Intelligent evaluation system, device and method for stability of steeply inclined slope section tunnel

Abstract

The invention discloses an intelligent evaluation system, device and method for stability of a steeply inclined slope section tunnel, and belongs to the technical field of tunnel engineering monitoring. The system is characterized in that a multi-source information acquisition module is arranged in a differentiated manner according to the vault, the side wall and the slope toe of the steeply inclined slope section; the evaluation and analysis module is used for training based on a fusion training set formed by numerical simulation data and real-time monitoring data; the decision control module is used for performing simulation pre-deduction before issuing the physical regulation and control instruction; and the cloud platform module dynamically updates the evaluation model parameters according to the effect feedback data after the maintenance or intervention equipment executes the instruction. A sensing-analysis-decision-execution-feedback full-closed-loop intelligent regulation and control system is constructed, the problems that in the prior art, evaluation precision is insufficient, response lags behind and self-adaptive learning ability is lacked are solved, and high-precision evaluation and high-safety automatic control of the stability of the steeply inclined slope section tunnel and self-adaptive evolution of the system are achieved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of tunnel engineering monitoring, in particular to a system, device and method for intelligently evaluating and automatically closed-loop controlling the stability of a tunnel in a steep slope section with complex geological conditions. BACKGROUND

[0002] In the field of tunnel engineering technology, especially in areas with complex geological structures and significant topographic elevation differences, ensuring the long-term stability of steep slope section tunnels has always been a core technical problem throughout the entire cycle of surveying, design, construction and operation. Traditional stability evaluation methods largely rely on the experience of engineering and technical personnel, static geological survey reports, and simplified mechanical models based on numerous assumptions. While these methods have some guiding significance under certain conditions, their inherent limitations have become increasingly apparent, such as the difficulty in fully capturing and quantifying the complexity and dynamic evolution process of the tunnel surrounding rock system, resulting in significant deviations in evaluation results and an inability to provide sufficient basis for fine and dynamic construction and support decisions.

[0003] With the rapid development of sensor technology, computer simulation and artificial intelligence, modern tunnel stability evaluation has gradually evolved towards intelligence and digitization. Currently, comprehensive evaluation systems integrating multi-source information monitoring, three-dimensional geological modeling and intelligent algorithm analysis have emerged. These systems obtain real-time multi-dimensional data such as surrounding rock deformation and stress by deploying sensors, and use machine learning or deep learning models to predict and classify the stability state of the tunnel. Compared to traditional methods, significant progress has been made in the automation and objectivity of evaluation.

[0004] However, a deep analysis of existing technical solutions in this field reveals several technical bottlenecks that need to be addressed. First, the training data for existing intelligent evaluation models is relatively single, mainly relying on historical measured monitoring data. When the project faces new working conditions or rare geological disasters that lack similar cases, the generalization ability and prediction accuracy of the model will be greatly limited, and the theoretical data provided by numerical simulation technology cannot effectively make up for the lack of real data. Second, most current systems still function in an open or semi-open loop mode of "monitoring-analysis-warning", with the final output being a risk level or decision recommendation for human reference, lacking the closed-loop capability of directly converting evaluation results into physical control instructions and automatically executing them, resulting in a long emergency response chain and intervention lag. More importantly, even if automatic control is achieved, the decision-making process often lacks necessary safety verification mechanisms.

[0005] In recent years, although there have been some developments in combining "digital twin" and artificial intelligence, there are still limitations. These solutions usually use digital models to visualize and predict trends of tunnel status, but in the aspect of automatic control, the decision instructions are mostly generated and executed directly based on static rule bases. This is essentially an "open-loop" execution, that is, once the instructions are generated, they are implemented, lacking a forward-looking safety check link that dynamically predicts the complex physical consequences that may be caused by the instructions (for example, whether drainage at a certain place will cause stress concentration at another place), which poses a risk of secondary disasters.

[0006] In a complex tunnel environment, unverified automated instructions may pose a potential risk of secondary disasters. In addition, the built-in evaluation models and control strategies of these systems are usually static after deployment and cannot learn and optimize themselves according to new data and real feedback accumulated during subsequent operation, lacking the ability to adapt to dynamic changes in the environment.

[0007] Therefore, there is an urgent need to develop a tunnel stability intelligent management system that can overcome the above-mentioned defects and achieve high-precision evaluation, high-safety closed-loop control, and self-adaptive learning. SUMMARY

[0008] The purpose of the present application is to solve the technical problems of insufficient generalization ability of evaluation models, lack of automated closed-loop physical control, safety risks in decision-making process, and inability of systems to adaptively learn and evolve in the prior art.

[0009] To achieve the above-mentioned purpose, the first aspect of the present application provides an intelligent evaluation system for the stability of a steep slope section tunnel, comprising: a multi-source information acquisition module, the multi-source information acquisition module comprising a plurality of sensors differentially arranged according to the vault, side wall and slope foot of the steep slope section; a data preprocessing module connected with the multi-source information acquisition module, for receiving and processing multi-dimensional information collected by the plurality of sensors; an evaluation analysis module connected with the data preprocessing module, the evaluation analysis module comprising a three-dimensional modeling unit, a numerical simulation unit and an intelligent evaluation unit, the intelligent evaluation unit being trained based on a fusion training set, the fusion training set being composed of numerical simulation data generated by the numerical simulation unit and real-time monitoring data processed by the data preprocessing module; a decision control module connected with the evaluation analysis module, for receiving the stability evaluation result output by the intelligent evaluation unit and inputting the physical regulation and control instruction generated based on the result to the numerical simulation unit for simulation and pre-inference, and after the simulation and pre-inference passes, issuing the physical regulation and control instruction to the maintenance or intervention equipment in the tunnel. a cloud platform module, connected with the decision control module and the evaluation analysis module respectively, configured to receive effect feedback data after the maintenance or intervention device executes the physical control instruction, and update model parameters of the intelligent evaluation unit according to the effect feedback data; and a warning and response module, connected with the multi-source information collection module.

[0010] Preferably, the multi-source information collection module is further configured to adjust the collection frequency of the plurality of sensors according to the risk level output by the evaluation analysis module.

[0011] Preferably, the input feature vector of the intelligent evaluation unit comprises a slope angle parameter, a lateral stress concentration parameter, and a rainfall infiltration model parameter.

[0012] Preferably, the intelligent evaluation unit is a neural network model.

[0013] Preferably, the cloud platform module is further configured to update the control strategy of the decision control module according to the effect feedback data.

[0014] Preferably, the maintenance or intervention device comprises a reinforcement device, a drainage facility, and a ventilation system.

[0015] In a second aspect, the present application provides an intelligent evaluation device for the stability of a steeply inclined slope section tunnel, comprising the system of the first aspect, and hardware devices and auxiliary facilities required for installing and fixing the system.

[0016] In a third aspect, the present application further provides an intelligent evaluation method for the stability of a steeply inclined slope section tunnel, comprising the following steps: S1. Collecting multi-dimensional information of the steeply inclined slope section tunnel through a multi-source information collection module, wherein a plurality of sensors in the multi-source information collection module are differentially arranged according to the tunnel crown, sidewall, and slope toe; S2. Processing the multi-dimensional information through a data preprocessing module; S3. Evaluating the stability of the processed information through an evaluation analysis module, wherein the evaluation step comprises: training an intelligent evaluation unit based on a fusion training set composed of numerical simulation data and real-time monitoring data, and evaluating using the intelligent evaluation unit; S4. Generating a physical control instruction through a decision control module, and simulating and pre-deriving the instruction in a three-dimensional model before issuing the physical control instruction; S5. Receiving effect feedback data after the physical control instruction is executed through a cloud platform module, and updating model parameters of the intelligent evaluation unit according to the effect feedback data.

[0017] Preferably, in step S3, the input feature vector of the intelligent evaluation unit contains the slope angle parameter, the lateral stress concentration degree parameter, and the rainfall infiltration model parameter.

[0018] Preferably, the method further comprises a step S11 of dynamically adjusting the acquisition frequency of the plurality of sensors according to the risk level output in step S3.

[0019] Compared with the prior art, the present application uses a fusion training set composed of numerical simulation data and real-time monitoring data to construct an intelligent evaluation model, and introduces specific risk factors of steep slope sections, which fundamentally solves the problem of insufficient generalization ability of traditional models due to the dependence on a single data source, greatly improves the accuracy and reliability of the evaluation results. On the basis of high-precision evaluation, the present application further introduces a "simulation pre-deployment" link to ensure that each automated physical control instruction has undergone strict virtual safety verification before being issued for execution, thereby realizing high-security automated closed-loop control and effectively avoiding the risks that may be caused by blind decision-making. The present application is not a static tool system, but a self-adapting iterative mechanism that can continuously learn from practice. It receives real feedback of instruction execution through a cloud platform, dynamically updates the evaluation model and control strategy, and endows the entire system with the ability of self-evolution. In addition, the intelligent framework can dynamically adjust the acquisition frequency of the sensors according to the risk level, realizing intelligent on-demand allocation of monitoring resources, ensuring timely response while taking into account the economy of operation. BRIEF DESCRIPTION OF DRAWINGS

[0020] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the drawings needed to be used in the embodiments or prior art description will be briefly introduced as follows.

[0021] Figure 1 is a schematic diagram of the overall architecture of the intelligent evaluation system for the stability of steep slope section tunnels provided by the embodiments of the present application; Figure 2 is a schematic diagram of the differentiated arrangement of sensors in the multi-source information acquisition module in the embodiments of the present application; Figure 3 is a schematic diagram of the working process of the data preprocessing module in the embodiments of the present application; Figure 4 is a schematic diagram of the intelligent evaluation method flow for the stability of steep slope section tunnels provided by the embodiments of the present application; Figure 5 is a schematic diagram of the user interface of the cloud platform module in the embodiments of the present application; In the drawings: 0, multi-source information acquisition module; 1, data preprocessing module; 2, evaluation analysis module; 3, decision control module; 4, cloud platform module; 5, early warning and response module; 01, geological radar; 02, displacement sensor; 03, stress meter; 04, vibration sensor; 05, temperature and humidity sensor; 06, gas concentration sensor; 1-0, tunnel; 1-2, displacement sensor layout point; 1-4, three-dimensional laser scanner; 1-5, surrounding rock geological body. DETAILED DESCRIPTION

[0022] In order to make the purpose, technical scheme and advantages of the present application clearer, more complete, the following will combine the drawings and specific embodiments to describe the present application in detail. It should be understood that the specific embodiments described herein are only used to explain the present application, and do not constitute any form of limitation on the present application. Any modification, equivalent replacement, improvement, etc. made within the spirit and principles of the present application shall be included in the protection scope of the present application.

[0023] Example one: The present embodiment provides an intelligent evaluation system for the stability of a steeply inclined slope section tunnel. The system is designed at the beginning to overcome a series of fundamental challenges faced by the prior art in dealing with such complex geological working conditions, including but not limited to the limited generalization ability of the evaluation model, the lack of automatic control closed loop, the lack of forward-looking safety check in the decision-making process, and the lack of self-adaptive evolution ability of the system as a whole to learn from experience. As shown in Figure 1 To achieve this goal, the present system builds an integrated technical architecture consisting of six core functional modules. These six modules are: the multi-source information acquisition module 0 as the perception front end of the system, the data preprocessing module 1 responsible for data purification and standardization, the evaluation analysis module 2 for intelligent analysis, the decision control module 3 as the command and execution center, the cloud platform module 4 as the data aggregation center and wisdom evolution engine, and the early warning and response module 5 to ensure rapid response in extreme situations. These modules are tightly interconnected through an industrial-grade, high-reliability data communication network (for example, an industrial Ethernet based on optical fiber or a 5G wireless network with low delay and high bandwidth characteristics), ensuring that data flows can be transferred in real time, accurately and without obstacles within the system.

[0024] At the physical implementation level, the deployment of the system fully takes into account the actual needs of the engineering site. The various sensor hardware covered by the multi-source information acquisition module 0 is precisely installed in the internal structure of the tunnel 1-0 (such as the lining surface, key cross section) and the surrounding rock geological body 1-5 which has a significant impact on its stability according to its function and monitoring target. While the software functions of the data preprocessing module 1, the evaluation and analysis module 2 and the decision control module 3 are deployed on one or more high-performance servers. These servers can be flexibly selected to be deployed locally in the central control room of the tunnel site for on-site maintenance and rapid response according to the needs of engineering management; or to be deployed centrally in a remote cloud computing data center to take advantage of the powerful computing resources and data storage capabilities of the cloud platform. The cloud platform module 4 itself is a complex software platform based on B / S or C / S architecture, authorized users can access the system through a standard web browser or a dedicated client software at any place with network connection, and its user interface as shown in Figure 5 provides an intuitive and comprehensive overview of the system status and an interactive entry. Finally, the physical execution units of the early warning and response module 5, such as high-decibel sound and light alarms, high-brightness LED warning screens, etc., are installed at key locations such as the tunnel portal, the working area of the tunnel face and the central control room where warning information needs to be delivered in the first time.

[0025] Specifically, the multi-source information acquisition module 0 is the "sensory system" of the entire system, and its design directly determines the depth and breadth of the system's awareness of the tunnel state. This embodiment deeply understands that the core risk of steep slope section tunnel comes from its asymmetric stress environment, which is prone to induce overall or local sliding driven by gravity, and significant lateral stress concentration on one side of the slope. Therefore, the sensor layout scheme must be able to accurately capture these specific mechanical behavior patterns. For this purpose, this module adopts a differentiated deployment strategy based on risk identification, rather than simple uniform deployment. Referring to Figure 2 , the specific embodiment of this strategy is: at the tunnel vault, as the core area bearing the vertical load of the overlying rock mass and reflecting the overall settlement deformation of the tunnel, the system will deploy a set of high-precision static leveling instruments or multi-point displacement meters in the form of displacement sensors 02 at a small interval, for example, every 5-10m along the longitudinal direction of the tunnel, which can continuously monitor the vertical displacement of the vault with sub-millimeter accuracy. At the sidewalls of the tunnel, especially near the slope foot, this area is the most severe part of lateral horizontal stress concentration and shear deformation, the system will use a "combined" monitoring method, that is, a displacement sensor 02 for monitoring the horizontal convergence deformation of the tunnel is deployed on the same section, as shown in Figure 2The displacement sensor layout points 1-2 and the stress meters 03 capable of directly measuring the magnitude and direction of the stress inside the surrounding rock are shown schematically. The stress meters 03 can be selected from the vibrating wire stress meters or more advanced fiber bragg grating (FBG) stress meters, which are pre-embedded at different depths inside the surrounding rock to obtain the stress distribution along the radial direction.

[0026] In addition to the direct monitoring of the structural response, in order to grasp the inherent properties of the surrounding rock mass 1-5 and their changes from a more macroscopic scale, this module also integrates the geological radar 01 and the optional three-dimensional laser scanner 1-4. The geological radar 01 is used to periodically perform "CT scanning" on the tunnel periphery to detect the development of the fracture network inside the surrounding rock, the existence of concealed structures, and the distribution and changes of water-bearing areas. The three-dimensional laser scanner 1-4 is used to quickly obtain high-precision three-dimensional point cloud data of the tunnel inner wall during construction or maintenance, providing a geometric basis for the fine construction of the model. In order to also consider external environmental factors, this module also optionally integrates vibration sensors 04 for monitoring dynamic disturbances caused by construction blasting or traffic loads, as well as temperature and humidity sensors 05 and harmful gas concentration sensors 06 for ensuring the safety of the operating environment inside the tunnel. This differentiated, multi-physical field integrated monitoring network design ensures that the system can comprehensively capture the true state of the steep slope section tunnel from multiple dimensions and multiple scales, providing an unprecedented high-quality data foundation for subsequent accurate assessment.

[0027] More importantly, the multi-source information acquisition module 0 in this embodiment is not a passive "data recorder", but an intelligent front-end with active adjustment capability. It establishes a key feedback control link with the evaluation and analysis module 2. The evaluation and analysis module 2 will perform real-time comprehensive evaluation of the tunnel stability and issue a dynamically changing risk level signal. The controller of the multi-source information acquisition module 0 continuously listens to this signal and automatically adjusts the collection frequency of each sensor inside it according to the pre-set logic. For example, the system can pre-set a set of frequency scheduling strategies: When the risk level is rated as "low", all sensors work in "energy-saving patrol" mode, and the collection frequency is maintained at a basic level (for example, the displacement sensor collects once an hour); When the risk level rises to "medium", the system automatically switches to the "focus" mode, and the collection frequency of the sensors related to the risk area is increased several times (for example, collected once every 10 minutes); when the risk level is determined to be "high", the system enters the "emergency encryption" mode, and the key area is monitored at the highest frequency (for example, collected once every 30 seconds) without interruption. This dynamic adaptive collection mechanism enables the monitoring resources of the system to be intelligently and on-demand allocated to the most needed places, achieving fine-grained management of monitoring, while ensuring high-resolution capture of the risk evolution process, effectively avoiding data redundancy, and reducing the operating cost of the system.

[0028] The data preprocessing module 1 is connected with the multi-source information collection module 0 in the system through a high-speed data interface, and receives original and heterogeneous data streams from hundreds of sensors. Its core task is to perform data purification, conversion and standardization to provide "clean" and "easy-to-understand" inputs for the complex algorithms of the evaluation and analysis module 2. As shown in FIG. 1, it is internally implemented as a multi-stage data processing pipeline. Figure 3

[0029] Firstly, data cleaning: there are inevitably missing values in the original data stream due to signal transmission interruption, or abnormal values due to transient sensor failure. The cleaning algorithm built in this module will first identify these problem data. For missing values, the system will intelligently select a filling strategy according to the type and context of the data, for example, for stress data with obvious trend, linear or spline interpolation method based on adjacent time points can be used a1; for vibration signals with stronger randomness, more complex prediction filling based on autoregressive model can be used. For abnormal values, the system uses statistical methods (such as local outlier factor LOF algorithm) for identification and labeling a2, these labeled data can be reviewed by human or directly removed or corrected by algorithm according to rules.

[0030] Secondly, data denoising: strong electromagnetic interference and mechanical vibration in the tunnel site will bring a lot of noise to the sensor signal. This module integrates a filter library b1, which contains a variety of advanced digital filtering algorithms. The system will adaptively call the most suitable filter according to the spectral characteristics of each sensor signal to filter out noise to the greatest extent and restore the true physical quantity change trend.

[0031] ​Finally, data calibration and fusion: To account for the possible performance degradation or drift of sensors over time, this module will utilize periodic field calibration data to dynamically calibrate the sensor measurements c1. At the same time, it will align the data from different sensors, which may have slight differences in timestamps, through precise time synchronization algorithms, and fuse related physical quantities (such as stress, displacement, temperature, etc. in the same section) into a structured, multi-dimensional state vector.

[0032] After this series of rigorous processing procedures, the originally chaotic and complex raw data is transformed into a high-quality, high-dimensional, time-synchronized standardized data set, providing an ideal input for the deep learning model of the evaluation and analysis module 2.

[0033] The evaluation and analysis module 2 is the key to the transition from "data" to "insight". It is connected with the data preprocessing module 1 to provide the high-quality data set as the basis for analysis. Functionally, this module is composed of three logically independent but closely coupled units: three-dimensional modeling unit, numerical simulation unit, and intelligent evaluation unit. The three-dimensional modeling unit is the foundation of building a digital twin world of the tunnel. It is not just a simple three-dimensional visualization tool, but a comprehensive geomechanical model builder integrating multi-source information. This unit first imports the design drawings of the tunnel (providing ideal geometric boundaries), then fuses detailed geological survey reports (providing initial distribution of rock mass mechanical parameters). Based on this, it further utilizes high-precision three-dimensional point cloud data obtained by devices such as three-dimensional laser scanners 1-4 during construction and operation to accurately reverse correct the geometric shape of the model to reflect the true excavation profile and initial support shape. In this way, this unit constructs a fine three-dimensional geomechanical model that can highly approximate the geometric, physical and mechanical properties of the real engineering entity.

[0034] The numerical simulation unit is a platform for theoretical analysis and risk pre-emption. It is embedded with a professional and secondarily developed numerical calculation engine for geotechnical engineering (e.g., FLAC3D based on the finite difference method or ABAQUS based on the finite element method). Based on the high-fidelity model constructed by the three-dimensional modeling unit, the unit can perform complex multi-field coupled numerical simulation. For example, it can simulate the changes in saturation and pore water pressure of surrounding rock caused by rainwater infiltration under heavy rainfall conditions, as well as the resulting reduction in effective stress and rock mass strength degradation process (fluid-structure interaction analysis); or simulate the dynamic response and cumulative damage of the tunnel structure under seismic load (dynamic analysis). One of the core tasks of this unit is to generate a large-scale and comprehensive numerical simulation database through a large number of systematic parameterized analysis and Monte Carlo random simulation. This database contains detailed mechanical response data (such as stress field, displacement field, plastic zone distribution, etc.) of the tunnel under various foreseeable and even extreme adverse working condition combinations, which constitutes an important source of training data for the intelligent assessment unit.

[0035] The core of the intelligent assessment unit is a specially designed deep learning model, which is particularly suitable for processing monitoring data of steep slope section tunnels. For example, a hybrid architecture of convolutional neural network (CNN) and recurrent neural network (RNN) is adopted because the risk of steep slope section is not only related to the time series evolution of a single monitoring point, but also closely related to the spatial correlation characteristics between monitoring points at different positions (such as slope top, slope foot, and side wall). This hybrid model can capture and fuse features in both time and space dimensions, thereby more accurately identifying the progressive failure mode under asymmetric stress.

[0036] First, in terms of training strategy, it adopts a fusion training mechanism of "simulation data augmentation and measured data driven". The fusion training set used for its training consists of two parts of data: the first part is the numerical simulation data generated by the numerical simulation unit, which is large in scale and covers all working conditions. This part of data provides the model with rich "prior knowledge", enabling it to learn and understand the internal mechanical behavior of the tunnel under various extreme and rare conditions, thereby greatly improving the model's generalization ability when facing unknown working conditions. The second part is the verified real-time monitoring data representing the physical laws of the real world obtained from the data preprocessing module 1. This part of data plays the role of "reality anchor point", which constantly calibrates and fine-tunes the theoretical knowledge learned from simulation data by the model, ensuring that the model's prediction results closely match the engineering practice. This training method of deep fusion of "theoretical knowledge" and "practical experience" enables the intelligent assessment unit to achieve a new level in accuracy and reliability of assessment.

[0037] Secondly, in the input feature design, in order to make the evaluation more targeted, the input feature vector of the unit is carefully screened and constructed. In addition to the conventional stress, displacement and other monitoring data, it also forcibly contains a plurality of specific risk factors specially used to depict the unique risks of steep slope sections. These factors can optionally include: a slope angle parameter that directly reflects the macro-geometric stability of the slope; a lateral stress concentration degree parameter that quantifies the degree of asymmetric stress of the tunnel (which can be calculated by analyzing the reading distribution of the sidewall stress meter); and a modified rainfall infiltration model parameter that takes into account the influence of slope on surface runoff collection and infiltration path. When new real-time monitoring data is input into this well-trained intelligent evaluation unit, it can quickly and forward-lookingly analyze and ultimately output an accurate quantitative evaluation result of the current and future stability state of the tunnel. This result can be a continuous stability index between 0 and 1, or a graded risk warning signal.

[0038] The decision control module 3 is the bridge connecting "intelligent cognition" and "physical intervention", and is the execution mechanism for the system to realize automatic closed-loop control. It is directly connected to the intelligent evaluation unit of the evaluation analysis module 2 and receives the stability evaluation results output by it in real time. The core function of this module is to autonomously and safely formulate and execute physical control strategies. When receiving the risk warning signal from the evaluation analysis module 2, such as "at tunnel mileage pile number YK1+500, the shear stress in the left slope toe area continues to grow, and the predicted instability probability within the next 24 hours exceeds the threshold", the decision logic engine inside the decision control module 3 will immediately start. This engine can be a rule base based on expert system, which stores a large number of "IF-THEN" rules (for example, IF shear stress growth AND pore water pressure rise THEN start drainage); or a more advanced decision model trained through reinforcement learning, which can balance between multiple objectives (such as safety, economy, timeliness) and find the optimal control action sequence.

[0039] However, the originality of the present embodiment lies in that any preliminary physical control instruction generated by the decision engine must pass a strict simulation pre-deployment safety check procedure. Specifically, the decision control module 3 will feed back the preliminary generated instruction in a parameterized form (for example, instruction target: reduce the pore water pressure in the YK1+500 area; execution device: No. 3 drainage pump; execution parameter: power 80%, continuous operation for 2 hours) to the numerical simulation unit of the evaluation analysis module 2. The numerical simulation unit will immediately load the current tunnel three-dimensional model and state data, and perform a "forward-looking" rapid simulation to accurately calculate how the stress field, displacement field and seepage field of the tunnel in this area and even the surrounding area will evolve if the instruction is executed. This process is like a rehearsal of the action to be taken in the virtual world. The simulation pre-deployment will output a detailed prediction report to evaluate the expected effect (for example, whether the pore water pressure can be effectively reduced) and potential side effects (for example, whether it will trigger the risk of stress redistribution in other areas) of the instruction. Only when the pre-deployment result confirms that the instruction is "effective and harmless", the simulation pre-deployment is considered to pass. At this time, the decision control module 3 will confirm the instruction as the final execution instruction, and through the control network, issue accurate and executable control commands to the corresponding maintenance or intervention equipment deployed inside the tunnel, such as the intelligent drainage pump valve system in the present embodiment. This "plan before action" decision-making mechanism adds a crucial "safety lock" to the automated control system, fundamentally avoiding the risk of erroneous decisions due to model bias or unknown factors. Specifically, its technical effect is particularly prominent in the special stress environment of steeply inclined slope section tunnels. For example, a preliminary instruction may be "grouting reinforcement in the left slope foot area". Traditional experience judges that this is usually safe. However, the simulation pre-deployment of the present embodiment can find through rapid numerical calculation that this grouting behavior may cause stress transfer to the right slope top, making the shear stress of the right slope top, which is in a sub-stable state, exceed the safety threshold, thereby triggering new risks. Simulation pre-deployment can discover and quantify this "stress redistribution" effect caused by intervention measures in advance, which cannot be achieved by traditional experience judgment or static rule base, thereby achieving higher levels of safety.

[0040] As Figure 5As shown, the cloud platform module 4 provides a powerful, web-based integrated management platform, providing a unified system monitoring, data analysis and remote management portal for all authorized users. In the system architecture, it maintains two-way communication with all core modules such as the decision control module 3 and the evaluation analysis module 2 through a high-speed data interface. Its core value lies in two aspects: data full life cycle management and system adaptive learning and evolution. As the data hub, it is responsible for gathering, integrating and structurally storing all data from all modules of the system, forming a traceable, analyzable tunnel full life cycle engineering big data archive.

[0041] The most important function is to drive the adaptive learning and evolution of the entire system. After a physical control instruction is successfully executed, the event does not end here. The multi-source information collection module 0 will continuously monitor the subsequent state changes of the intervention area and upload these data as the most real "effect feedback data" to the cloud platform module 4. A special analysis engine inside the cloud platform module 4 will accurately compare the real effect feedback data with the "expected effect" predicted by the simulation pre-inference before the decision, thereby quantifying the actual effectiveness of each decision and control and identifying the deviation. A closed-loop data case containing "risk identification-intelligent evaluation-simulation inference-decision execution-actual effect" complete information is generated and permanently stored. When such cases accumulate to a certain number, the cloud platform module 4 will automatically trigger a background model retraining task using its powerful cloud computing capabilities: incremental learning or periodic global optimization of the model of the intelligent evaluation unit in the evaluation analysis module 2, and integrating these real-world case data containing new knowledge into the model, so that the next evaluation prediction is more accurate.

[0042] The self-evolution mechanism of the present embodiment is realized through two parallel feedback paths: Short-term model parameter fine-tuning: After the execution of a single physical control instruction, the system quantitatively compares the real effect feedback data collected with the prediction results of the previous simulation pre-inference, generating a deviation vector. The deviation vector will be used to perform a real-time, small-scale online update of the model weight parameters of the intelligent evaluation unit through gradient descent and other algorithms. This ensures that the evaluation model can quickly adapt to small changes in the field.

[0043] Long-term control strategy optimization: The system stores each complete "risk identification-pre-inference-execution-actual effect" closed-loop event as a structured learning case in the cloud database. When the cases accumulate to a predetermined number (e.g. 100), the system will periodically trigger an offline global optimization to evaluate and iteratively update the control strategy rule library in the decision control module.

[0044] Through this dual-channel mechanism that combines short-term fine-tuning with long-term optimization, the system achieves a comprehensive and logically clear self-evolution capability.

[0045] The early warning and response module 5, serving as the system's rapid emergency channel, maintains the most direct and fundamental connection with the multi-source information acquisition module 0. Its purpose is to handle sudden, high-intensity emergencies characterized by rapid development, leaving insufficient time for complex intelligent analysis and decision-making processes. Internally, this module maintains a relatively simple yet extremely fast-responding list of rules based on fixed thresholds. For example, if the peak vibration velocity detected by a vibration sensor 04 instantaneously exceeds the blasting safety control standard, or if the reading of a gas concentration sensor 06 suddenly spikes to a dangerous level, this module will be directly triggered by a hardware interrupt or a high-priority software interrupt. It bypasses the relatively time-consuming processes of assessment and decision-making, immediately activating the highest-level alarm on-site and sending predefined emergency warning information via the communication module to the mobile phones, walkie-talkies, and other terminal devices of all relevant personnel. This ensures that in the most critical moments, personnel receive the most timely evacuation and intervention instructions, providing a final, solid guarantee for the safety of life and property.

[0046] Example 2: This embodiment provides an intelligent assessment method for the stability of tunnels on steep slopes. The method details the specific operational logic and workflow of the system described in Embodiment 1. For example... Figure 4 As shown, the method specifically includes the following core steps: Step S1: Dynamic and Differentiated Data Acquisition. This step is executed through the system's multi-source information acquisition module 0. The method begins with comprehensive and continuous multi-dimensional information acquisition of the target steep slope tunnel section. Unlike traditional fixed-mode acquisition, this step includes two core intelligent features. First, differentiated deployment: As described in Example 1, the method requires multiple sensors within the system to be deployed differently based on key areas such as the tunnel's arch, sidewalls, and toe, to ensure that the acquired data can reflect the specific risk patterns of the steep slope section to the greatest extent. Second, dynamic adjustment: The execution of this step is not isolated; it forms a dynamic feedback loop with the subsequent evaluation step S3. Specifically, during the continuous operation of the method, step S3 continuously outputs a dynamically updated tunnel risk level. This step will adjust the information acquisition frequency of all or some sensors in real time and automatically based on this risk level. For example, when the risk level increases from "low" to "medium," the method will instruct the sampling interval of displacement and stress sensors related to the risk area to be shortened from 60 minutes to 10 minutes, capturing the evolution details of the risk with higher temporal resolution. This dynamic adjustment mechanism ensures the efficiency and effectiveness of the method during the information acquisition phase.

[0047] Step S2: Standardization processing is performed in this step by collecting massive, heterogeneous, multi-dimensional information in step S1. This is a key data "purification" and "preparation" phase. The method will sequentially call a series of algorithm modules such as data cleaning, data denoising, data fusion, and data calibration. For example, for stress, displacement, and temperature data collected by different sensors at the same time, the method will perform precise timestamp alignment and data fusion operations to form a unified, multivariate system state snapshot. Through this series of refined processing, the method ensures that the data input to the next step for intelligent analysis is of high quality, consistent, and reliable.

[0048] Step S3: Intelligent evaluation driven by fused data. The internal logic of this evaluation step is: first, the method will call an intelligent evaluation unit that has been specially trained. The training of this intelligent evaluation unit is based on a fusion training set composed of two types of data: one is simulation data generated by numerical simulation techniques, covering various theoretical working conditions; the other is real monitoring data obtained from actual engineering and verified. Through this "theory + practice" fusion training, the intelligent evaluation unit is ensured to have strong generalization ability. In addition, when performing evaluation, the feature vector input to the intelligent evaluation unit will specially include certain risk factors that can deeply reflect the characteristics of steep slope sections, such as slope angle, lateral stress concentration degree, and modified rainfall infiltration model parameters. After training, when new real-time monitoring data is input, the intelligent evaluation unit will use the complex nonlinear mapping relationship it has learned to perform rapid calculation and finally output an accurate quantitative evaluation result of the current and future stability state of the tunnel.

[0049] Step S4: Decision and control after safety check. This step is executed by decision control module 3. When the evaluation result output by step S3 indicates the presence of potential risks, this step is activated. The method first generates a preliminary physical control instruction according to the nature and severity of the evaluation result. Then, also the core guarantee of the safety of this method, before issuing the instruction, a simulation pre-deployment is performed. That is, the parameters of the instruction are input into the digital twin model of the tunnel for a rapid virtual execution to predict its possible effects and impacts. Only when the pre-deployment result confirms that the instruction is both effective and safe, the method will continue to execute, issue the final control command to the corresponding equipment in the tunnel, and complete a closed-loop operation from intelligent analysis to physical intervention.

[0050] Step S5: System self-learning and iteration based on feedback. This step is executed by the cloud platform module 4. After the physical control instructions in step S4 are executed, this step is started. The method continuously receives real effect feedback data about the subsequent state changes of the intervention area from the multi-source information collection module. Then, the feedback data is compared and analyzed with the expected results of the simulation pre-deployment to form a complete learning case. Finally, the method updates and optimizes the model parameters of the intelligent evaluation unit in step S3 regularly using the continuously accumulated cases, so that the evaluation model can evolve in the use process, and its evaluation ability will become more and more accurate over time.

[0051] The method provided by the embodiment is formed by the organic combination and cyclic execution of the above steps, forming a new paradigm of dynamic, closed-loop, intelligent and adaptive tunnel stability management. It deeply integrates advanced sensing technology, big data analysis, artificial intelligence and automation control, not only significantly improves the safety control level of complex engineering objects such as steeply inclined slope section tunnels, but also provides a clear and feasible technical path for the intelligent transformation of the tunnel engineering field.

Claims

1. An intelligent evaluation system for the stability of a steeply inclined section of a tunnel, characterized in that, include: Multi-source information acquisition module (0), the multi-source information acquisition module (0) includes multiple sensors that are differently deployed according to the arch, sidewall and toe of the steep slope section; A data preprocessing module (1) is connected to the multi-source information acquisition module (0) and is used to receive and process multi-dimensional information acquired by the multiple sensors. The evaluation and analysis module (2) is connected to the data preprocessing module (1). The evaluation and analysis module (2) includes a three-dimensional modeling unit, a numerical simulation unit and an intelligent evaluation unit. The intelligent evaluation unit is trained based on a fusion training set. The fusion training set consists of the numerical simulation data generated by the numerical simulation unit and the real-time monitoring data processed by the data preprocessing module (1). The decision control module (3) is connected to the evaluation and analysis module (2) and is used to receive the stability evaluation results output by the intelligent evaluation unit and input the physical control command generated based on the results into the numerical simulation unit for simulation pre-deduction. After the simulation pre-deduction is passed, the physical control command is sent to the maintenance or intervention equipment in the tunnel. The cloud platform module (4) is connected to the decision control module (3) and the evaluation and analysis module (2) respectively, and is used to receive the effect feedback data after the maintenance or intervention equipment executes the physical control command, and update the model parameters of the intelligent evaluation unit according to the effect feedback data; and The early warning and response module (5) is connected to the multi-source information acquisition module (0).

2. The steeply inclined slope section tunnel stability intelligent evaluation system according to claim 1, characterized in that: The multi-source information acquisition module (0) is also used to adjust the acquisition frequency of the multiple sensors according to the risk level output by the assessment and analysis module (2).

3. The intelligent evaluation system for the stability of a steeply inclined ramp section tunnel according to claim 1, characterized in that: The input feature vector of the intelligent evaluation unit includes slope inclination angle parameters, lateral stress concentration parameters, and rainfall infiltration model parameters.

4. The intelligent evaluation system for the stability of a steeply inclined ramp section tunnel according to claim 1 or 3, characterized in that: The intelligent evaluation unit is a neural network model.

5. The intelligent evaluation system for the stability of a steeply inclined ramp section tunnel according to claim 1, characterized in that: The cloud platform module (4) is also used to update the control strategy of the decision control module (3) based on the effect feedback data.

6. The steeply inclined slope section tunnel stability intelligent evaluation system according to claim 1, characterized in that: The maintenance or intervention equipment includes reinforcement devices, drainage facilities, and ventilation systems.

7. An intelligent assessment device for the stability of tunnels on steep slopes, characterized in that, Includes the system as described in any one of claims 1-6, as well as the hardware equipment and auxiliary facilities required for installing and securing the system.

8. A method for intelligently assessing the stability of tunnels on steep slopes, characterized in that: Includes the following steps: S1. Multi-dimensional information of the steep slope section of the tunnel is collected by the multi-source information acquisition module (0). The multiple sensors in the multi-source information acquisition module (0) are arranged differently according to the tunnel's arch, sidewalls and slope toe. S2. The multi-dimensional information is processed by the data preprocessing module (1); S3. The stability of the processed information is evaluated by the evaluation and analysis module (2). The evaluation steps include: training an intelligent evaluation unit based on a fusion training set consisting of numerical simulation data and real-time monitoring data, and using the intelligent evaluation unit to perform the evaluation. S4. Generate physical control instructions through the decision control module (3), and simulate and pre-determine the instructions in the three-dimensional model before issuing the physical control instructions; S5. Receive the effect feedback data after the execution of the physical control command through a cloud platform module (4), and update the model parameters of the intelligent evaluation unit according to the effect feedback data.

9. The intelligent assessment method for the stability of tunnels on steep slopes according to claim 8, characterized in that: In step S3, the input feature vector of the intelligent evaluation unit includes slope inclination angle parameters, lateral stress concentration parameters, and rainfall infiltration model parameters.

10. The intelligent assessment method for the stability of tunnels on steep slopes according to claim 8, characterized in that, It also includes the following step: S11. In step S1, the acquisition frequency of the multiple sensors is dynamically adjusted according to the risk level output in step S3.

Images