curves, and improving the intelligence, efficiency, and safety of tunnel construction. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] Figure 1 is a flow chart of the construction method for shallow tunneling with super-small radius curves based on deep learning of the present invention;

[0031] Figure 2 is a schematic diagram of the deep neural network model shown in the embodiment of the present invention;

[0032] Figure 3 is a schematic diagram of the regenerated steel arch layout plan shown in the embodiment of the present invention;

[0033] Figure 4A schematic diagram of a regenerated steel arch reinforcement solution shown in an embodiment of the present invention;

[0034] Figure numbers: 11-half-frame I22b steel arch frame, 12-longitudinal I18 I-beam, 13-connecting plate, 14-front and rear two steel arch frames, 15-inner contour line of curve, 16-outer contour line of curve, 17-central axis of tunnel, 21-I10 steel, 22-C25 shotcrete, 23-channel steel, 24-φ60 small duct. DETAILED DESCRIPTION

[0035] The following is a more detailed description of the embodiments of the present invention in conjunction with the accompanying drawings and the accompanying drawings so that a person skilled in the art can implement the embodiments after reading the description. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0036] This application discloses Figures 1-4 The ultra-small radius curve shallow buried dark excavation tunnel construction method shown in the figure includes the following steps:

[0037] S1. Establishment of deep neural network model: Establish a deep neural network model, including the input layer (I1, I2, ... I k ), several hidden layers (h1 [1] ,h2 [1] ,…h n [3] ) and the output layer (O1, O2, ...O k ); the deep neural network model adopts a hybrid of a fully connected network and a long short-term memory network, which are used to process static parameters and time series dynamic parameters respectively; the input layer is the first layer of the deep neural network model, which is used to receive input data; the output layer is the last layer of the deep neural network model, which is used to generate prediction results; the hidden layer is a layer located between the input layer and the output layer, which is used for data processing and feature extraction; the hidden layer is set to 3 layers, and the ReLU activation function is used to process nonlinear features;

[0038] S2. Collection of training data: Filter the historical data accumulated and stored in past tunnel construction operations to obtain historical construction technical parameters and historical construction plans of ultra-small radius curve shallow buried dark excavation tunnels;

[0039] S3. Training of the deep neural network model: Using the normalized historical construction technical parameters as input values and the historical construction plan as output values, train the deep neural network model. Specifically, in implementation, the historical construction technical parameters are normalized using the Min-Max normalization method, scaled to the range [0, 1] through linear transformation, and then input into the deep neural network model to calculate the evaluation value of the loss function. The total loss L of the evaluation value of the loss function total is the weighted multi-task loss, as shown in the following expression:

[0040]

[0041] where L i is the loss of the i-th evaluation value, and λ i is the weight of the i-th loss. The weight adjustment uses the backpropagation algorithm to optimize the coefficient λ according to the real-time evaluation value of the loss function i ;

[0042] The evaluation value of the loss function includes the evaluation value of the difference between the model output and the historical construction plan, the construction rate evaluation value, the tunnel deformation parameter evaluation value, the mechanical parameter evaluation value of the steel arch, and the evaluation value of the surrounding stratum cavity parameter. The tunnel deformation parameter evaluation value includes the evaluation value of the maximum surface settlement, the evaluation value of the maximum surface settlement change rate, the evaluation value of the poor surface settlement on the inside and outside of the curve, the evaluation value of the tunnel crown settlement, and the evaluation value of the tunnel horizontal convergence deformation. The mechanical parameter evaluation value of the steel arch includes the evaluation value of the asymmetric stress of the steel arch, the evaluation value of the overall displacement of the steel arch, and the evaluation value of the poor vertical displacement of the steel arch foot. Optimize and update the parameters of the deep neural network model according to the evaluation value of the loss function, and guide the learning process of the model until the evaluation value of the loss function meets the threshold requirements;

[0043] S4. Acquisition of current cycle construction technical parameters: In the current tunnel construction cycle, obtain the current cycle construction technical parameters in real time through the Internet of Things module. Both the historical construction technical parameters and the current cycle construction technical parameters include tunnel geometric parameters, surrounding stratum mechanical parameters, tunnel deformation parameters, mechanical parameters of the steel arch, surrounding stratum cavity parameters, and construction rate parameters. The Internet of Things module includes displacement sensors, stress sensors, and ground penetrating radar. The displacement sensors are installed on the surface above the tunnel, the tunnel crown, the tunnel shoulders, and the steel arch. The stress sensors are installed on the steel arch. The ground penetrating radar is installed on the construction machinery and equipment. The Internet of Things module is connected to the control center system through a wireless network;

[0044] S5. Generation of subsequent cycle construction plan: Using the normalized current cycle construction technical parameters as input values, and outputting the subsequent cycle construction plan through the deep neural network model; both the historical construction technical parameters and the subsequent cycle construction plan include the excavation method plan, steel arch support layout plan, steel arch support reinforcement plan, ground traffic control plan, and grouting reinforcement plan; after completing the current tunnel construction cycle, carry out the operations of the subsequent tunnel construction cycle according to the subsequent cycle construction plan.

[0045] S6. Real-time and continuous decision-making on construction plan: Repeat steps S4 and S5 until the construction of the extra-small radius curve shallow-buried tunneling is completed.

[0046] In specific implementation, in step S5, the excavation method plan is generated through the deep neural network model. The excavation method plan includes the excavation method, the footage of each excavation, the cross-sectional area of each excavation, and the time interval between each excavation; the excavation methods include the full-face excavation method, the bench method, the upper and lower bench reserved core soil method, the middle diaphragm method, the cross middle diaphragm method, and the double-side drift method; when the tunnel deformation parameters obtained in real time through the Internet of Things module exceed the threshold requirements, suspend the tunnel construction, reduce the weight of the construction rate evaluation value, increase the weight of the tunnel deformation parameter evaluation value, and retrain the deep neural network model and regenerate the excavation method plan according to steps S3 to S5, and then adjust the construction according to the regenerated excavation method plan.

[0047] Under typical working conditions, when monitoring shows that the crown settlement value exceeds the threshold requirements when adopting the upper and lower bench reserved core soil method, suspend the tunnel construction at this time, reduce the weight of the construction rate evaluation value, increase the weight of the tunnel deformation parameter evaluation value, train the deep neural network model according to steps S3 to S5, and the regenerated excavation method plan for the subsequent cycle is the middle diaphragm method. Add I18 steel arch support as the temporary middle diaphragm, divide the tunnel into 4 drifts for construction, and reduce the footage of each excavation, the cross-sectional area of each excavation, and the time interval between each excavation; considering that the stress of the surrounding strata on the inner side of the curve is more complex, the drift on one side of the inner contour line 15 of the curve is excavated first and the steel arch support is constructed.

[0048] In specific implementation, in step S5, the steel arch support layout plan is generated through the deep neural network model, that is, according to the asymmetric stress evaluation value of the steel arch support, combined with the effect of the reinforcement plan in the historical data, output the model, layout position, and spacing of the asymmetrically arranged steel arch support; the asymmetric stress evaluation value I s of the steel arch support needs to satisfy the following expression:

[0049]

[0050] Wherein, R is the radius of curvature of the tunnel bend, D is the tunnel excavation width, and k is the amplification coefficient of the ultra-small radius curve, which is used to quantify the influence brought by the stress difference between the inner and outer sides of the bend; σ i is the maximum stress value of the steel arch at the inner side of the bend obtained in real time through the Internet of Things module, and σ o is the maximum stress value of the steel arch at the outer side of the bend obtained in real time through the Internet of Things module, and [I s is the threshold value of the asymmetric stress evaluation value of the steel arch, which is determined based on the safety range of historical construction data; when the asymmetric stress evaluation value of the steel arch does not meet the threshold requirement, an alarm signal is triggered through the control center system and the tunnel construction is suspended, the weight of the asymmetric stress evaluation value of the steel arch is increased, the deep neural network model is retrained according to steps S3 to S5, and a new steel arch layout plan is generated. Then, the construction is adjusted according to the newly generated steel arch layout plan;

[0051] Under typical working conditions, the radius of curvature R of the tunnel bend is 7.25 m, the tunnel excavation width D is 8.5 m, and the maximum stress value σ o of the steel arch at the outer side of the bend obtained in real time through the Internet of Things module is 152 MPa, and the maximum stress value σ i of the steel arch at the inner side of the bend obtained in real time through the Internet of Things module is 98 MPa. The threshold value [I s of the asymmetric stress evaluation value of the steel arch is 4. Then, the asymmetric stress evaluation value I s of the steel arch is calculated according to Equation (2) as follows:

[0052]

[0053] The asymmetric stress evaluation value of the steel arch does not meet the threshold requirement. An alarm signal is triggered through the control center system and the tunnel construction is suspended. The weight of the asymmetric stress evaluation value of the steel arch is increased. The deep neural network model is retrained according to steps S3 to S5. The newly generated steel arch layout plan is to densify the steel arch at the outer side of the bend, that is, add half a bay of I22b steel arch, so that the spacing of the steel arch on the outer contour line 16 of the bend changes from 1.218 m to 0.609 m; the end of half a bay of I22b steel arch at the tunnel central axis 17 is welded to the front and rear two bays of steel arches 14 through longitudinal I18 steel beams to form an integral body, and half a bay of I22b steel arch 11 is welded to the longitudinal I18 steel beam 12 through a connecting plate 13; after adding half a bay of steel arch, the maximum stress values of the steel arch at the inner and outer sides of the bend obtained in real time through the Internet of Things module are σ i = 94 MPa and σ o = 110 MPa respectively, and the corresponding asymmetric stress evaluation value I s of the steel arch is 3.92, meeting the threshold requirement.

[0054] In specific implementation, in step S5, the steel arch support reinforcement plan is generated by the deep neural network model, that is, according to the overall displacement evaluation value of the steel arch support and the poor evaluation value of the vertical displacement of the arch feet of the steel arch support, combined with the reinforcement plan effect in historical data, the specifications and quantities of the locking foot anchor pipes, the types and areas of the enlarged foundations at the arch feet, and the specifications and quantities of the longitudinal connecting steel bars of the steel arch support are output; when the overall displacement of the steel arch support and the difference in the vertical displacement of the arch feet of the steel arch support obtained in real time through the Internet of Things module do not meet the requirements, the steel arch support reinforcement plan is regenerated according to steps S4 to S5, and the overall displacement evaluation value I of the steel arch support x and the poor evaluation value I of the vertical displacement of the arch feet of the steel arch support f are calculated to satisfy the following expression:

[0055]

[0056] where Δx is the overall displacement of the steel arch support, S is the spacing of the steel arch supports, and Δh is the difference in the vertical displacement of the arch feet of the steel arch support; [I x is the threshold value of the overall displacement evaluation value of the steel arch support, and [I f is the threshold value of the poor evaluation value of the vertical displacement of the arch feet of the steel arch support, and the threshold values are determined based on the safety range of historical construction data;

[0057] Under typical working conditions, the amplification coefficient k for ultra-small radius curves is 1.83, the overall displacement Δx of the steel arch support is 0.020 m, the spacing S of the steel arch supports is 0.609 m, the difference in the vertical displacement Δh of the arch feet of the steel arch support is 0.047 m, the tunnel excavation width D is 8.5 m, and the threshold value [I x of the overall displacement evaluation value of the steel arch support is 0.05, and [I f is the threshold value of the poor evaluation value of the vertical displacement of the arch feet of the steel arch support, which is 0.01. Then, the overall displacement evaluation value I of the steel arch support x and the poor evaluation value I of the vertical displacement of the arch feet of the steel arch support f are calculated according to Equation (4) as follows:

[0058]

[0059] That is, the overall displacement evaluation value I of the steel arch support x and the poor evaluation value I of the vertical displacement of the arch feet of the steel arch support fThe requirements of the threshold are not met. The steel arch support reinforcement plan regenerated according to steps S4 to S5 is as follows: 1) Expand the foundation of the steel arch support at the arch feet on both sides of the upper bench, that is, excavate a triangular area with a width of about 30 cm, fix it with I10 steel bars No. 21 by welding, and fill it densely with C25 shotcrete No. 22; 2) Use channel steel No. 23 as a pad at the arch feet of the steel arch support, and when excavating to 20 cm above the arch feet of the steel arch support, manually and finely clean the rock and soil body to ensure that the channel steel lies on the original formation; 3) Use φ60 small ducts No. 24 for the foot-locking anchor pipes; 4) Add I10 steel bars longitudinally at the arch part and arch feet of each steel arch support, and add a group of inclined φ28 threaded steel bars at the arch feet.

[0060] In specific implementation, in step S5, the ground traffic control plan and the grouting reinforcement plan are generated by the depth neural network model. The ground traffic control plan includes the control range and duration, and the grouting reinforcement plan includes the grouting liquid type, grouting pressure, and grouting volume for drilling grouting in the void area of the overlying formation and grouting behind the initial support.

[0061] It can be seen that through the depth neural network model adopting a hybrid neural network architecture and a weighted multi-task loss function, the collaborative processing of static parameters and dynamic parameters and multi-objective collaborative optimization are realized, improving the accuracy and rationality of tunnel construction plan decision-making; the subsequent cycle construction plan is continuously predicted in real time based on the current construction technical parameters, and the algorithm weights of the model are dynamically adjusted according to the measured construction data, improving the adaptability and pertinence of the deep learning algorithm to the ultra-small radius curve shallow tunneling condition, and continuously generating the construction plan for the subsequent cycle in real time to provide a quantitative reference basis for construction technicians; by establishing an evaluation index system for the asymmetric stress and displacement of the steel arch support, an intuitive and reasonable basis is provided for the layout and reinforcement plan generation of the steel arch support, which is beneficial to dynamically adjusting the parameters of the steel arch support to adapt to the characteristics of the ultra-small radius curve shallow tunneling tunnel, and improving the intelligence, efficiency, and safety of tunnel construction.

[0062] The above are one or more embodiments of the present invention, and the description is relatively specific and detailed, but it should not be construed as a limitation to the scope of the patent of the present invention. It should be noted that for those of ordinary skill in the art, without departing from the concept of the present invention, several deformations and improvements can still be made, and these all belong to the protection scope of the present invention. Therefore, the protection scope of the patent of the present invention should be subject to the appended claims.

Claims

1. A method for constructing shallow buried tunnels with ultra-small radius curves based on deep learning, characterized in that: The following steps are involved: S1. Establishment of deep neural network model: Establish a deep neural network model, including an input layer, several hidden layers and an output layer; S2. Collection of training data: Filter the historical data accumulated and stored in past tunnel construction operations to obtain historical construction technical parameters and historical construction plans of ultra-small radius curved shallow buried dark excavation tunnels; S3, training of a deep neural network model: using the normalized historical construction technical parameters as input values ​​and the historical construction scheme as output values ​​to train the deep neural network model; S4. Collection of current cycle construction technical parameters: In the current tunnel construction cycle, the current cycle construction technical parameters are obtained in real time through the Internet of Things module; the historical construction technical parameters and the current cycle construction technical parameters both include tunnel geometry parameters, surrounding stratum mechanical parameters, tunnel deformation parameters, steel arch mechanical parameters, surrounding stratum cavity parameters and construction rate parameters; S5, generation of subsequent cycle construction plan: using the normalized current cycle construction technical parameters as input values, outputting the subsequent cycle construction plan through the deep neural network model; the historical construction technical parameters and the subsequent cycle construction plan both include an excavation method plan, a steel arch layout plan, a steel arch reinforcement plan, a ground traffic control plan, and a grouting reinforcement plan; after completing the current tunnel construction cycle, carrying out the subsequent tunnel construction cycle operations according to the subsequent cycle construction plan; S6. Real-time and continuous decision-making of construction plans: Repeat steps S4 and S5 until the construction of the ultra-small radius curve shallow buried dark excavation tunnel is completed.

2. The method for constructing a shallow-buried tunnel with an ultra-small radius curve based on deep learning according to claim 1 is characterized in that: In step S1, the deep neural network model adopts a mixture of a fully connected network and a long short-term memory network, which are respectively used to process static parameters and time series dynamic parameters; the input layer is the first layer of the deep neural network model, which is used to receive input data; the output layer is the last layer of the deep neural network model, which is used to generate prediction results; the hidden layer is a layer located between the input layer and the output layer, which is used for data processing and feature extraction.

3. The method for constructing a shallow tunnel with an ultra-small radius curve based on deep learning according to claim 1 is characterized in that: The training of the deep neural network model in step S3 is specifically as follows: the historical construction technical parameters are normalized using the Min-Max standardization method, and then input into the deep neural network model to calculate the loss function evaluation value; the total loss L of the loss function evaluation value total is the weighted multi-task loss, as shown in the following expression: Where L i is the loss of the i-th evaluation value, λ i is the weight of the i-th loss; the weight adjustment adopts the back propagation algorithm, and the coefficient λ is optimized according to the real-time loss function evaluation value i ; The loss function evaluation value includes the evaluation value of the difference between the model output and the historical construction plan, the construction rate evaluation value, the tunnel deformation parameter evaluation value, the steel arch mechanical parameter evaluation value and the surrounding stratum cavity parameter evaluation value; the tunnel deformation parameter evaluation value includes the maximum surface settlement evaluation value, the maximum surface settlement change rate evaluation value, the inner and outer surface settlement difference evaluation value of the curve, the tunnel vault settlement evaluation value and the tunnel horizontal convergence deformation evaluation value; the steel arch mechanical parameter evaluation value includes the steel arch asymmetric stress evaluation value, the steel arch overall displacement evaluation value and the steel arch foot vertical displacement difference evaluation value; the deep neural network model is optimized and updated according to the loss function evaluation value to guide the model learning process until the loss function evaluation value meets the threshold requirement.

4. The method for constructing a shallow-buried tunnel with an ultra-small radius curve based on deep learning according to claim 1 is characterized in that: In step S4, the Internet of Things module includes a displacement sensor, a stress sensor and a geological radar; the displacement sensor is installed on the ground surface above the tunnel, the tunnel vault, the tunnel arch shoulder and the steel arch frame, the stress sensor is installed on the steel arch frame, and the geological radar is installed on the construction machinery and equipment; the Internet of Things module is connected to the control center system through a wireless network.

5. The method for constructing a shallow-buried tunnel with an ultra-small radius curve based on deep learning according to any one of claims 1 to 4, characterized in that: In step S5, an excavation method scheme is generated by a deep neural network model, wherein the excavation method scheme includes an excavation method, a footage of each excavation, a cross-sectional area of ​​each excavation, and a time interval of each excavation; the excavation method includes a full-section excavation method, a step method, an upper step reserved core soil method, a middle partition wall method, a cross middle partition wall method, and a double-side wall pilot pit method; When the tunnel deformation parameters obtained in real time by the Internet of Things module exceed the threshold requirements, the tunnel construction is suspended, the weight of the construction rate evaluation value is reduced, and the weight of the tunnel deformation parameter evaluation value is increased, and the deep neural network model is retrained and the excavation method scheme is regenerated according to steps S3 to S5, and then the construction is adjusted according to the regenerated excavation method scheme.

6. The method for constructing a shallow-buried tunnel with an ultra-small radius curve based on deep learning according to any one of claims 1 to 4, characterized in that: In step S5, a steel arch layout scheme is generated by a deep neural network model, wherein the steel arch layout scheme includes a steel arch model, a layout position and a spacing; the steel arch asymmetric stress evaluation value I s The calculation must satisfy the following expression: Where R is the curvature radius of the tunnel curve, D is the tunnel excavation width, k is the ultra-small radius curve magnification factor; σ i is the maximum stress value of the steel arch frame on the inner part of the curve obtained in real time through the Internet of Things module, σ o is the maximum stress value of the steel arch frame on the outer part of the curve obtained in real time through the Internet of Things module, [I s ] is the threshold of the asymmetric stress evaluation value of the steel arch frame; when the asymmetric stress evaluation value of the steel arch frame does not meet the threshold requirement, the alarm signal is triggered by the control central system and the tunnel construction is suspended, the weight of the asymmetric stress evaluation value of the steel arch frame is increased, and the deep neural network model is re-trained and the steel arch frame layout plan is re-generated according to steps S3 to S5, and then the construction is adjusted according to the re-generated steel arch frame layout plan.

7. The method for constructing a shallow tunnel with an ultra-small radius curve based on deep learning according to any one of claims 1 to 6, characterized in that: In step S5, a steel arch reinforcement scheme is generated by a deep neural network model, wherein the steel arch reinforcement scheme includes specifications and quantity of locking foot anchor pipes, arch foot expansion foundation type and area, and specifications and quantity of longitudinal connecting steel bars of the steel arch; When the overall displacement of the steel arch frame and the vertical displacement difference of the arch foot of the steel arch frame obtained in real time by the Internet of Things module do not meet the requirements, a steel arch frame reinforcement plan is regenerated according to steps S4 to S5; the overall displacement evaluation value of the steel arch frame I x and the vertical displacement difference evaluation value I of the steel arch foot f The calculation satisfies the following expression: Among them, Δx is the overall displacement of the steel arch frame, S is the distance between the steel arch frames, Δh is the vertical displacement difference of the arch foot of the steel arch frame, [I x ] is the threshold of the overall displacement evaluation value of the steel arch frame, [I f ] is the threshold of the vertical displacement difference evaluation value of the arch foot of the steel arch frame.

8. The method for constructing a shallow tunnel with an ultra-small radius curve based on deep learning according to any one of claims 1 to 7, characterized in that: In step S5, a ground traffic control plan and a grouting reinforcement plan are generated through a deep neural network model. The surface traffic control plan includes a control range and duration, and the grouting reinforcement plan includes drilling grouting in the hollow area of ​​the overlying stratum and grouting of the grouting fluid type, grouting pressure and grouting volume for grouting of the voids behind the initial support.

Images