A tunnel construction risk early warning method and system based on a data middle platform

Abstract

This invention relates to the field of tunnel engineering safety monitoring technology, and discloses a tunnel construction risk early warning method and system based on a data platform. The method includes: acquiring multi-source monitoring data from the tunnel construction site and preprocessing it; constructing a spatial coordinate system and performing multi-physics coupling and correlation; calculating the coupled multi-physics fields; constructing a multi-agent game theory model to simulate the interaction and evolution of various risk factors; quantifying resource requirements and predicting future changes in resource requirements; dynamically allocating resources; determining risk levels and generating early warning information; tracking effects; and continuously optimizing the multi-agent game theory model and system parameters to obtain an optimized early warning scheme. This invention achieves precise matching and dynamic adjustment of early warning business requirements and underlying computing resources by constructing an elastic resource scheduling mechanism driven by early warning tasks.

Description

Technical Field

[0001] This invention relates to the field of tunnel engineering safety monitoring technology, and more specifically, to a tunnel construction risk early warning method and system based on a data platform. Background Technology

[0002] As a crucial component of transportation infrastructure, tunnel engineering plays a vital role in traversing complex geological areas. With the rapid advancement of my country's transportation network construction, an increasing number of tunnel projects need to cross areas with adverse geological conditions, such as karst development zones, fault fracture zones, and high ground stress zones. During construction, these projects face various geological hazards, including sudden water inrushes, large deformations of surrounding rock, collapses, and gas outbursts. These hazards are often characterized by their suddenness, destructive power, and high difficulty in prediction, posing a serious threat to the lives of construction workers and the progress of project construction. Therefore, the establishment of an effective risk early warning mechanism is urgently needed.

[0003] With the development of sensing and Internet of Things (IoT) technologies, modern tunnel construction sites have deployed a large number of monitoring devices capable of collecting multi-dimensional data such as surrounding rock deformation, stress and strain, seepage pressure, and ambient gas levels in real time. However, existing tunnel construction risk early warning systems have significant shortcomings in data processing and risk analysis: various monitoring devices operate independently, data formats are heterogeneous and standards are not uniform, making it difficult to achieve effective fusion of multi-source data; early warning methods are mainly based on threshold judgments of single indicators, failing to effectively capture the coupled evolution patterns between multiple risk factors, and lacking foresight in early warning of complex disasters.

[0004] Furthermore, the computing resource configuration of existing early warning systems is usually static and cannot be dynamically adjusted according to the real-time needs of early warning operations. When the system needs to perform complex risk simulation analysis, fixed computing resources may become a performance bottleneck, leading to early warning delays; while during routine monitoring, over-configured resources result in waste.

[0005] Therefore, a technical solution is needed that can dynamically bind early warning business requirements with underlying computing resources, so that the early warning system itself has elasticity and resilience, and can ensure that risk early warning can operate accurately, timely and reliably under any operating conditions. Summary of the Invention

[0006] This invention provides a method and system for early warning of tunnel construction risks based on a data platform, which solves the technical problem of performance bottlenecks in high-load scenarios in related technologies.

[0007] This invention provides a tunnel construction risk early warning method based on a data platform, comprising: Acquire multi-source monitoring data from the tunnel construction site, perform preprocessing, and obtain a time-aligned standardized dataset; Based on the time-aligned standardized data set, a spatial coordinate system is constructed and multi-physics coupling correlation is performed; the coupled multi-physics fields are calculated to obtain comprehensive risk field distribution data; Based on the comprehensive risk field distribution data, a multi-agent game inference model is constructed to simulate the interaction and evolution process between various risk factors and obtain the game inference results. Based on the game theory results, resource demand is quantified, and future changes in resource demand are predicted to obtain a resource demand quantification report. Based on the resource demand quantification report, resources are dynamically configured to obtain resource scheduling execution results; Based on the game theory results and resource scheduling execution results, risk level determination and early warning information generation are performed to obtain structured early warning information; the effect of the structured early warning information is tracked to obtain effect tracking data. Based on the structured early warning information and effect tracking data, the multi-agent game inference model and system parameters are continuously optimized to obtain an optimized early warning scheme.

[0008] In a preferred embodiment, the step of acquiring and preprocessing multi-source monitoring data at the tunnel construction site includes: A categorized data acquisition strategy is adopted, which uses ground-penetrating radar to acquire time-domain waveform data of electromagnetic wave reflection signals, surrounding rock deformation monitoring equipment to acquire deformation data, stress-strain monitoring equipment to acquire mechanical parameter data, environmental monitoring equipment to acquire environmental parameter data, and video surveillance equipment to acquire real-time image frame data of the construction site. The resulting raw data stream from the monitoring equipment includes time-domain waveform data, deformation data, mechanical parameter data, environmental parameter data, and real-time image frame data. The raw data stream from the monitoring equipment is parsed for heterogeneous data formats, converting binary waveform data into numerical matrix form and raw sampled values ​​into engineering unit values. Establish a data quality detection rule base, set reasonable range thresholds for various types of data in the raw data stream of the monitoring equipment, and correct abnormal values ​​in the raw data stream of the monitoring equipment that exceed the reasonable range thresholds; The timestamps of ground-penetrating radar detection equipment, surrounding rock deformation monitoring equipment, stress and strain monitoring equipment, environmental monitoring equipment, and video surveillance equipment are calibrated using the Network Time Protocol. A resampling method is used to unify data sources with different sampling frequencies to a preset standard sampling period.

[0009] In a preferred embodiment, the construction of the spatial coordinate system and the multiphysics coupling correlation include: A local coordinate system is established with the tunnel axis as the reference. The installation positions of the geological radar detection equipment, surrounding rock deformation monitoring equipment, stress and strain monitoring equipment, environmental monitoring equipment, and video surveillance equipment are transformed to a unified coordinate system. Spatial grid units are divided according to the tunnel cross-section shape and the layout of monitoring points. The displacement values ​​of discrete monitoring points are extended into a continuous deformation field distribution using the Kriging interpolation method, and the stress values ​​of discrete monitoring points are extended into a continuous stress field distribution using the inverse distance weighted interpolation method. Establish a constitutive correlation model between the deformation field and the stress field, a fluid-structure interaction model between the seepage field and the stress field, and a thermo-mechanical interaction model between the temperature field and the deformation field, stress field, and seepage field.

[0010] In a preferred embodiment, the step of obtaining the comprehensive risk field distribution data includes: Define the calculation rules for the comprehensive risk index, and convert the state values ​​of the deformation field, stress field, seepage field and temperature field into dimensionless risk contribution values ​​through normalization. The corresponding weighting coefficients are determined based on the degree of influence of deformation field, stress field, seepage field and temperature field on the overall risk. The relative importance of each risk factor is analyzed by using the analytic hierarchy process (AHP). The weight values ​​are obtained by constructing a judgment matrix and calculating the eigenvectors. The risk contribution values ​​of the deformation field, stress field, seepage field, and temperature field are weighted and summed to obtain the comprehensive risk field distribution data.

[0011] In a preferred embodiment, the construction of the multi-agent game inference model includes: The risk of surrounding rock instability is abstracted into a surrounding rock instability intelligent agent, the risk of groundwater seepage is abstracted into a seepage intelligent agent, and the response of the support structure is abstracted into a support intelligent agent. The initial states of the surrounding rock instability intelligent agent, the seepage intelligent agent, and the support intelligent agent are assigned based on the comprehensive risk field distribution data; Define the action space of the surrounding rock instability agent, the seepage agent, and the support agent, and the interaction function between the agents; Define the revenue functions for the surrounding rock instability agent, the seepage agent, and the support agent.

[0012] In a preferred embodiment, the steps for obtaining the game theory deduction result include: A rapid coarse-grained inference is performed, using a simplified agent state space and a first preset time step, to obtain a preliminary judgment on the risk evolution trend; When the rapid simulation identifies that the comprehensive risk index exceeds the first warning threshold, a fine-grained simulation is initiated, using a complete agent model and a second preset time step. Analyze the temporal relationship of the state transitions of the surrounding rock instability intelligent agent, the seepage intelligent agent, and the support intelligent agent during the simulation process, identify the state transition sequences with causal relationships, and form a cascaded amplification path; Dynamic risk entropy is calculated based on information entropy theory. The state combinations and probability distributions during the statistical deduction process are calculated. The state combinations are defined based on the state combinations of the surrounding rock instability intelligent agent, the seepage intelligent agent, and the support intelligent agent. The rate of change of entropy value over time is calculated.

[0013] In a preferred embodiment, quantifying the resource requirements includes: Establish a classification system for early warning tasks, dividing tasks into data preprocessing, threshold monitoring, statistical analysis, and model inference, and calculate the complexity score of the tasks based on evaluation indicators; Establish a mapping model between early warning tasks and resource requirements. Take task type, complexity level, and data scale parameters as input, and output quantitative requirements for the number of computing processor cores, memory capacity, graphics processor resources, and data throughput bandwidth. An autoregressive integral moving average model is used to model the resource consumption time series, and the resource demand for several future time steps is predicted based on the trained model.

[0014] In a preferred embodiment, the dynamic configuration of resources includes: Compare the resource allocation goals with the current resource status, calculate the gaps or redundancies of various resources, and formulate resource allocation adjustment plans. The container elastic scaling method is used to send expansion or shrinkage commands to the container orchestration platform to create or terminate container instances based on predefined container images and resource configuration templates. The identified hot data is preloaded from cloud storage to the high-speed cache of the edge server, and the caching strategy uses the least recently used algorithm for cache eviction. Establish a priority classification system for stream processing jobs and adjust the resource allocation strategy of the stream processing framework to allocate more processing slots and memory resources to jobs with a preset priority threshold.

[0015] In a preferred embodiment, the risk level determination and early warning information generation include: Establish risk level classification standards, classifying risk status into five levels: normal, attention, warning, alert, and emergency. The determination of risk level comprehensively considers dynamic risk entropy value, entropy value change rate, and the severity of cascading amplification path. The early warning information is organized in a structured format; Establish an information push strategy based on the recipients of the early warning information, and generate differentiated versions of early warning information according to the information needs and receiving capabilities of different recipients; Each warning message is logged throughout its entire process, and a mechanism for tracking the effectiveness of warnings is established to compare the warning predictions with the actual results.

[0016] In a preferred embodiment, a tunnel construction risk early warning system based on a data platform is used to execute the aforementioned tunnel construction risk early warning method based on a data platform, including: The data access module is used to acquire multi-source monitoring data from the tunnel construction site, perform preprocessing, and obtain a time-aligned standardized data set. The risk field construction module constructs a spatial coordinate system and performs multi-physics coupling and correlation based on the time-aligned standardized data set; it then calculates the coupled multi-physics fields to obtain comprehensive risk field distribution data. The game simulation module, based on the comprehensive risk field distribution data, constructs a multi-agent game simulation model to simulate the interaction and evolution process between various risk factors and obtain the game simulation results. The resource quantification module quantifies resource demand based on the game theory results and predicts future changes in resource demand to obtain a resource demand quantification report. The resource scheduling module dynamically configures resources based on the resource demand quantification report to obtain the resource scheduling execution result; The early warning generation module, based on the game simulation results and resource scheduling execution results, determines the risk level and generates early warning information to obtain structured early warning information; it then tracks the effect of the structured early warning information to obtain effect tracking data. The optimization module continuously optimizes the multi-agent game inference model and system parameters based on the structured early warning information and effect tracking data to obtain an optimized early warning scheme.

[0017] The beneficial effects of this invention are as follows: By introducing a multi-agent game theory mechanism, risk factors such as surrounding rock instability, groundwater seepage, and support structure response in tunnel construction are abstracted into intelligent agents with autonomous decision-making capabilities. The interaction and evolution process between various risk factors are simulated, which can effectively capture the coupled evolution law and cascade amplification effect of multiple risk factors. Compared with the traditional single-factor threshold early warning method, it improves the foresight and accuracy of early warning for complex disasters and provides more reliable decision support for construction safety management.

[0018] By constructing an elastic resource scheduling mechanism driven by early warning tasks and establishing a mapping model between early warning tasks and resource requirements, the system achieves precise matching and dynamic adjustment between early warning business needs and underlying computing resources. This solves the problem of early warning delay caused by insufficient computing resources in traditional early warning systems under high load scenarios, while avoiding resource waste during low load periods. This enables the early warning system to maintain stable response capabilities when facing data surges and computationally intensive tasks, thereby improving the overall reliability and resource utilization efficiency of the early warning system. Attached Figure Description

[0019] Figure 1 This is a flowchart of the main process of a tunnel construction risk early warning method based on a data platform in this invention; Figure 2 This is a detailed flowchart of a tunnel construction risk early warning method based on a data platform in this invention; Figure 3 This is a module diagram of a tunnel construction risk early warning system based on a data platform in this invention. Detailed Implementation

[0020] The subject matter described herein will now be discussed with reference to exemplary embodiments. It should be understood that these embodiments are discussed only to enable those skilled in the art to better understand and implement the subject matter described herein, and changes may be made to the function and arrangement of the elements discussed without departing from the scope of this specification. Various processes or components may be omitted, substituted, or added as needed in the examples. Furthermore, some features described in the examples may be combined in other examples.

[0021] At least one embodiment of the present invention discloses a tunnel construction risk early warning method based on a data middle platform, such as... Figures 1 to 2 As shown, it includes: Step 1: Acquire multi-source monitoring data from the tunnel construction site, perform preprocessing, and obtain a time-aligned standardized data set; Step 1.1: Data acquisition from multi-source monitoring equipment; Based on the monitoring equipment network deployed at the tunnel construction site, a categorized data acquisition strategy is adopted to obtain the raw data streams of each type of monitoring equipment. Specifically, for the ground-penetrating radar detection equipment, time-domain waveform data of electromagnetic wave reflection signals are acquired through the radar data acquisition interface, reflecting the geological structural characteristics ahead of the tunnel face; for the surrounding rock deformation monitoring equipment, deformation data such as convergence displacement, arch settlement, and surface settlement of each monitoring section are acquired through the displacement sensor data interface; for the stress-strain monitoring equipment, mechanical parameter data such as surrounding rock stress, anchor bolt axial force, and steel frame stress are acquired through the stress gauge data interface; for the environmental monitoring equipment, environmental parameter data such as methane concentration, carbon monoxide concentration, and oxygen concentration are acquired through the gas detector data interface; and for the video surveillance equipment, real-time image frame data of the construction site is acquired through the video stream interface. The resulting raw data streams from the monitoring equipment include time-domain waveform data, deformation data, mechanical parameter data, environmental parameter data, and real-time image frame data. Step 1.2, Heterogeneous data format parsing; Based on the raw data stream from monitoring equipment, a heterogeneous data format parsing method is used to obtain structured data records. For ground-penetrating radar waveform data, metadata information such as sampling rate, time window, and channel number is parsed, and the binary waveform data is converted into a numerical matrix form. For sensor numerical data, attribute information such as data type, measurement range, and accuracy level is parsed, and the raw sampled values ​​are converted into engineering unit values. For video image data, parameter information such as resolution, frame rate, and encoding format is parsed, and keyframe images are extracted for subsequent processing.

[0022] Step 1.3, Data quality inspection and cleaning; Based on structured data records, data quality inspection and cleaning methods are used to obtain valid data that meets quality standards. A data quality inspection rule base was established, including rules for numerical range detection, rate of change detection, and continuity detection. For time-domain waveform data, the reasonable amplitude range was set to 5% to 95% of the equipment's range, and the frequency range was determined based on the operating frequency band of the ground-penetrating radar equipment. For deformation data, the reasonable displacement range was set to three times the single measurement accuracy to the expected maximum deformation, and the reasonable deformation rate range was determined based on the theoretical values ​​of the surrounding rock type and construction method. For mechanical parameter data, the reasonable stress range was set to zero to 1.5 times the theoretical bearing capacity of the surrounding rock, and the reasonable strain range was calculated based on the material's elastic modulus and the expected stress level. For outliers exceeding the physical range, linear interpolation or cubic spline interpolation of adjacent time points was used for correction. For values ​​exceeding the statistical anomaly range but within the physical range, anomaly marking and preserving the original value was used. For missing data, linear interpolation, spline interpolation, or marking missing data were used depending on the duration of the missing data and its importance. For data with noise interference, a sliding window filtering method was used for smoothing.

[0023] Step 1.4, Timestamp unification and alignment; Based on qualified and valid data, a time-synchronized standardized data set is obtained using a timestamp unification and alignment method. Since the internal clocks of each monitoring device may deviate, the timestamps of each device are calibrated using a network time protocol. The sampling times of each data source are aligned using a unified time base. For data sources with different sampling frequencies, a resampling method is used to unify them to a preset standard sampling period. Finally, standardized data records containing fields such as timestamp, data source identifier, monitoring point location information, monitoring parameters, and numerical values ​​are generated.

[0024] Step 1.5, Deploy the tiered storage architecture; Based on a standardized dataset synchronized with time, a tiered storage architecture is employed to achieve data storage configurations adaptable to varying access needs. A three-tiered storage architecture is established: a hot storage layer using an in-memory database and high-speed solid-state drives (SSDs) to store real-time monitoring data from the past seven days and currently processed datasets, supporting millisecond-level read / write responses; a warm storage layer using an enterprise-grade hard disk array to store historical monitoring data and intermediate calculation results from the past three months, supporting second-level data retrieval responses; and a cold storage layer using object storage services and a tape library system to store long-term archived data (over three months) and backup data, supporting minute-level data recovery responses. Data migration strategies are developed based on data access frequency and importance levels. Real-time monitoring data is automatically migrated to the warm storage layer after seven days in the hot storage layer. Data in the warm storage layer is selectively migrated to the cold storage layer or deleted after three months based on data value assessment results.

[0025] Furthermore, due to potential network latency differences in data transmission between different monitoring devices, the order in which data arrives at the data platform may differ from the actual acquisition order. An event-time-based stream processing mechanism can be used instead of a processing-time-based mechanism. The aim is to ensure that data is sorted and processed according to its actual occurrence time, avoiding data out-of-order issues caused by network latency. Specifically, a watermark mechanism is introduced at the data access layer, marking each data record with an event timestamp. The system determines whether all data within a given time window has arrived based on the watermark's progression. Subsequent processing of the data in that window is only triggered when the watermark crosses the window boundary. For late-arriving data, the degree of lateness determines whether it should be included in the calculation or bypassed.

[0026] Step 2: Based on the time-aligned standardized dataset, construct a spatial coordinate system and perform multi-physics coupling correlation; calculate the coupled multi-physics to obtain comprehensive risk field distribution data; Step 2.1, Establishing the tunnel spatial coordinate system; Based on the monitoring point location information in the standardized dataset, a unified three-dimensional spatial reference frame is obtained by establishing a tunnel spatial coordinate system. A local coordinate system is established with the tunnel axis as the reference, where the longitudinal coordinate represents the mileage along the tunnel axis, the transverse coordinate represents the horizontal offset perpendicular to the axis, and the vertical coordinate represents the elevation relative to the tunnel floor. The installation positions of each monitoring point are transformed into this unified coordinate system to form a spatial distribution map of the monitoring points. According to the tunnel cross-section shape and the monitoring point layout scheme, spatial grid cells are divided to provide a geometric basis for subsequent spatial interpolation calculations.

[0027] Step 2.2, Spatial interpolation extension processing; Based on the spatial distribution of monitoring points and the monitoring values ​​at each point, a spatial interpolation extension method is used to obtain a continuous single-parameter spatial distribution field. For deformation monitoring data, the Kriging interpolation method is used to extend the displacement values ​​of discrete monitoring points into a continuous deformation field distribution. This method can perform optimal unbiased estimation based on the spatial correlation between monitoring points. For stress monitoring data, the inverse distance weighted interpolation method is used to extend the stress values ​​of discrete monitoring points into a continuous stress field distribution. For seepage monitoring data, a constrained interpolation method is used to generate the seepage field distribution in combination with geological structural information. The specific implementation process of Kriging interpolation is as follows: the experimental variogram is calculated based on the known monitoring point data, and the spatial autocorrelation characteristics of the data are analyzed; a suitable theoretical variogram model is selected for fitting, and parameters such as range, sill value, and nugget value are determined; for the spatial location to be estimated, the optimal weighting coefficients are calculated based on its spatial relationship with surrounding known points and the variogram model, and a weighted sum is performed to obtain the estimated value.

[0028] Step 2.3, Multiphysics Coupling Correlation; Based on the spatial distribution fields of each single parameter, a multi-physics coupling correlation method is used to obtain the interaction relationships between physical fields. A constitutive correlation model between the deformation field and the stress field is established, transforming stress field changes into driving factors of the deformation field according to the stress-strain relationship of the surrounding rock. A fluid-structure interaction model between the seepage field and the stress field is established, considering the influence of pore water pressure on effective stress and the influence of stress changes on the permeability coefficient. A thermo-mechanical coupling model between the temperature field and other physical fields is established, considering the thermal stress caused by temperature changes and the influence of temperature changes on material parameters. The calculation of the coupling correlation adopts an iterative solution method, updating the state of each physical field sequentially in each time step until the convergence condition is met.

[0029] Step 2.4, Calculate the overall risk field; Based on the coupled multiphysics distribution, a comprehensive risk field calculation method is adopted to obtain comprehensive risk field distribution data. The calculation rules for the comprehensive risk index are defined, and the state values ​​of each physical field are converted into dimensionless risk contribution values ​​through normalization. According to the degree of influence of each physical field on the overall risk, corresponding weight coefficients are determined. The weight coefficients are obtained by using the analytic hierarchy process (AHP) based on historical disaster case data to analyze the relative importance of each risk factor, constructing a judgment matrix and calculating eigenvectors to obtain weight values. The default values ​​can be set as follows: deformation field weight 0.35, stress field weight 0.30, seepage field weight 0.25, and other factors weight 0.10. The risk contribution values ​​of each physical field are weighted and summed to obtain the comprehensive risk field distribution data.

[0030] Furthermore, since traditional fixed-weight methods cannot adapt to changes in risk characteristics under different geological conditions and construction stages, a dynamic adaptive weight adjustment mechanism can be used to replace the fixed-weight scheme. The aim is to enable the calculation of the comprehensive risk index to automatically adjust the weight ratio of each factor according to the actual working conditions. Specifically, a dynamic weight adjustment model is established. This model takes current geological condition type, construction method, tunneling progress, and other working condition parameters as inputs and outputs a combination of weight coefficients adapted to the current working conditions. The training data of the model comes from historical monitoring data and risk event records under different working conditions. By analyzing the correlation between each risk factor and actual risk events under different working conditions, the mapping relationship between working conditions and weights is learned. In actual operation, the system queries or calculates the corresponding weight coefficients according to the current working condition parameters to achieve dynamic weight adaptation.

[0031] Step 3: Based on the comprehensive risk field distribution data, construct a multi-agent game inference model to simulate the interaction and evolution process between various risk factors and obtain the game inference results; Step 3.1, Risk Factor Agent Modeling; Based on the main risk factors identified in the comprehensive risk field distribution data, an agent abstract modeling method is used to obtain a set of risk factor agents. The risk of surrounding rock instability is abstracted as a surrounding rock instability agent, whose state is described by parameters such as surrounding rock deformation, deformation rate, and plastic zone range. Its behavioral rules reflect the deformation evolution characteristics of the surrounding rock under different stress states. The risk of groundwater seepage is abstracted as a seepage agent, whose state is described by parameters such as water pressure, seepage flow, and aquifer connectivity. Its behavioral rules reflect the flow characteristics of groundwater under different boundary conditions. The response of the support structure is abstracted as a support agent, whose state is described by parameters such as support stress, deformation, and safety factor. Its behavioral rules reflect the mechanical response characteristics of the support structure under different loads. The initial state of each agent is assigned based on the comprehensive risk field data output in step 2.

[0032] Step 3.2, Define the game interaction rules; Based on a set of agents representing risk factors, a game-theoretic interaction rule definition method is used to obtain a game relationship model among the agents. The action space of each agent is defined: the actions of the rock instability agent include state transitions such as stability, slow deformation, accelerated deformation, and instability failure; the actions of the seepage agent include state transitions such as stable seepage, enhanced seepage, and sudden water inrush; and the actions of the support agent include state transitions such as normal operation, stress concentration, localized failure, and overall failure. An interaction influence function is defined to describe how the state change of one agent affects the state transition probabilities of other agents; for example, the enhanced seepage action of the seepage agent increases the probability of the rock instability agent transitioning to accelerated deformation, and the accelerated deformation action of the rock instability agent increases the probability of the support agent transitioning to stress concentration. The payoff function of each agent is defined, designed based on the principle that agents tend to evolve towards higher-risk states to simulate the natural development trend of risk factors, while being constrained and influenced by the states of other agents. Steps 3.1 and 3.2 complete the construction of the multi-agent game inference model, which includes a set of risk factor agents and a game relationship model between agents.

[0033] Step 3.3, Layered Progressive Game Theory Deduction; Based on a game theory model, a hierarchical progressive game theory deduction method is employed to obtain the trajectory of risk evolution. A rapid coarse-grained deduction is performed, using a simplified agent state space and a large time step to complete multiple rounds of game iterations in a short time, obtaining a preliminary judgment of the risk evolution trend. The specific process of the rapid deduction is as follows: at each time step, each agent selects an action based on its current state and the game rules; the system updates the global state based on the combination of actions from all agents, repeating the iteration until a preset deduction duration is reached or the system enters a stable state. When the rapid deduction identifies that the comprehensive risk index exceeds the first warning threshold, a fine-grained deduction is initiated, using a complete agent model and a smaller time step to conduct in-depth analysis of high-risk areas. The fine-grained deduction is conducted within the high-risk time periods and spatial ranges identified by the rapid deduction, enabling the capture of more subtle risk evolution characteristics.

[0034] Step 3.4, Cascade Amplification Path Identification; Based on the inference trajectory of risk evolution, a cascading amplification path identification method is used to obtain the cascading propagation chain between risk factors. The temporal relationship of state transitions of each agent during the inference process is analyzed to identify causally related state transition sequences. When the state transition of one agent precedes that of another agent in time, and there is an interactive influence between the two, cascading propagation is determined. The identified cascading propagation relationships are arranged according to the propagation order to form cascading amplification paths. For cases with multiple cascading paths, they are ranked according to the probability of occurrence and risk consequences of each path to determine the main cascading amplification path.

[0035] Step 3.5, Calculation of dynamic risk entropy; Based on the inference trajectory of risk evolution, a dynamic risk entropy calculation method is used to obtain a quantitative index of the overall risk evolution of the system. The calculation of dynamic risk entropy is based on information entropy theory, which quantifies the uncertainty of the system's risk state into an entropy value. The specific calculation process is as follows: statistically analyze the various states that the system may reach during the inference process and their probability distributions. The definition of a state is based on the combination of states of all agents. The information entropy is calculated according to the probability distribution. The higher the entropy value, the more uncertain the risk state of the system and the more diverse the possibilities of risk evolution. The rate of change of the entropy value over time is further calculated to obtain the dynamic risk entropy index. This index reflects the activity level of risk evolution. The larger the rate of change, the faster the risk is evolving.

[0036] Through the above steps, the game theory simulation results are obtained, including the risk evolution trajectory, cascading amplification path, and dynamic risk entropy value, providing a data foundation for subsequent resource demand prediction and early warning information generation.

[0037] Furthermore, since standard game theory deduction methods are computationally too complex when dealing with a large number of agents, they may not meet the timeliness requirements of real-time early warning. Therefore, an approximate game theory solution method based on Monte Carlo tree search can be used to replace exhaustive game theory solutions. The aim is to obtain sufficiently accurate game theory deduction results within an acceptable computational time. Specifically, the Monte Carlo tree search method explores the game state space by combining random sampling and tree structure search. In each iteration, starting from the current state, the search proceeds through four stages: selection, expansion, simulation, and backtracking. In the selection stage, the most valuable branch is selected from the explored tree nodes based on a confidence upper bound algorithm. In the expansion stage, new child nodes are added to the selected leaf nodes. In the simulation stage, random deduction is performed starting from the new nodes until the termination state. In the backtracking stage, the simulation results are propagated back along the path to update the value estimates of each node. Through multiple iterations, the search tree gradually focuses on the most probable risk evolution path, obtaining near-optimal deduction results with limited computational resources.

[0038] Step 4: Based on the game theory results, quantify resource requirements and predict future changes in resource requirements to obtain a resource requirement quantification report. Step 4.1, Task classification and complexity assessment; Based on the current queue of pending early warning analysis tasks, a task classification and complexity assessment method is used to obtain the computational complexity level of each task. An early warning task classification system is established, dividing tasks into categories such as data preprocessing, threshold monitoring, statistical analysis, and model inference. For each task category, a complexity assessment index is defined: the complexity of data preprocessing tasks is related to the amount of data and the number of processing rules; the complexity of threshold monitoring tasks is related to the number of monitoring points and the complexity of judgment rules; the complexity of statistical analysis tasks is related to the analysis dimensions and time span; and the complexity of model inference tasks is related to the model size and the number of inference steps. The complexity score of each task is calculated based on the assessment index, and the score is mapped to a preset complexity level, which is divided into four levels: low complexity, medium complexity, high complexity, and ultra-high complexity.

[0039] Step 4.2, Resource Demand Mapping Calculation; Based on task complexity levels, a resource requirement mapping model is used to calculate the specific resource requirements for each task. A mapping model between early warning tasks and resource requirements is established. This model takes parameters such as task type, complexity level, and data size as input and outputs quantitative requirements for the number of processor cores, memory capacity, graphics processing unit (GPU) resources, and data throughput bandwidth. The parameters of the mapping model are calibrated using historical task execution data. Specifically, actual resource consumption data from various historical task executions is collected, and regression analysis is used to establish a mathematical relationship between task characteristics and resource consumption, yielding the coefficients of the mapping model. For new task types, initial estimations are made using analogies with similar tasks, and the model parameters are updated based on feedback data after actual execution. The default parameters of the mapping model can be set as follows: low-complexity tasks require two to four processor cores and two to four gigabytes of memory; medium-complexity tasks require four to eight processor cores and four to eight gigabytes of memory; high-complexity tasks require eight to sixteen processor cores and eight to sixteen gigabytes of memory and may require GPU acceleration; and ultra-high-complexity tasks require more than sixteen processor cores and more than sixteen gigabytes of memory and require GPU cluster support.

[0040] Step 4.3, Time Series Resource Demand Forecasting; Based on the current resource requirements and historical resource consumption data of the task queue, a time series forecasting method is used to obtain the predicted resource requirements for the near future. Time series data on system resource consumption over a past period are collected, including indicators such as processor utilization, memory usage, and network bandwidth utilization. An autoregressive integral moving average model is used to model the time series, which can capture the trend and periodic characteristics of the data. The model's order parameters are selected using an information criterion, choosing the parameter combination that minimizes the information criterion value. Based on the trained model, resource requirements for several future time steps are predicted. The prediction duration needs to consider the resource scheduling response time, typically set to two to three times the resource scheduling response time, to allow sufficient lead time for resource pre-scheduling.

[0041] Step 4.4, Resource Requirements Summary and Report Generation; Based on the resource requirements and forecasts for each task, a resource requirement aggregation and report generation method is used to obtain a quantitative resource requirement report. The resource requirements of all tasks in the current task queue are aggregated, and the total requirement for each type of resource is calculated. The aggregated result is combined with the forecasted resource requirement, and the larger of the two is taken as the recommended resource allocation target. A structured quantitative resource requirement report is generated, including a summary of resource requirements at the current moment, a forecast of short-term future resource requirements, the recommended resource allocation target, and an assessment of the urgency of resource requirements. The urgency assessment is based on the ratio of the current resource gap to available resources; a gap ratio exceeding a first urgency threshold is marked as urgent, and a ratio exceeding a second urgency threshold is marked as extremely urgent.

[0042] Furthermore, since historical data-based forecasting methods struggle to handle sudden changes in resource demand—such as the need to immediately initiate large-scale simulations upon detecting a high-risk signal—an event-driven resource demand triggering mechanism can be used as a supplement to time-series forecasting. The aim is to enable the system to respond quickly to sudden resource demands. Specifically, a triggering rule base for risk events and resource demands is established, defining the resource demand increments corresponding to various risk events. When the system detects a specific risk event, the corresponding resource demand increment is immediately triggered and superimposed on the conventional forecasting results. The triggering rules are designed based on the severity of the risk event and the complexity of the required analysis task. For example, detecting a sudden increase in the surrounding rock deformation rate triggers an increase in resource demand for a more complex simulation task. The triggering mechanism and the time-series forecasting mechanism operate in parallel, and the union of their outputs generates the final resource demand report.

[0043] Step 5: Based on the resource demand quantification report, dynamically allocate resources to obtain the resource scheduling execution results; Step 5.1, Resource Gap Analysis; Based on the resource allocation targets in the resource demand quantification report, a resource gap analysis method is used to obtain the resource allocation plan that needs to be adjusted. The current resource usage status of the data platform is obtained, including the number of allocated compute instances, the load status of each instance, and the available resource pool capacity. The resource allocation targets are compared with the current resource status to calculate the gaps or redundancies of various resources. Based on the gap analysis results, a resource allocation adjustment plan is formulated, including the number of compute instances to be increased or decreased, the memory allocation to be adjusted, and the network bandwidth configuration to be changed. For resource gap situations, the plan prioritizes ensuring the resource needs of high-priority early warning tasks; for resource redundancy situations, the plan releases excess resources while retaining a certain safety margin.

[0044] Step 5.2, container elastic scaling scheduling; Based on the computing resource requirements in the resource configuration adjustment plan, a container elastic scaling method is adopted to obtain dynamic adjustment results for computing resources. The computing resources of the data platform are deployed in a containerized manner on a container orchestration platform, with the early warning analysis service running as a microservice within container instances. When computing resources need to be increased, a scaling command is sent to the container orchestration platform. The platform creates new container instances based on predefined container images and resource configuration templates. After the new instances start, they automatically register with the service discovery component and begin undertaking tasks. When computing resources need to be reduced, a scaling command is sent to the container orchestration platform. The platform selects container instances to be terminated according to a preset scaling strategy, and terminates the instances and releases resources after ensuring that tasks on the instances are completed or migrated. The triggering conditions and scaling range for elastic scaling are differentiated based on the urgency level in the resource demand quantification report; the higher the urgency level, the faster the scaling response and the larger the scaling range.

[0045] Step 5.3, optimize hot data caching; Based on the data access requirements in the resource allocation adjustment plan, a hot data caching optimization method is adopted to obtain the optimized configuration results for data storage and access. The data access patterns involved in the current early warning task are analyzed to identify frequently accessed data sets. For identified hot data, it is preloaded from cloud storage to the high-speed cache of the edge server. The caching medium uses an in-memory database or solid-state drive to reduce data access latency. The caching strategy uses the Least Recently Used (LRU) algorithm for cache eviction, prioritizing the eviction of data with the lowest access frequency when cache space is insufficient. For data predicted to become hot data, a prefetching mechanism is used to load it into the cache in advance. The prefetching trigger condition is based on changes in construction progress and risk status. For example, when tunneling approaches a known unfavorable geological section, the historical monitoring data and geological data of that section are preloaded from the warm or cold storage layer to the hot storage layer for caching.

[0046] Step 5.4, stream processing priority adjustment; Based on the task processing requirements in the resource allocation adjustment scheme, a stream processing priority adjustment method is adopted to obtain the optimized configuration result of the data stream processing pipeline. Real-time data processing in the data platform is implemented using a stream processing framework, with various data processing tasks running as stream processing jobs. When system resources are scarce, differentiated resource allocation is required for stream processing jobs of different priorities. A priority classification system for stream processing jobs is established, dividing jobs into priority levels such as real-time alerts, near-real-time analysis, and batch processing. The resource allocation strategy of the stream processing framework is adjusted to allocate more processing slots and memory resources to high-priority jobs and reduce the resource quota for low-priority jobs. In cases of extreme resource scarcity, the execution of low-priority jobs can be paused, and all resources can be concentrated on high-priority real-time alert jobs.

[0047] Step 5.5, Monitoring and feedback of scheduling effects; Based on the resource configuration adjustment operations performed in steps 5.1 to 5.4, a scheduling effect monitoring and feedback method is adopted to obtain resource scheduling effect evaluation data. After resource scheduling is executed, the system's operating status is continuously monitored, and various performance index data are collected, including task processing latency, resource utilization, and queue backlog. Actual performance indicators are compared with expected targets to evaluate the effectiveness of resource scheduling. For cases where the expected results are not achieved, the reasons are analyzed and adjustment suggestions are generated. Possible reasons include deviations in resource demand prediction, inappropriate scheduling strategy parameters, and sudden changes in external load. The effect evaluation data is fed back to the resource demand mapping model in step 4 for online updating and optimization of model parameters.

[0048] Through the above steps, resource scheduling is completed, and the resource scheduling execution results are obtained, including resource configuration scheme, dynamic adjustment results of computing resources, optimized configuration results of data storage and access, optimized configuration results of data stream processing pipeline, and scheduling effect evaluation data, providing resource guarantee for stable system operation.

[0049] Furthermore, since starting container instances takes time, scheduling response delays may occur when resource demand surges. A resource preheating pool mechanism can be used instead of a pure on-demand scheduling mechanism to shorten resource scheduling response time and improve the system's ability to handle sudden load increases. Specifically, a certain number of spare container instances are pre-created in the data platform. These instances are in a preheating state, having completed startup and initialization but not yet undertaking actual tasks. When resource demand increases and expansion is needed, instances in the preheating pool are prioritized for use, while new instances are started to replenish the preheating pool. The size of the preheating pool is dynamically adjusted based on historical load fluctuations, maintaining a larger preheating pool size during periods of high load volatility and appropriately reducing the preheating pool size during periods of stable load to conserve resources. The resource configuration of the preheating pool instances uses a general configuration that can adapt to most types of early warning analysis tasks.

[0050] Step 6: Based on the game theory results and resource scheduling execution results, determine the risk level and generate early warning information to obtain structured early warning information; track the effect of the structured early warning information to obtain effect tracking data. Step 6.1, Risk Level Determination; Based on the dynamic risk entropy value and cascading amplification path from the game theory simulation results, a risk level determination method is adopted to obtain the current risk warning level. A risk level classification standard is established, dividing the risk status into five levels: normal, attention, warning, alert, and emergency. The risk level determination comprehensively considers multiple factors such as the dynamic risk entropy value, the rate of change of entropy value, and the severity of the cascading amplification path. The specific determination rules are as follows: when the dynamic risk entropy value is below the first risk threshold and the rate of change of entropy value is negative or close to zero, it is determined to be at the normal level; when the dynamic risk entropy value is between the first and second risk thresholds, or the rate of change of entropy value is positive but small, it is determined to be at the attention level; when the dynamic risk entropy value is between the second and third risk thresholds, or the rate of change of entropy value is positive and large, it is determined to be at the warning level; when the dynamic risk entropy value exceeds the third risk threshold, or a high-risk cascading amplification path is identified, it is determined to be at the alert level; when the dynamic risk entropy value exceeds the fourth risk threshold, or the cascading amplification path has begun to evolve, it is determined to be at the emergency level. The method for obtaining each risk threshold is as follows: based on historical risk event data, the distribution of dynamic risk entropy values ​​when events of different risk levels occur is statistically analyzed, and the threshold boundaries of each level are determined by statistical analysis methods. The default values ​​can be calibrated according to specific engineering conditions.

[0051] Step 6.2: Structured generation of early warning information; Based on the risk level assessment results and detailed information from the game theory simulation, a structured early warning information generation method is adopted to obtain structured early warning information. The early warning information is organized in a structured format and includes the following fields: Early Warning Number, a unique identifier generated by combining a timestamp and a sequence number; Early Warning Time, recording the precise moment the early warning information was generated; Risk Type, determined based on the type of dominant risk factor in the game theory simulation, possible types include surrounding rock instability risk, sudden water inrush risk, support failure risk, and compound risks; Risk Location, determined based on the comprehensive risk field distribution data, describing the spatial area with the highest risk using tunnel mileage and cross-section location; Risk Level, using the level determined in step 61; Risk Trend, judging whether the risk tends to worsen, remain stable, or tend to mitigate based on the rate of change of dynamic risk entropy; Cascading Risk Warning, providing a description of possible risk propagation chains if a cascading amplification path is identified; and Recommended Measures, matching corresponding response suggestions from a pre-set measure library based on the risk type and level.

[0052] Step 6.3, Differentiated Information Push; Based on structured early warning information and the characteristics of the recipients, a differentiated information push method is adopted to obtain early warning information versions for different recipients. An information push strategy is established according to the recipients of the early warning information, including construction site workers, project managers, owners, supervision units, and emergency management departments. Differentiated early warning information versions are generated based on the information needs and receiving capabilities of different recipients: the version for on-site workers highlights the location and urgency of the risk, using a concise and intuitive expression; the version for managers includes complete risk analysis information and recommended measures; and the version for emergency management departments focuses on the risk level and the potential scope of impact. Appropriate push channels and methods are selected according to the early warning level. Normal and attention levels are displayed through the system interface, early warning levels are notified via message push, and alarm and emergency levels are pushed simultaneously through multiple channels and trigger audible and visual alarms.

[0053] Step 6.4: Recording and tracking the effects of early warning logs; Based on the release records of structured early warning information, an early warning log recording and effect tracking method is used to obtain the operational log data of the early warning system. The entire process of each early warning message's generation, release, reception, and response is logged, including the complete content of the early warning message, release time, push channel, reception confirmation status, and subsequent response measures. An early warning effect tracking mechanism is established to continuously track the evolution of actual risks after early warning release, comparing early warning predictions with actual results. The accuracy of the early warning is evaluated based on the comparison results, and indicators such as early warning hit rate, false alarm rate, and missed alarm rate are statistically analyzed. The evaluation results are fed back to the multi-agent game theory model and risk level determination rules for continuous optimization of the model and rules.

[0054] Furthermore, since fixed risk level determination rules may not be suitable for the differences in risk characteristics across different engineering projects and construction stages, an adaptive risk level determination method based on case-based reasoning can be used to replace fixed rule determination. The aim is to enable dynamic adjustment of risk level determination based on specific engineering conditions and historical experience. Specifically, a risk case database is established to collect characteristic data and actual consequences of various risk events in history. When risk level determination is required, the current risk characteristics are matched with historical cases in the database to identify the most similar historical cases. Based on the actual risk level and consequences of similar cases, a weighted voting method is used to determine the risk level of the current situation. The similarity calculation considers multiple dimensions such as geological conditions, construction methods, and combinations of risk factors. As the system continues to accumulate new cases, the determination capability of the case database is continuously enhanced.

[0055] Step 7: Based on structured early warning information and effect tracking data, continuously optimize the multi-agent game inference model and system parameters to obtain the optimized early warning scheme; Step 7.1, Accuracy assessment of early warnings; Based on early warning effect tracking data, an early warning accuracy evaluation method was adopted to obtain the performance evaluation results of the multi-agent game inference model. All early warning information release records and corresponding actual risk evolution results were collected over a period of time to establish a correspondence between early warnings and actual results. Early warning accuracy indicators were calculated, including the true positive rate (the proportion of correct early warnings), the false positive rate (the proportion of false alarms), and the false negative rate (the proportion of missed warnings). The differences in early warning accuracy for different risk types and risk levels were analyzed to identify weaknesses in the multi-agent game inference model. For risk types or levels with low accuracy, the reasons were analyzed in depth, including inaccurate parameters of the multi-agent game inference model, improper risk level threshold settings, and data quality issues.

[0056] Step 7.2, Optimize the parameters of the game theory model; Based on the accuracy assessment results of the early warning system, a game theory inference model parameter optimization method was adopted to obtain the updated model parameters. For the agent behavior rules and interaction influence functions in the multi-agent game theory inference model, parameter correction was performed based on actual risk evolution data. Specifically, the actual evolution trajectories of historical risk events were used as training samples, and parameter estimation methods were used to inversely deduce the model parameters that best fit the actual trajectories. Parameter optimization employed gradient descent or evolutionary algorithms, with the objective function being to minimize the deviation between the model's inferred trajectory and the actual trajectory. During the optimization process, reasonable range constraints were set for the parameters to avoid values ​​exceeding the physically permissible range. After optimization, the new parameters were deployed into the multi-agent game theory inference model, replacing the original parameters.

[0057] Step 7.3, risk level threshold adaptively adjusted; Based on the accuracy assessment results of the early warning system, an adaptive adjustment method for risk level thresholds was adopted to obtain updated risk level judgment thresholds. The characteristics of false alarms and missed alarms were analyzed to identify threshold setting issues leading to judgment bias. For cases with high false alarm rates, the corresponding risk thresholds were appropriately increased to reduce overly sensitive early warnings; for cases with high missed alarm rates, the corresponding risk thresholds were appropriately decreased to increase the sensitivity of early warnings. A gradual strategy was adopted for threshold adjustments, with each adjustment controlled within a certain range to avoid drastic changes in system behavior due to large adjustments. The adjusted thresholds officially took effect after a period of trial operation and verification.

[0058] Step 7.4, optimize resource scheduling strategy; Based on system operation data, a resource scheduling strategy optimization method is adopted to obtain updated scheduling strategy parameters. Historical data on resource scheduling execution is collected, including scheduling decisions, execution results, and actual effects. The efficiency and effectiveness of resource scheduling are analyzed to identify shortcomings in the scheduling strategy, such as resource demand prediction deviations, insufficient scaling response speed, and low resource utilization. Corresponding strategy parameters are adjusted to address the identified problems, such as adjusting the coefficients of the resource demand mapping model, adjusting the trigger threshold for elastic scaling, and adjusting the size of the preheating pool. The adjustment of strategy parameters also adopts a gradual approach, and the adjustment effect is verified through comparative experiments.

[0059] Step 7.5, System Version Management and Canary Release; Based on the updated model and parameters, a system version management and canary release approach is adopted to obtain a secure and controllable system update. The updated model parameters and strategy configurations are packaged into a new system version, and the version changes and reasons are recorded. A canary release strategy is used to gradually promote the new version, conducting trial runs on a small scale to monitor its performance. If the new version performs normally and is superior to the old version, its application is gradually expanded until it completely replaces the old version. If problems occur with the new version, it can be quickly rolled back to the old version to ensure stable system operation. A version history is established to support the tracing and analysis of the system evolution process.

[0060] This step outputs the optimized early warning scheme, including optimized model parameters, risk level thresholds, strategy parameters, and system version configuration.

[0061] Furthermore, since single parameter optimization methods may get stuck in local optima, making it difficult to achieve continuous improvement in model performance, ensemble learning multi-model fusion methods can be used to replace single-model optimization. The aim is to improve the overall performance and robustness of the early warning system through the collaborative work of multiple models. Specifically, multiple multi-agent game simulation models with different configurations are maintained simultaneously, each using different parameter settings or different algorithm variants. During risk simulation, multiple models execute in parallel, each outputting its own simulation results. Weighted voting or stacked generalization methods are used to fuse the outputs of multiple models to obtain the final simulation result. The weights of each model are dynamically adjusted based on its historical early warning accuracy, with models with higher accuracy receiving higher weights. This multi-model fusion mechanism can reduce the risk caused by the bias of a single model and improve the reliability of early warning results.

[0062] A tunnel construction risk early warning system based on a data platform, such as Figure 3 As shown, a tunnel construction risk early warning method based on a data platform, as described above, includes: The data access module is used to acquire multi-source monitoring data from the tunnel construction site, perform preprocessing, and obtain a time-aligned standardized data set. The risk field construction module constructs a spatial coordinate system and performs multi-physics coupling and correlation based on a time-aligned standardized dataset; it then calculates the coupled multi-physics fields to obtain comprehensive risk field distribution data. The game simulation module, based on comprehensive risk field distribution data, constructs a multi-agent game simulation model to simulate the interaction and evolution process between various risk factors and obtain game simulation results. The resource quantification module quantifies resource demand based on game theory results, predicts future changes in resource demand, and generates a resource demand quantification report. The resource scheduling module dynamically configures resources based on the resource demand quantification report and obtains the resource scheduling execution results; The early warning generation module, based on the game theory simulation results and resource scheduling execution results, determines the risk level and generates early warning information to obtain structured early warning information; it also tracks the effect of the structured early warning information to obtain effect tracking data. The optimization module continuously optimizes the multi-agent game inference model and system parameters based on structured early warning information and effect tracking data to obtain an optimized early warning scheme.

[0063] This application provides a computer-readable storage medium for storing computer-readable instructions, which, when read by a computer, can execute a tunnel construction risk early warning method based on a data platform as described above.

[0064] A tunnel construction risk early warning method based on a data platform can be implemented wholly or partially through software, hardware, firmware, or any combination thereof. If implemented in software, the functionality can be stored as one or more instructions or code on a computer-readable storage medium. Computer-readable storage media include computer storage media and communication media, wherein communication media include any medium that facilitates the transfer of computer programs from one location to another. Storage media can be any available medium accessible to general-purpose or special-purpose computers.

[0065] In one alternative embodiment, the computer-readable storage medium may include, but is not limited to, random access memory, read-only memory, electrically erasable programmable read-only memory, or flash memory, optical disk storage, magnetic disk storage or other magnetic storage devices, or any other medium capable of storing desired program code in the form of instructions or data structures and accessible by a general-purpose or special-purpose computer or a general-purpose or special-purpose processor.

[0066] In one embodiment of the present invention, a specific example is provided: A high-speed railway tunnel project traverses a karst development area. Various types of monitoring equipment were deployed at the construction site. Examples of some of the monitoring data collected by the system are shown in Table 1. Table 1: Examples of some monitoring data; The system performs fusion processing on the above multi-source data, and the resulting comprehensive risk field distribution data is shown in Table 2. Table 2: Example of comprehensive risk field distribution data; Based on the aforementioned risk field data, the system initiated a multi-agent game simulation to model the interactive evolution process among the rock instability agent, seepage agent, and support agent. The system identified a potential cascading amplification path: enhanced seepage leads to rock softening, which intensifies deformation, ultimately increasing the support load. When the dynamic risk entropy value calculated from the game simulation results exceeded the warning threshold, the system generated a warning message and pushed it to relevant personnel, recommending increased monitoring frequency for this section and preparation for emergency response.

[0067] The embodiments of the present invention have been described above. However, the embodiments are not limited to the specific implementation methods described above. The specific implementation methods described above are merely illustrative and not restrictive. Those skilled in the art can make more equivalent embodiments under the guidance of the present embodiments, and all of them are within the protection scope of the present embodiments.

Claims

1. A method for early warning of tunnel construction risks based on a data platform, characterized in that, include: Acquire multi-source monitoring data from the tunnel construction site, perform preprocessing, and obtain a time-aligned standardized dataset; Based on the time-aligned standardized data set, a spatial coordinate system is constructed and multi-physics coupling correlation is performed; The coupled multiphysics field is calculated to obtain the comprehensive risk field distribution data; Based on the comprehensive risk field distribution data, a multi-agent game inference model is constructed to simulate the interaction and evolution process between various risk factors and obtain the game inference results. Based on the game theory results, resource demand is quantified, and future changes in resource demand are predicted to obtain a resource demand quantification report. Based on the resource demand quantification report, resources are dynamically configured to obtain resource scheduling execution results; Based on the game theory results and resource scheduling execution results, risk level determination and early warning information generation are performed to obtain structured early warning information; The structured early warning information is subjected to effect tracking to obtain effect tracking data; Based on the structured early warning information and effect tracking data, the multi-agent game inference model and system parameters are continuously optimized to obtain an optimized early warning scheme.

2. The tunnel construction risk early warning method based on a data platform according to claim 1, characterized in that, The acquisition and preprocessing of multi-source monitoring data at the tunnel construction site includes: A categorized data acquisition strategy is adopted, which uses ground-penetrating radar to acquire time-domain waveform data of electromagnetic wave reflection signals, surrounding rock deformation monitoring equipment to acquire deformation data, stress-strain monitoring equipment to acquire mechanical parameter data, environmental monitoring equipment to acquire environmental parameter data, and video surveillance equipment to acquire real-time image frame data of the construction site. The resulting raw data stream from the monitoring equipment includes time-domain waveform data, deformation data, mechanical parameter data, environmental parameter data, and real-time image frame data. The raw data stream from the monitoring equipment is parsed for heterogeneous data formats, converting binary waveform data into numerical matrix form and raw sampled values ​​into engineering unit values. Establish a data quality detection rule base, set reasonable range thresholds for various types of data in the raw data stream of the monitoring equipment, and correct abnormal values ​​in the raw data stream of the monitoring equipment that exceed the reasonable range thresholds; The timestamps of ground-penetrating radar detection equipment, surrounding rock deformation monitoring equipment, stress and strain monitoring equipment, environmental monitoring equipment, and video surveillance equipment are calibrated using the Network Time Protocol. A resampling method is used to unify data sources with different sampling frequencies to a preset standard sampling period.

3. The tunnel construction risk early warning method based on a data platform according to claim 1, characterized in that, The construction of the spatial coordinate system and the multiphysics coupling association include: A local coordinate system is established with the tunnel axis as the reference. The installation positions of the geological radar detection equipment, surrounding rock deformation monitoring equipment, stress and strain monitoring equipment, environmental monitoring equipment, and video surveillance equipment are transformed to a unified coordinate system. Spatial grid units are divided according to the tunnel cross-section shape and the layout of monitoring points. The displacement values ​​of discrete monitoring points are extended into a continuous deformation field distribution using the Kriging interpolation method, and the stress values ​​of discrete monitoring points are extended into a continuous stress field distribution using the inverse distance weighted interpolation method. Establish a constitutive correlation model between the deformation field and the stress field, a fluid-structure interaction model between the seepage field and the stress field, and a thermo-mechanical interaction model between the temperature field and the deformation field, stress field, and seepage field.

4. The tunnel construction risk early warning method based on a data platform according to claim 1, characterized in that, The steps to obtain the comprehensive risk field distribution data include: Define the calculation rules for the comprehensive risk index, and convert the state values ​​of the deformation field, stress field, seepage field and temperature field into dimensionless risk contribution values ​​through normalization. The corresponding weighting coefficients are determined based on the degree of influence of deformation field, stress field, seepage field and temperature field on the overall risk. The relative importance of each risk factor is analyzed by using the analytic hierarchy process (AHP). The weight values ​​are obtained by constructing a judgment matrix and calculating the eigenvectors. The risk contribution values ​​of the deformation field, stress field, seepage field, and temperature field are weighted and summed to obtain the comprehensive risk field distribution data.

5. The tunnel construction risk early warning method based on a data platform according to claim 1, characterized in that, The construction of the multi-agent game inference model includes: The risk of surrounding rock instability is abstracted into a surrounding rock instability intelligent agent, the risk of groundwater seepage is abstracted into a seepage intelligent agent, and the response of the support structure is abstracted into a support intelligent agent. The initial states of the surrounding rock instability intelligent agent, the seepage intelligent agent, and the support intelligent agent are assigned based on the comprehensive risk field distribution data; Define the action space of the surrounding rock instability agent, the seepage agent, and the support agent, and the interaction function between the agents; Define the revenue functions for the surrounding rock instability agent, the seepage agent, and the support agent.

6. The tunnel construction risk early warning method based on a data platform according to claim 1, characterized in that, The steps to obtain the game theory deduction results include: A rapid coarse-grained inference is performed, using a simplified agent state space and a first preset time step, to obtain a preliminary judgment on the risk evolution trend; When the rapid simulation identifies that the comprehensive risk index exceeds the first warning threshold, a fine-grained simulation is initiated, using a complete agent model and a second preset time step. Analyze the temporal relationship of the state transitions of the surrounding rock instability intelligent agent, the seepage intelligent agent, and the support intelligent agent during the simulation process, identify the state transition sequences with causal relationships, and form a cascaded amplification path; Dynamic risk entropy is calculated based on information entropy theory. The state combinations and probability distributions during the statistical deduction process are calculated. The state combinations are defined based on the state combinations of the surrounding rock instability intelligent agent, the seepage intelligent agent, and the support intelligent agent. The rate of change of entropy value over time is calculated.

7. The tunnel construction risk early warning method based on a data platform according to claim 1, characterized in that, The quantification of the aforementioned resource requirements includes: Establish a classification system for early warning tasks, dividing tasks into data preprocessing, threshold monitoring, statistical analysis, and model inference, and calculate the complexity score of the tasks based on evaluation indicators; Establish a mapping model between early warning tasks and resource requirements. Take task type, complexity level, and data scale parameters as input, and output quantitative requirements for the number of computing processor cores, memory capacity, graphics processor resources, and data throughput bandwidth. An autoregressive integral moving average model is used to model the resource consumption time series, and the resource demand for several future time steps is predicted based on the trained model.

8. The tunnel construction risk early warning method based on a data platform according to claim 1, characterized in that, The dynamic configuration of resources includes: Compare the resource allocation goals with the current resource status, calculate the gaps or redundancies of various resources, and formulate resource allocation adjustment plans. The container elastic scaling method is used to send expansion or shrinkage commands to the container orchestration platform to create or terminate container instances based on predefined container images and resource configuration templates. The identified hot data is preloaded from cloud storage to the high-speed cache of the edge server, and the caching strategy uses the least recently used algorithm for cache eviction. Establish a priority classification system for stream processing jobs and adjust the resource allocation strategy of the stream processing framework to allocate more processing slots and memory resources to jobs with a preset priority threshold.

9. The tunnel construction risk early warning method based on a data platform according to claim 1, characterized in that, The risk level determination and early warning information generation include: Establish risk level classification standards, classifying risk status into five levels: normal, attention, warning, alert, and emergency. The determination of risk level comprehensively considers dynamic risk entropy value, entropy value change rate, and the severity of cascading amplification path. The early warning information is organized in a structured format; Establish an information push strategy based on the recipients of the early warning information, and generate differentiated versions of early warning information according to the information needs and receiving capabilities of different recipients; Each warning message is logged throughout its entire process, and a mechanism for tracking the effectiveness of warnings is established to compare the warning predictions with the actual results.

10. A tunnel construction risk early warning system based on a data middle platform, characterized in that, A method for implementing a tunnel construction risk early warning system based on a data platform as described in any one of claims 1-9 includes: The data access module is used to acquire multi-source monitoring data from the tunnel construction site, perform preprocessing, and obtain a time-aligned standardized data set. The risk field construction module constructs a spatial coordinate system and performs multi-physics coupling and correlation based on the time-aligned standardized data set; it then calculates the coupled multi-physics fields to obtain comprehensive risk field distribution data. The game simulation module, based on the comprehensive risk field distribution data, constructs a multi-agent game simulation model to simulate the interaction and evolution process between various risk factors and obtain the game simulation results. The resource quantification module quantifies resource demand based on the game theory results and predicts future changes in resource demand to obtain a resource demand quantification report. The resource scheduling module dynamically configures resources based on the resource demand quantification report to obtain the resource scheduling execution result; The early warning generation module, based on the game simulation results and resource scheduling execution results, determines the risk level and generates early warning information to obtain structured early warning information; it then tracks the effect of the structured early warning information to obtain effect tracking data. The optimization module continuously optimizes the multi-agent game inference model and system parameters based on the structured early warning information and effect tracking data to obtain an optimized early warning scheme.

Images