A method, system, terminal and storage medium for predicting deformation of a neighboring tunnel based on a multi-fidelity residual neural network

Abstract

This invention relates to the field of data prediction technology, and discloses a method, system, terminal, and storage medium for predicting the deformation of adjacent tunnels based on a multi-fidelity residual neural network. The method includes: acquiring historical geotechnical physical and mechanical parameters of a historical project and calculating residual information; training a deep neural network based on project control information to obtain an initial residual neural network, and obtaining a multi-fidelity residual replacement model based on the initial residual neural network; training the multi-fidelity residual replacement model based on the historical geotechnical physical and mechanical parameters to obtain a target multi-fidelity residual replacement model; acquiring target working condition information of the target project, inputting it into the target multi-fidelity residual replacement model, and outputting the deformation prediction result of the adjacent tunnel. This invention improves prediction efficiency and accuracy by using a multi-fidelity residual neural network for deformation prediction, thus reducing the risk of deformation of adjacent tunnels.

Description

Technical Field

[0001] This invention relates to the field of data prediction technology, and in particular to a method, system, terminal, and computer-readable storage medium for predicting the deformation of adjacent tunnels based on a multi-fidelity residual neural network. Background Technology

[0002] When excavating deep foundation pits in densely populated urban areas, problems such as the presence of adjacent existing tunnels (e.g., subways, utility tunnels) are often encountered. The unloading and soil displacement caused by excavation can easily lead to excessive settlement, heave, or elliptic deformation of the tunnel, which may threaten the structural safety of the tunnel in severe cases.

[0003] Currently, the processing methods used in existing technologies are as follows: (1) Pre-construction prediction and passive monitoring: Before construction, a one-time prediction is made using numerical methods such as finite element method, and monitoring points are set up during construction. However, the prediction accuracy of this method is limited by the accuracy of the initial soil and rock parameters, and during construction, it can only passively observe whether the deformation exceeds the limit. When the monitoring data alarms, it is often already in a passive response situation. The remedial measures are not only costly, but also ineffective. (2) Extensive control: If the tunnel deformation is found to be close to the warning value during construction, measures such as increasing the prestress of the support and grouting the stratum are usually taken. However, the parameters of these measures (e.g., the size of the support force, the amount of grouting, etc.) rely too much on engineering experience and lack quantitative basis, resulting in insufficient or excessive reinforcement. (3) Disconnect between back analysis and prediction: Existing back analysis methods are mainly used to interpret the deformation that has occurred "after the fact" and calibrate soil parameters, but rarely can the calibrated parameters be used immediately for the optimization of control parameters (e.g., support force) "before the fact" and to guide the next step of construction.

[0004] Therefore, existing technologies still need to be improved and developed. Summary of the Invention

[0005] The main objective of this invention is to provide a method, system, terminal, and storage medium for predicting the deformation of adjacent tunnels based on a multi-fidelity residual neural network. This aims to solve the problems of low prediction efficiency and low accuracy of prediction results in existing adjacent tunnel deformation prediction methods, which lead to a high risk of deformation of adjacent tunnels.

[0006] To achieve the above objectives, this invention provides a method for predicting the deformation of neighboring tunnels based on a multi-fidelity residual neural network. The method includes the following steps: The historical geotechnical physical and mechanical parameters of the historical project are obtained, the historical geotechnical physical and mechanical parameters are calculated to obtain high-fidelity data pairs, and residual information is obtained based on the high-fidelity data pairs; A deep neural network is constructed, and the engineering control information of the historical project is input into the deep neural network. The engineering control information is used to predict the residuals and obtain the predicted residuals. The deep neural network is adjusted according to the predicted residuals and the residual information to obtain an initial residual neural network. A model is constructed according to the initial residual neural network to obtain a multi-fidelity residual substitution model. The historical geotechnical physical and mechanical parameters are input into the multi-fidelity residual replacement model to predict deformation of the historical geotechnical physical and mechanical parameters, and the deformation prediction results are obtained. Based on the deformation prediction results and the deformation monitoring data of the historical project, the multi-fidelity residual replacement model is optimized to obtain the target multi-fidelity residual replacement model. Obtain the target working condition information of the target project, input the target working condition information into the target multi-fidelity residual substitution model, and output the deformation prediction results of the adjacent tunnel.

[0007] Optionally, the method for predicting deformation of adjacent tunnels based on multi-fidelity residual neural networks, wherein obtaining historical geotechnical physical and mechanical parameters of historical engineering projects, performing data calculations on the historical geotechnical physical and mechanical parameters to obtain high-fidelity and low-fidelity data pairs, and obtaining residual information based on the high-fidelity and low-fidelity data pairs, specifically includes: Historical geotechnical physical and mechanical parameters of historical engineering projects are obtained, and low-fidelity and high-fidelity models are constructed. The historical geotechnical physical and mechanical parameters include cohesion, internal friction angle, and elastic modulus. The historical geotechnical physical and mechanical parameters are input into the low-fidelity model to output low-precision response values, and the historical geotechnical physical and mechanical parameters are input into the high-fidelity model to output high-precision response values. The low-precision response value and the high-precision response value are combined to obtain a high- and low-fidelity data pair, and the difference between the high- and low-fidelity data pairs is calculated to obtain residual information.

[0008] Optionally, the method for predicting deformation of adjacent tunnels based on a multi-fidelity residual neural network, wherein constructing a deep neural network involves inputting engineering control information from the historical project into the deep neural network, performing residual prediction on the engineering control information to obtain predicted residuals, adjusting the deep neural network based on the predicted residuals and the residual information to obtain an initial residual neural network, and constructing a model based on the initial residual neural network to obtain a multi-fidelity residual replacement model, specifically including: The engineering control information of historical projects is obtained and a deep neural network is constructed. The engineering control information includes spatial coordinate information, construction time and working condition parameters. The engineering control information is input into the deep neural network. Based on the spatial coordinate information, the construction time, and the working condition parameters, residual fitting of high and low fidelity data is performed to obtain the predicted residual. Error calculation is performed on the predicted residual and the residual information to obtain the mean square error. The parameters of the deep neural network are adjusted according to the mean square error to obtain an initial residual neural network. The initial residual neural network is then combined with the low-fidelity model to obtain a multi-fidelity residual replacement model.

[0009] Optionally, the method for predicting the deformation of adjacent tunnels based on a multi-fidelity residual neural network, wherein the step of inputting the historical geotechnical physical and mechanical parameters into the multi-fidelity residual substitution model to predict the deformation of the historical geotechnical physical and mechanical parameters and obtain the deformation prediction result specifically includes: The historical geotechnical physical and mechanical parameters are input into the multi-fidelity residual substitution model, and the historical geotechnical physical and mechanical parameters are calculated in parallel to obtain preliminary response values ​​and nonlinear compensation values. The preliminary response value and the nonlinear compensation value are summed to obtain the predicted response value. Based on the predicted response value, the deformation of the adjacent tunnel is predicted to obtain the deformation prediction result.

[0010] Optionally, the adjacent tunnel deformation prediction method based on multi-fidelity residual neural network, wherein optimizing the multi-fidelity residual substitution model based on the deformation prediction result and the deformation monitoring data of the historical project to obtain the target multi-fidelity residual substitution model specifically includes: Obtain deformation monitoring data of the historical project, and construct a function based on the deformation monitoring data to obtain the target function; The deformation prediction result is calculated based on the objective function to obtain the error calculation result. The network weights of the multi-fidelity residual substitution model are then optimized based on the error calculation result to obtain the target multi-fidelity residual substitution model.

[0011] Optionally, in the aforementioned method for predicting deformation of adjacent tunnels based on a multi-fidelity residual neural network, the step of calculating the error of the deformation prediction result according to the objective function specifically involves: ; in, The result is the error calculation result. This is the initial response value. This is a nonlinear compensation value. This is deformation monitoring data.

[0012] Optionally, the method for predicting the deformation of adjacent tunnels based on a multi-fidelity residual neural network, wherein the step of obtaining the target working condition information of the target project, inputting the target working condition information into the target multi-fidelity residual substitution model, and outputting the deformation prediction result of the adjacent tunnel, further includes: A risk assessment is performed on the deformation prediction results of the adjacent tunnel to obtain an engineering early warning result. The engineering early warning result is then analyzed to obtain a solution analysis result. Based on the solution analysis result, the solution is confirmed to obtain the target treatment solution.

[0013] Optionally, the adjacent tunnel deformation prediction method based on multi-fidelity residual neural network, wherein the adjacent tunnel deformation prediction system based on multi-fidelity residual neural network includes: The residual calculation module is used to obtain the historical geotechnical physical and mechanical parameters of the historical project, perform data calculation on the historical geotechnical physical and mechanical parameters to obtain high-fidelity data pairs, and obtain residual information based on the high-fidelity data pairs; The model building module is used to build a deep neural network. The engineering control information of the historical project is input into the deep neural network, residual prediction is performed on the engineering control information to obtain the predicted residual, the deep neural network is adjusted according to the predicted residual and the residual information to obtain an initial residual neural network, and a model is built according to the initial residual neural network to obtain a multi-fidelity residual replacement model. The model training module is used to input the historical geotechnical physical and mechanical parameters into the multi-fidelity residual replacement model, perform deformation prediction on the historical geotechnical physical and mechanical parameters, obtain deformation prediction results, and optimize the multi-fidelity residual replacement model based on the deformation prediction results and the deformation monitoring data of the historical project to obtain the target multi-fidelity residual replacement model. The deformation prediction module is used to acquire the target working condition information of the target project, input the target working condition information into the target multi-fidelity residual substitution model, and output the deformation prediction results of the adjacent tunnel.

[0014] Furthermore, to achieve the above objectives, the present invention also provides a terminal, wherein the terminal includes: a memory, a processor, and a neighboring tunnel deformation prediction program based on a multi-fidelity residual neural network stored in the memory and executable on the processor. When the neighboring tunnel deformation prediction program based on the multi-fidelity residual neural network is executed by the processor, it implements the steps of the neighboring tunnel deformation prediction method based on the multi-fidelity residual neural network as described above.

[0015] Furthermore, to achieve the above objectives, the present invention also provides a computer-readable storage medium, wherein the computer-readable storage medium stores a neighboring tunnel deformation prediction program based on a multi-fidelity residual neural network, and when the neighboring tunnel deformation prediction program based on a multi-fidelity residual neural network is executed by a processor, it implements the steps of the neighboring tunnel deformation prediction method based on a multi-fidelity residual neural network as described above.

[0016] In this invention, historical geotechnical physical and mechanical parameters of a historical engineering project are obtained. Data calculations are performed on these parameters to obtain high- and low-fidelity data pairs, and residual information is obtained based on these pairs. A deep neural network is constructed, and engineering control information of the historical engineering project is input into the deep neural network. Residual prediction is performed on the engineering control information to obtain predicted residuals. The deep neural network is adjusted based on the predicted residuals and the residual information to obtain an initial residual neural network. A model is constructed based on the initial residual neural network to obtain a multi-fidelity residual replacement model. The historical geotechnical physical and mechanical parameters are input into the multi-fidelity residual replacement model, and deformation prediction is performed on these parameters to obtain deformation prediction results. The multi-fidelity residual replacement model is optimized based on the deformation prediction results and deformation monitoring data of the historical engineering project to obtain a target multi-fidelity residual replacement model. Target working condition information of the target engineering project is obtained, and this target working condition information is input into the target multi-fidelity residual replacement model to output the deformation prediction results of adjacent tunnels. This invention uses a multi-fidelity residual neural network for deformation prediction, which improves prediction efficiency and accuracy, and reduces the risk of deformation in adjacent tunnels. Attached Figure Description

[0017] Figure 1 This is a flowchart of a preferred embodiment of the adjacent tunnel deformation prediction method based on a multi-fidelity residual neural network of the present invention; Figure 2 This is a schematic diagram of the overall process of the adjacent tunnel deformation prediction method based on multi-fidelity residual neural network of the present invention; Figure 3 This is a schematic diagram of the architecture of the multi-fidelity residual substitution model in a preferred embodiment of the present invention; Figure 4 This is a schematic diagram of iterative calibration of the multi-fidelity residual substitution model in a preferred embodiment of the present invention; Figure 5 This is a structural diagram of a preferred embodiment of the adjacent tunnel deformation prediction system based on a multi-fidelity residual neural network of the present invention; Figure 6 This is a structural diagram of a preferred embodiment of the terminal of the present invention. Detailed Implementation

[0018] To make the objectives, technical solutions, and advantages of this invention clearer and more explicit, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

[0019] It should be noted that if the embodiments of the present invention involve directional indicators (such as up, down, left, right, front, back, etc.), the directional indicators are only used to explain the relative positional relationship and movement of the components in a certain specific posture (as shown in the figure). If the specific posture changes, the directional indicators will also change accordingly.

[0020] Furthermore, if the embodiments of this invention involve descriptions such as "first" or "second," these descriptions are for descriptive purposes only and should not be construed as indicating or implying their relative importance or implicitly specifying the number of technical features indicated. Therefore, a feature defined with "first" or "second" may explicitly or implicitly include at least one of those features. Additionally, the technical solutions of the various embodiments can be combined with each other, but this must be based on the ability of those skilled in the art to implement them. If the combination of technical solutions is contradictory or impossible to implement, it should be considered that such a combination of technical solutions does not exist and is not within the scope of protection claimed by this invention.

[0021] The preferred embodiment of the present invention describes a method for predicting the deformation of neighboring tunnels based on a multi-fidelity residual neural network, such as... Figure 1 As shown, the neighboring tunnel deformation prediction method based on multi-fidelity residual neural network includes the following steps: Step S10: Obtain historical geotechnical physical and mechanical parameters of the historical project, perform data calculation on the historical geotechnical physical and mechanical parameters to obtain high-fidelity data pairs, and obtain residual information based on the high-fidelity data pairs.

[0022] Specifically, existing methods for predicting deformation of adjacent tunnels suffer from low prediction efficiency and low accuracy, leading to a higher risk of deformation in adjacent tunnels. Therefore, this invention proposes a method for predicting deformation of adjacent tunnels based on a multi-fidelity residual neural network to overcome the shortcomings of existing technologies, improve prediction efficiency and accuracy, and reduce the risk of deformation in adjacent tunnels. The core of this invention is to construct a residual substitution model (i.e., a multi-fidelity residual substitution model) consisting of a low-cost low-fidelity (LF) model and a residual neural network (Res_NN) for learning the residuals. Before obtaining the multi-fidelity residual substitution model, corresponding high-fidelity and low-fidelity data pairs need to be obtained. Specifically, after determining the survey area, the historical geotechnical physical and mechanical parameters of historical projects in this survey area are obtained, and a low-fidelity model and a high-fidelity model are constructed. The historical geotechnical physical and mechanical parameters include cohesion, internal friction angle, and elastic modulus. The low-fidelity model is a model with low computational cost and fast response speed, used to capture the main physical trends of geotechnical engineering problems. For example, the low-fidelity model is a simplified analytical solution, empirical formula, and coarse-grid finite difference model. The high-fidelity model is a high-precision and high-cost model. For example, the high-fidelity model is a fine-grid finite element model. Subsequently, the historical soil and rock physical and mechanical parameters are calculated using both the low-fidelity model and the high-fidelity model. Specifically, the historical soil and rock physical and mechanical parameters are input into the low-fidelity model, which outputs low-precision response values, and the historical soil and rock physical and mechanical parameters are input into the high-fidelity model, which outputs high-precision response values. The low-precision response values ​​and the high-precision response values ​​are combined to obtain high- and low-fidelity data pairs. The purpose of this calculation is to obtain the "performance difference" (i.e., residual) of the physical model at different levels of precision. The neural network can then specifically learn the error patterns caused by simplified physical laws or coarse meshes, thereby enabling it to compensate for near-high-precision results with low computational cost in subsequent use. Afterward, the difference between the high- and low-fidelity data pairs is calculated to obtain residual information.

[0023] Step S20: Construct a deep neural network. Input the engineering control information of the historical project into the deep neural network, perform residual prediction on the engineering control information to obtain the predicted residual, adjust the deep neural network according to the predicted residual and the residual information to obtain an initial residual neural network, and construct a model according to the initial residual neural network to obtain a multi-fidelity residual substitution model.

[0024] Specifically, after obtaining the high-fidelity and low-fidelity data pairs, such as Figure 2As shown in step S1, a residual substitution model (i.e., a multi-fidelity residual substitution model) needs to be constructed. The specific construction process is as follows: obtain the engineering control information of historical projects and construct a deep neural network. The engineering control information includes spatial coordinate information, construction time and working condition parameters (e.g., excavation depth, support force, etc.). The engineering control information is input into the deep neural network. The purpose of the input is because this engineering control information is a key control variable that determines the engineering response. Spatial coordinates are used to determine the location, construction steps (i.e., construction time) are used to reflect timeliness, and working condition parameters are used to reflect the stress state. The deep neural network needs this information as independent variables to learn the nonlinear distribution law of residuals under different spatiotemporal and stress conditions. Subsequently, based on the spatial coordinate information, the construction time, and the working condition parameters, residual fitting of high-fidelity and low-fidelity data is performed to obtain the predicted residual. Error calculation is then performed on the predicted residual and the residual information to obtain the mean squared error. The parameters of the deep neural network are adjusted based on the mean squared error to obtain an initial residual neural network. Specifically, the gradient descent algorithm is used to calculate the derivative of the mean squared error with respect to the network weights and biases. This derivative is used to continuously adjust the internal parameters of the deep neural network, enabling it to accurately fit the error of the physical model. Finally, the initial residual neural network is combined with the low-fidelity model to obtain a multi-fidelity residual replacement model.

[0025] Step S30: Input the historical soil and rock physical and mechanical parameters into the multi-fidelity residual replacement model, perform deformation prediction on the historical soil and rock physical and mechanical parameters, obtain the deformation prediction result, and optimize the multi-fidelity residual replacement model based on the deformation prediction result and the deformation monitoring data of the historical project to obtain the target multi-fidelity residual replacement model.

[0026] Specifically, after obtaining the multi-fidelity residual replacement model, the prediction process of the multi-fidelity residual replacement model is as follows: Figure 3 As shown, the historical geotechnical physical and mechanical parameters are input into the multi-fidelity residual replacement model, and parallel calculations are performed on these parameters. Specifically, while calculating the historical geotechnical physical and mechanical parameters using a low-fidelity model, an initial residual neural network is also used to calculate them, yielding preliminary response values ​​and nonlinear compensation values. The preliminary response values ​​and the nonlinear compensation values ​​are summed to obtain predicted response values, and deformation prediction of adjacent tunnels is performed based on these predicted response values ​​to obtain deformation prediction results. For the deformation prediction results, the multi-fidelity residual replacement model needs to be optimized, such as... Figure 2As shown in steps S2 and S3, real-time on-site monitoring data (i.e., deformation monitoring data) is needed as the highest fidelity information to synchronously optimize and fine-tune the historical soil and rock physical and mechanical parameters and the multi-fidelity residual replacement model, thereby achieving model calibration and parameter inversion. The deformation monitoring data is collected in real-time or periodically by sensors deployed at key locations on the engineering site, serving as the highest fidelity information source. This deformation monitoring data includes, but is not limited to, displacement, stress, strain, or pore water pressure. The subsequent calibration process is as follows: Figure 4 As shown, the deformation monitoring data is used as the target with the highest fidelity, and a function is constructed to obtain the objective function. The purpose of constructing the objective function is to minimize the error between the predicted response value of the multi-fidelity residual replacement model and the deformation monitoring data. The error is calculated on the deformation prediction result according to the objective function, and the corresponding expression is: ; in, The result is the error calculation result. This is the initial response value. This is a nonlinear compensation value. The deformation monitoring data is used as the basis for the optimization of the network weights of the multi-fidelity residual substitution model based on the error calculation results, thereby obtaining the target multi-fidelity residual substitution model. The geotechnical physical and mechanical parameters calibrated with actual data are also obtained. The calibrated target multi-fidelity residual substitution model can quickly and accurately predict the deformation status of adjacent tunnels in the future construction stage.

[0027] Step S40: Obtain the target working condition information of the target project, input the target working condition information into the target multi-fidelity residual substitution model, and output the deformation prediction results of the adjacent tunnel.

[0028] Specifically, after obtaining the target multi-fidelity residual replacement model, the deformation status of adjacent tunnels of the target project can be predicted. That is, the target working condition information of the target project is obtained, the target working condition information is input into the target multi-fidelity residual replacement model, and the deformation prediction result of adjacent tunnels is output. Then, it is necessary to confirm whether the target project has ended. If it has ended, the prediction ends. If it has not ended, dynamic cyclic prediction is performed. The newly generated deformation monitoring data is used to continuously fine-tune and evolve the target multi-fidelity residual replacement model, and the response in the more distant future is predicted based on this, until the project ends.

[0029] Furthermore, after obtaining the deformation prediction results of adjacent tunnels, a risk assessment can be performed on the deformation prediction results of adjacent tunnels to obtain engineering early warning results. The risk assessment process involves comparing the deformation prediction results of adjacent tunnels with engineering early warning indicators (e.g., tunnel settlement threshold ±10mm). If the deformation prediction results of adjacent tunnels are close to or exceed the engineering early warning indicators, a corresponding early warning is triggered. Subsequently, a scheme analysis is performed on the engineering early warning results to obtain scheme analysis results. Based on the scheme analysis results, the scheme is confirmed to obtain the target treatment scheme (e.g., increasing the prestress of the support, changing the grouting volume, slowing down the excavation speed, etc.). By comparing the predicted responses under different schemes, the scheme that can keep the deformation prediction results of adjacent tunnels within a safe range and has the best cost is selected as the construction guidance suggestion.

[0030] Furthermore, such as Figure 5 As shown, based on the above-mentioned method for predicting the deformation of adjacent tunnels using a multi-fidelity residual neural network, this invention also provides a system for predicting the deformation of adjacent tunnels using a multi-fidelity residual neural network, wherein the system for predicting the deformation of adjacent tunnels using a multi-fidelity residual neural network includes: The residual calculation module 51 is used to obtain the historical geotechnical physical and mechanical parameters of the historical project, perform data calculation on the historical geotechnical physical and mechanical parameters to obtain high and low fidelity data pairs, and obtain residual information based on the high and low fidelity data pairs. The model building module 52 is used to build a deep neural network. The engineering control information of the historical project is input into the deep neural network, the residual prediction of the engineering control information is performed to obtain the predicted residual, the deep neural network is adjusted according to the predicted residual and the residual information to obtain an initial residual neural network, and the model is built according to the initial residual neural network to obtain a multi-fidelity residual replacement model. The model training module 53 is used to input the historical soil and rock physical and mechanical parameters into the multi-fidelity residual replacement model, perform deformation prediction on the historical soil and rock physical and mechanical parameters, obtain deformation prediction results, and optimize the multi-fidelity residual replacement model based on the deformation prediction results and the deformation monitoring data of the historical project to obtain the target multi-fidelity residual replacement model. The deformation prediction module 54 is used to obtain the target working condition information of the target project, input the target working condition information into the target multi-fidelity residual substitution model, and output the deformation prediction result of the adjacent tunnel.

[0031] Furthermore, such as Figure 6 As shown, based on the above-mentioned method for predicting the deformation of adjacent tunnels based on multi-fidelity residual neural networks, the present invention also provides a terminal, which includes a processor 10, a memory 20 and a display 30. Figure 6 Only some of the terminal components are shown; however, it should be understood that it is not required to implement all of the components shown, and more or fewer components may be implemented instead.

[0032] In some embodiments, the memory 20 may be an internal storage unit of the terminal, such as a hard disk or memory. In other embodiments, the memory 20 may be an external storage device of the terminal, such as a plug-in hard disk, smart media card (SMC), secure digital card (SD), flash card, etc. Further, the memory 20 may include both internal and external storage devices. The memory 20 is used to store application software and various types of data installed on the terminal, such as program code installed on the terminal. The memory 20 can also be used to temporarily store data that has been output or will be output. In one embodiment, the memory 20 stores a neighboring tunnel deformation prediction program 40 based on a multi-fidelity residual neural network. This neighboring tunnel deformation prediction program 40 based on a multi-fidelity residual neural network can be executed by the processor 10, thereby implementing the neighboring tunnel deformation prediction method based on a multi-fidelity residual neural network in this application.

[0033] In some embodiments, the processor 10 may be a central processing unit (CPU), a microprocessor, or other data processing chip, used to run program code stored in the memory 20 or process data, such as executing the adjacent tunnel deformation prediction method based on multi-fidelity residual neural network.

[0034] In some embodiments, the display 30 may be an LED display, a liquid crystal display, a touch-sensitive liquid crystal display, or an OLED (Organic Light-Emitting Diode) touchscreen. The display 30 is used to display information on the terminal and to display a visual user interface.

[0035] In one embodiment, when the processor 10 executes the neighbor tunnel deformation prediction program 40 based on a multi-fidelity residual neural network in the memory 20, the following steps are performed: The historical geotechnical physical and mechanical parameters of the historical project are obtained, the historical geotechnical physical and mechanical parameters are calculated to obtain high-fidelity data pairs, and residual information is obtained based on the high-fidelity data pairs; A deep neural network is constructed, and the engineering control information of the historical project is input into the deep neural network. The engineering control information is used to predict the residuals and obtain the predicted residuals. The deep neural network is adjusted according to the predicted residuals and the residual information to obtain an initial residual neural network. A model is constructed according to the initial residual neural network to obtain a multi-fidelity residual substitution model. The historical geotechnical physical and mechanical parameters are input into the multi-fidelity residual replacement model to predict deformation of the historical geotechnical physical and mechanical parameters, and the deformation prediction results are obtained. Based on the deformation prediction results and the deformation monitoring data of the historical project, the multi-fidelity residual replacement model is optimized to obtain the target multi-fidelity residual replacement model. Obtain the target working condition information of the target project, input the target working condition information into the target multi-fidelity residual substitution model, and output the deformation prediction results of the adjacent tunnel.

[0036] The process of acquiring historical geotechnical physical and mechanical parameters of historical engineering projects, performing data calculations on these parameters to obtain high- and low-fidelity data pairs, and obtaining residual information based on these high- and low-fidelity data pairs specifically includes: Historical geotechnical physical and mechanical parameters of historical engineering projects are obtained, and low-fidelity and high-fidelity models are constructed. The historical geotechnical physical and mechanical parameters include cohesion, internal friction angle, and elastic modulus. The historical geotechnical physical and mechanical parameters are input into the low-fidelity model to output low-precision response values, and the historical geotechnical physical and mechanical parameters are input into the high-fidelity model to output high-precision response values. The low-precision response value and the high-precision response value are combined to obtain a high- and low-fidelity data pair, and the difference between the high- and low-fidelity data pairs is calculated to obtain residual information.

[0037] The construction of the deep neural network involves inputting the engineering control information of the historical project into the deep neural network, performing residual prediction on the engineering control information to obtain predicted residuals, adjusting the deep neural network based on the predicted residuals and the residual information to obtain an initial residual neural network, and constructing a model based on the initial residual neural network to obtain a multi-fidelity residual substitution model. Specifically, this includes: The engineering control information of historical projects is obtained and a deep neural network is constructed. The engineering control information includes spatial coordinate information, construction time and working condition parameters. The engineering control information is input into the deep neural network. Based on the spatial coordinate information, the construction time, and the working condition parameters, residual fitting of high and low fidelity data is performed to obtain the predicted residual. Error calculation is performed on the predicted residual and the residual information to obtain the mean square error. The parameters of the deep neural network are adjusted according to the mean square error to obtain an initial residual neural network. The initial residual neural network is then combined with the low-fidelity model to obtain a multi-fidelity residual replacement model.

[0038] Specifically, the step of inputting the historical geotechnical physical and mechanical parameters into the multi-fidelity residual substitution model to predict the deformation of the historical geotechnical physical and mechanical parameters and obtain the deformation prediction result includes: The historical geotechnical physical and mechanical parameters are input into the multi-fidelity residual substitution model, and the historical geotechnical physical and mechanical parameters are calculated in parallel to obtain preliminary response values ​​and nonlinear compensation values. The preliminary response value and the nonlinear compensation value are summed to obtain the predicted response value. Based on the predicted response value, the deformation of the adjacent tunnel is predicted to obtain the deformation prediction result.

[0039] Specifically, the optimization of the multi-fidelity residual replacement model based on the deformation prediction results and the deformation monitoring data of the historical project to obtain the target multi-fidelity residual replacement model includes: Obtain deformation monitoring data of the historical project, and construct a function based on the deformation monitoring data to obtain the target function; The deformation prediction result is calculated based on the objective function to obtain the error calculation result. The network weights of the multi-fidelity residual substitution model are then optimized based on the error calculation result to obtain the target multi-fidelity residual substitution model.

[0040] Specifically, the step of calculating the error of the deformation prediction result based on the objective function includes: ; in, The result is the error calculation result. This is the initial response value. This is a nonlinear compensation value. This is deformation monitoring data.

[0041] The process of acquiring target working condition information of the target project, inputting the target working condition information into the target multi-fidelity residual substitution model, and outputting the deformation prediction results of adjacent tunnels further includes: A risk assessment is performed on the deformation prediction results of the adjacent tunnel to obtain an engineering early warning result. The engineering early warning result is then analyzed to obtain a solution analysis result. Based on the solution analysis result, the solution is confirmed to obtain the target treatment solution.

[0042] The present invention also provides a computer-readable storage medium, wherein the computer-readable storage medium stores a neighboring tunnel deformation prediction program based on a multi-fidelity residual neural network, and the neighboring tunnel deformation prediction program based on the multi-fidelity residual neural network, when executed by a processor, implements the steps of the neighboring tunnel deformation prediction method based on the multi-fidelity residual neural network as described above.

[0043] In summary, this invention provides a method, system, terminal, and storage medium for predicting deformation of adjacent tunnels based on a multi-fidelity residual neural network. The method includes: acquiring historical geotechnical physical and mechanical parameters of a historical engineering project; performing data calculations on the historical geotechnical physical and mechanical parameters to obtain high-fidelity and low-fidelity data pairs; and obtaining residual information based on the high-fidelity and low-fidelity data pairs; constructing a deep neural network; inputting the engineering control information of the historical engineering project into the deep neural network; performing residual prediction on the engineering control information to obtain predicted residuals; and adjusting the deep neural network based on the predicted residuals and the residual information to obtain... An initial residual neural network is used, and a multi-fidelity residual substitution model is obtained based on the initial residual neural network. Historical geotechnical physical and mechanical parameters are input into the multi-fidelity residual substitution model to predict deformation based on these parameters. The deformation prediction results are then used to optimize the multi-fidelity residual substitution model based on the deformation prediction results and deformation monitoring data from the historical project, resulting in a target multi-fidelity residual substitution model. Target working condition information of the target project is obtained and input into the target multi-fidelity residual substitution model, outputting the deformation prediction results of adjacent tunnels. This invention improves prediction efficiency and accuracy by using a multi-fidelity residual neural network for deformation prediction, thus reducing the deformation risk of adjacent tunnels.

[0044] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element.

[0045] Of course, those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware (such as a processor, controller, etc.). The program can be stored in a computer-readable storage medium, and when executed, it can include the processes described in the above method embodiments. The computer-readable storage medium can be a memory, magnetic disk, optical disk, etc.

[0046] It should be understood that the application of the present invention is not limited to the examples above. Those skilled in the art can make improvements or modifications based on the above description, and all such improvements and modifications should fall within the protection scope of the appended claims.

Claims

1. A method for predicting the deformation of adjacent tunnels based on a multi-fidelity residual neural network, characterized in that, The neighboring tunnel deformation prediction method based on multi-fidelity residual neural network includes: The historical geotechnical physical and mechanical parameters of the historical project are obtained, the historical geotechnical physical and mechanical parameters are calculated to obtain high-fidelity data pairs, and residual information is obtained based on the high-fidelity data pairs; A deep neural network is constructed, and the engineering control information of the historical project is input into the deep neural network. The engineering control information is used to predict the residuals and obtain the predicted residuals. The deep neural network is adjusted according to the predicted residuals and the residual information to obtain an initial residual neural network. A model is constructed according to the initial residual neural network to obtain a multi-fidelity residual substitution model. The historical geotechnical physical and mechanical parameters are input into the multi-fidelity residual replacement model to predict deformation of the historical geotechnical physical and mechanical parameters, and the deformation prediction results are obtained. Based on the deformation prediction results and the deformation monitoring data of the historical project, the multi-fidelity residual replacement model is optimized to obtain the target multi-fidelity residual replacement model. Obtain the target working condition information of the target project, input the target working condition information into the target multi-fidelity residual substitution model, and output the deformation prediction results of the adjacent tunnel.

2. The method for predicting the deformation of adjacent tunnels based on a multi-fidelity residual neural network according to claim 1, characterized in that, The process of acquiring historical geotechnical physical and mechanical parameters of historical engineering projects, performing data calculations on these parameters to obtain high- and low-fidelity data pairs, and obtaining residual information based on these high- and low-fidelity data pairs specifically includes: Historical geotechnical physical and mechanical parameters of historical engineering projects are obtained, and low-fidelity and high-fidelity models are constructed. The historical geotechnical physical and mechanical parameters include cohesion, internal friction angle, and elastic modulus. The historical geotechnical physical and mechanical parameters are input into the low-fidelity model to output low-precision response values, and the historical geotechnical physical and mechanical parameters are input into the high-fidelity model to output high-precision response values. The low-precision response value and the high-precision response value are combined to obtain a high- and low-fidelity data pair, and the difference between the high- and low-fidelity data pairs is calculated to obtain residual information.

3. The method for predicting the deformation of adjacent tunnels based on a multi-fidelity residual neural network according to claim 2, characterized in that, The construction of the deep neural network involves inputting the engineering control information of the historical project into the deep neural network, performing residual prediction on the engineering control information to obtain predicted residuals, adjusting the deep neural network based on the predicted residuals and the residual information to obtain an initial residual neural network, and constructing a model based on the initial residual neural network to obtain a multi-fidelity residual substitution model, specifically including: Obtain engineering control information from historical projects and construct a deep neural network, wherein the engineering control information includes spatial coordinate information, construction time and working condition parameters; The engineering control information is input into the deep neural network. Based on the spatial coordinate information, the construction time, and the working condition parameters, residual fitting of high and low fidelity data is performed to obtain the predicted residual. Error calculation is performed on the predicted residual and the residual information to obtain the mean square error. The parameters of the deep neural network are adjusted according to the mean square error to obtain an initial residual neural network. The initial residual neural network is then combined with the low-fidelity model to obtain a multi-fidelity residual replacement model.

4. The method for predicting the deformation of adjacent tunnels based on a multi-fidelity residual neural network according to claim 1, characterized in that, The step of inputting the historical geotechnical physical and mechanical parameters into the multi-fidelity residual substitution model to predict the deformation of the historical geotechnical physical and mechanical parameters and obtain the deformation prediction result specifically includes: The historical geotechnical physical and mechanical parameters are input into the multi-fidelity residual substitution model, and the historical geotechnical physical and mechanical parameters are calculated in parallel to obtain preliminary response values ​​and nonlinear compensation values. The preliminary response value and the nonlinear compensation value are summed to obtain the predicted response value. Based on the predicted response value, the deformation of the adjacent tunnel is predicted to obtain the deformation prediction result.

5. The method for predicting the deformation of adjacent tunnels based on a multi-fidelity residual neural network according to claim 1, characterized in that, The optimization of the multi-fidelity residual replacement model based on the deformation prediction results and the deformation monitoring data of the historical project, to obtain the target multi-fidelity residual replacement model, specifically includes: Obtain deformation monitoring data of the historical project, and construct a function based on the deformation monitoring data to obtain the target function; The deformation prediction result is calculated based on the objective function to obtain the error calculation result. The network weights of the multi-fidelity residual substitution model are then optimized based on the error calculation result to obtain the target multi-fidelity residual substitution model.

6. The method for predicting the deformation of adjacent tunnels based on a multi-fidelity residual neural network according to claim 5, characterized in that, The step of calculating the error of the deformation prediction result based on the objective function is specifically as follows: ； in, The result is the error calculation result. This is the initial response value. This is a nonlinear compensation value. This is deformation monitoring data.

7. The method for predicting the deformation of adjacent tunnels based on a multi-fidelity residual neural network according to claim 1, characterized in that, The process of obtaining target working condition information for the target project, inputting the target working condition information into the target multi-fidelity residual substitution model, and outputting the deformation prediction results of adjacent tunnels further includes: A risk assessment is performed on the deformation prediction results of the adjacent tunnel to obtain an engineering early warning result. The engineering early warning result is then analyzed to obtain a solution analysis result. Based on the solution analysis result, the solution is confirmed to obtain the target treatment solution.

8. A neighboring tunnel deformation prediction system based on a multi-fidelity residual neural network, characterized in that, The adjacent tunnel deformation prediction system based on multi-fidelity residual neural network includes: The residual calculation module is used to obtain the historical geotechnical physical and mechanical parameters of the historical project, perform data calculation on the historical geotechnical physical and mechanical parameters to obtain high-fidelity data pairs, and obtain residual information based on the high-fidelity data pairs; The model building module is used to build a deep neural network. The engineering control information of the historical project is input into the deep neural network, residual prediction is performed on the engineering control information to obtain the predicted residual, the deep neural network is adjusted according to the predicted residual and the residual information to obtain an initial residual neural network, and a model is built according to the initial residual neural network to obtain a multi-fidelity residual replacement model. The model training module is used to input the historical geotechnical physical and mechanical parameters into the multi-fidelity residual replacement model, perform deformation prediction on the historical geotechnical physical and mechanical parameters, obtain deformation prediction results, and optimize the multi-fidelity residual replacement model based on the deformation prediction results and the deformation monitoring data of the historical project to obtain the target multi-fidelity residual replacement model. The deformation prediction module is used to acquire the target working condition information of the target project, input the target working condition information into the target multi-fidelity residual substitution model, and output the deformation prediction results of the adjacent tunnel.

9. A terminal, characterized in that, The terminal includes a memory, a processor, and a program stored in the memory and executable on the processor. When the program is executed by the processor, it implements the steps of the adjacent tunnel deformation prediction method based on a multi-fidelity residual neural network as described in any one of claims 1-7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program thereon, and the computer-readable storage medium stores a neighboring tunnel deformation prediction program based on a multi-fidelity residual neural network. When the neighboring tunnel deformation prediction program based on the multi-fidelity residual neural network is executed by a processor, it implements the steps of the neighboring tunnel deformation prediction method based on any one of claims 1-7.

Images