Highway tunnel decision-making maintenance system and maintenance method

Abstract

The present invention provides a highway tunnel decision-making and maintenance system and a maintenance method thereof, which obtains tunnel static and dynamic data and geological environment data around the tunnel; based on the geological environment data, adopts a geostatistical simulation algorithm and a time series decomposition algorithm to perform geological risk analysis to obtain a safety risk assessment result of the tunnel structure affected by geological factors and a groundwater dynamic change trend prediction result; based on the geological risk analysis result, adopts a heterogeneous integration model of a dynamic Bayesian network and an extreme gradient boosting tree to fuse the tunnel technical status data, predict the tunnel technical status, and generate maintenance decision parameters; adopts a cuckoo search algorithm to optimize the maintenance decision parameters to minimize the maintenance cost under the influence of geological risk; and outputs a decision result including the total maintenance cost allocation based on geological risk, the expected improvement of technical status, the predicted value of the rock formation stability safety factor, and the groundwater seepage risk warning level.

Description

Technical Field

[0001] The present invention relates to the field of tunnel decision-making and maintenance, and in particular to a highway tunnel decision-making and maintenance system and a maintenance method thereof. Background Art

[0002] As a critical component of transportation infrastructure, highway tunnel maintenance and management are directly linked to operational safety and service life. Traditional tunnel maintenance relies on manual inspections and empirical judgment, resulting in data fragmentation, delayed risk assessments, and insufficiently scientific decision-making. This makes it difficult to address the potential threats posed by complex geological conditions to tunnel structures. With the advancement of the Internet of Things, big data, and artificial intelligence technologies, there is an urgent need to build intelligent maintenance systems that enable comprehensive management of static and dynamic tunnel data and geological environmental data.

[0003] In existing technologies, although some systems have achieved data visualization and simple maintenance planning, they lack the ability to conduct in-depth analysis of geological risks, multi-model collaborative prediction, and global optimization. In particular, there are technical gaps in integrating the evolution laws of the geological environment to generate cross-cycle maintenance plans and quantifying the impact of uncertainty on maintenance decisions.

[0004] Therefore, a highway tunnel decision-making and maintenance system was developed that integrates geological risk assessment, technical condition prediction, maintenance decision optimization and visualization output. Summary of the Invention

[0005] The main purpose of the present invention is to provide a highway tunnel decision-making maintenance system and maintenance method thereof, so as to solve the problems of traditional tunnel maintenance relying on manual inspection and experience judgment, data fragmentation, delayed risk assessment, and insufficient scientific decision-making.

[0006] To solve the above technical problems, the technical solution adopted by the present invention is: a highway tunnel decision-making and maintenance method, the method comprising: S1, acquiring static and dynamic data of the tunnel and geological environment data around the tunnel;

[0007] S2. Based on the geological environment data, a geological statistical simulation algorithm and a time series decomposition algorithm are used to perform a geological risk analysis to obtain a safety risk assessment result of the tunnel structure affected by geological factors and a prediction result of the groundwater dynamic change trend;

[0008] S3. Based on the geological risk analysis results, a heterogeneous integrated model of a dynamic Bayesian network and an extreme gradient boosting tree is used to integrate the tunnel technical condition data, predict the tunnel technical condition, and generate maintenance decision parameters;

[0009] S4. Optimizing the maintenance decision parameters using a cuckoo search algorithm to minimize the maintenance cost under the influence of geological risks;

[0010] S5. The output includes the decision results of the total maintenance cost allocation based on geological risks, the expected improvement of technical conditions, the predicted value of the rock stability safety factor, and the groundwater seepage risk warning level.

[0011] In the preferred embodiment, in step S2, geological risk analysis is performed to obtain rock formation type, fracture development degree, groundwater level and seepage rate parameters;

[0012] Use Sequential Gaussian Simulation (SGS) to construct a spatial probability model of rock structure and groundwater distribution;

[0013] Assess the safety risk level of the tunnel structure affected by geological factors;

[0014] Use the seasonal adjustment method X-13 to predict groundwater dynamic trends and identify potential leakage or rock slip risks.

[0015] In the preferred solution, the geological risk analysis method is specifically as follows:

[0016] Through a distributed sensor network and drilling sampling, the rock type, fracture development, groundwater level, and seepage rate parameters around the tunnel are obtained and pre-processed to remove outliers, fill in missing values, and perform standardization.

[0017] Based on the pre-processed geological parameters, the sequential Gaussian simulation algorithm is used to transform the parameters into a normal distribution. The spatial autocorrelation is analyzed through the variogram and the theoretical model is fitted. The points to be estimated are traversed along a random path. The mean and variance are estimated through Kriging interpolation based on the conditional distribution of the simulated points. The simulated values ​​are randomly sampled to generate multiple three-dimensional spatial probabilistic realizations of rock structure and groundwater distribution.

[0018] Key parameters are extracted from simulation results to construct a hierarchical risk indicator system. Fuzzy comprehensive evaluation is used to assign weights to indicators and determine membership functions. Geological parameters are mapped to risk levels, and a three-dimensional risk heat map is generated in combination with tunnel structure design parameters.

[0019] Preprocessing groundwater level and seepage rate time series data, using seasonal adjustment to decompose them into trend, seasonal, and irregular components. An ARIMA model is fitted to the trend component, and the seasonal component is periodically extrapolated. Combined with the irregular component, multi-scenario forecast results are generated, and thresholds are set to identify potential leakage or rock slip risks.

[0020] The spatial probability model is integrated with the time series prediction results to establish a four-dimensional dynamic risk model. The uncertainty is quantified through Monte Carlo simulation, and an interactive interface is developed to achieve multi-dimensional visualization of risk results and output reports.

[0021] In a preferred embodiment, predicting the tunnel technical condition and generating maintenance decision parameters in step S3 includes: updating the conditional probability distribution of the tunnel technical condition prediction using a dynamic Bayesian network based on the geological risk assessment results;

[0022] Integrate rock stability, seepage data and disease history records, and use extreme gradient boosting tree to construct a maintenance decision tree;

[0023] The dynamic Bayesian network deduces the probability of technical status transition over many years and generates a cross-cycle maintenance plan.

[0024] In the preferred solution, the geological risk analysis results are aligned with the tunnel technical status data in time and space to construct a multi-source heterogeneous feature set containing static, dynamic and historical features. After discretization, coding and feature screening, the input data set is formed;

[0025] Construct a dynamic Bayesian network, define hidden state variables and observable variables, build a conditional probability table based on geological risk factors, use the expectation maximization algorithm to estimate parameters, calculate the hidden state posterior probability through the forward-backward algorithm, and establish a technical status transition probability matrix;

[0026] An extreme gradient boosting tree heterogeneous integration model was constructed. This model takes geological risk data, historical technical status data, and maintenance measures records as inputs. A multi-objective loss function with a regularization term was used to iteratively generate decision trees through gradient boosting. A spatial weight matrix was introduced to adjust sample weights to address spatial correlation.

[0027] A stacking ensemble strategy was used to fuse the dynamic Bayesian network and the extreme gradient boosting tree model. The weight parameters were optimized through cross-validation. The expected utility was calculated based on the prediction results, and the ε-greedy strategy was used to select the optimal maintenance measure.

[0028] Based on the dynamic Bayesian network, the multi-period technical status transition probability is recursively calculated. A multi-stage decision optimization model is constructed and resource constraints are introduced. The optimal maintenance strategy sequence across the period is solved using strategy iteration and Lagrangian relaxation method.

[0029] The prediction uncertainty is quantified through Monte Carlo simulation, and the characteristic contribution is analyzed using Shapley value. The maintenance decision parameters and scenario analysis results containing probability distribution are output.

[0030] In the preferred embodiment, step S4 optimizes the maintenance decision parameters, including:

[0031] Determining the maintenance decision parameters including maintenance priorities and measure combinations;

[0032] A cuckoo search algorithm is used to globally optimize the maintenance priorities and measure combinations to minimize the maintenance costs under the influence of geological risks;

[0033] The maintenance decision parameters are modeled as a maintenance priority vector ( ) and the measure combination vector ( ) , construct a function with the weighted sum of maintenance cost and geological risk as the target , and set resource constraints , priority normalization constraint and mutually exclusive constraints on measures;

[0034] Initialize the cuckoo search algorithm population and update the solution vector through the Levy flight mechanism, where the Levy step size follows a stable distribution ( , ), the update formula is , perform S-shaped transformation on maintenance priority , the measures are updated using roulette wheel selection;

[0035] By probability Execute the nest elimination mechanism to generate new solutions through random perturbations;

[0036] Perform constraint repair on the updated solution and calculate the cost-benefit ratio when the budget is violated. Adjust the priority in descending order, passing when normalization is violated Repair; among them, is the cost function of the measure, which means that the maintenance measures are taken for the i-th section Cost;

[0037] Calculate population diversity during the iteration process When the probability of discovery is lower than the threshold, the optimal solution is output after the algorithm terminates. , extract maintenance priority and combination of measures , calculate the total maintenance cost , risk-weighted cost and cost-effectiveness ratio and through sensitivity analysis and robustness indicators Evaluate solution stability, For the The population diversity index at the iteration, is the population size, is the total number of simulated geological risk parameter fluctuation scenarios, For the Optimal measures for each segment maintenance costs.

[0038] In a preferred embodiment, in step S5, the total maintenance cost distribution, the expected improvement of the technical condition, the predicted value of the rock formation stability safety factor, and the groundwater seepage risk warning level are output;

[0039] Among them, construct a four-dimensional decision result matrix , generate and visualize maintenance decision parameters through the following steps: Based on the optimized maintenance priority and cost estimates of measures Calculate section maintenance costs and total cost , generate the total maintenance cost allocation matrix by geological risk level group ;

[0040] Using the dynamic Bayesian network state transition probability matrix Computing Technology Status Improvement Index and global improvement index , forming a matrix of expected improvements in technical conditions ;

[0041] Extract statistical parameters of rock compressive strength from sequential Gaussian simulation results and calculate safety factors based on tunnel loads , generate the predicted mean and confidence interval through Monte Carlo simulation, and construct the rock stability safety factor prediction value matrix ;

[0042] The predicted values ​​of groundwater level and seepage rate are used as inputs of the fuzzy logic system, and the risk level sequence is output through Gaussian membership function and rule base reasoning. , forming a groundwater seepage risk warning level matrix ;

[0043] Develop a 3D GIS interface to visualize the results of each dimension in the form of ring charts, heat maps, cloud maps, and dynamic arrows, and support interactive queries in the time dimension;

[0044] Calculating cost forecast error rate and risk level prediction accuracy , generate a PDF report containing uncertainty indicators and establish a feedback optimization mechanism.

[0045] is the cost forecast error rate, is the total maintenance cost forecast value output by the system, is the actual total cost after the implementation of the maintenance measures, Accuracy of risk level prediction.

[0046] In the preferred embodiment, the maintenance measures management includes:

[0047] Manage maintenance measures based on disease name, maintenance type, maintenance measures and unit price;

[0048] By combining the non-parametric survival analysis algorithm with the rock formation life cycle data, the effectiveness attenuation curve of different maintenance measures in delaying the development of geological diseases is quantified;

[0049] Computational geometry algorithms are used to automatically optimize the repair length threshold according to the geological boundaries of the construction area.

[0050] In a preferred solution, the acquisition of tunnel static and dynamic data includes:

[0051] Data collection is performed through the APP, which includes inputting the pile number, selecting the inspection item, selecting the disease, recording the disease attributes, and recording the disease assessment information;

[0052] The data collected by the data is encrypted for transmission using the elliptic curve encryption algorithm ECC;

[0053] Implementing fuzzy query of the disease records through suffix automaton algorithm;

[0054] Achieve incremental real-time synchronization between the APP and the Web through a two-way differential synchronization algorithm;

[0055] The spatial distribution map of hazards integrating geological risk distribution is drawn using the spline interpolation algorithm.

[0056] A highway tunnel decision-making and maintenance system, which includes: a cockpit, maintenance decision-making, road property management, system management, and structure inspection modules;

[0057] Road property management module, used to maintain basic tunnel data;

[0058] Structural inspection module, used to obtain tunnel site data;

[0059] A maintenance decision module, configured to generate tunnel maintenance decision parameters based on the tunnel basic data and the tunnel site data;

[0060] A cockpit module, for visually displaying the tunnel basic data, the tunnel site data, and the tunnel maintenance decision parameters;

[0061] System management module, used for system management;

[0062] Among them, the road property management module, the structure inspection module, the maintenance decision module, the cockpit module and the system management module operate in coordination to form a data-driven tunnel maintenance management closed loop.

[0063] The present invention provides a highway tunnel decision-making and maintenance system and maintenance method. The highway tunnel decision-making and maintenance system and maintenance method collect geological parameters such as the rock type, fracture development level, groundwater level and seepage rate around the tunnel, and combine them with advanced algorithms to achieve multi-dimensional risk assessment and precise maintenance decision-making. The specific beneficial effects are as follows: a spatial probability model of rock layer and groundwater distribution is constructed through a sequential Gaussian simulation algorithm, the impact of geological factors on tunnel structure is quantified, and a three-dimensional visual assessment of safety risk level is achieved, changing the subjectivity and one-sidedness of traditional manual assessment;

[0064] Use seasonal adjustment methods to decompose groundwater dynamic data into time series, predict seepage trends and potential risk points, identify leakage or rock slip risks in advance, and transform passive maintenance into active prevention;

[0065] By integrating geological risk assessment results with tunnel technical condition data and using a dynamic Bayesian network and extreme gradient boosting tree model, we can achieve intelligent technical condition prediction and maintenance decision parameter generation, improving prediction accuracy and plan rationality.

[0066] A cuckoo search algorithm is used to globally optimize maintenance priorities and measure combinations. Combining geological risk weights with resource constraints, this minimizes maintenance costs while reducing the impact of geological risks, achieving scientific resource allocation. Decision results, such as maintenance cost allocation and expected technical condition improvements, are output and dynamically displayed through a 3D geographic information system, providing managers with intuitive and comprehensive decision-making evidence and promoting the transformation of maintenance management from experience-driven to data-driven.

[0067] Establish a closed-loop mechanism for data collection, risk analysis, decision optimization, and feedback improvement to continuously improve the system's adaptability to complex geological environments, ensure tunnel operation safety, and extend its service life. BRIEF DESCRIPTION OF THE DRAWINGS

[0068] The present invention will be further described below with reference to the accompanying drawings and examples:

[0069] Figure 1 This is a flow chart of the highway tunnel decision-making and maintenance method of the present invention;

[0070] Figure 2 It is a block diagram of the highway tunnel decision-making and maintenance system of the present invention. DETAILED DESCRIPTION

[0071] Example 1

[0072] like Figure 1-2 As shown, a highway tunnel decision-making and maintenance method includes: S1, obtaining tunnel static and dynamic data and tunnel surrounding geological environment data;

[0073] S2. Based on the geological environment data, a geological statistical simulation algorithm and a time series decomposition algorithm are used to perform a geological risk analysis to obtain a safety risk assessment result of the tunnel structure affected by geological factors and a prediction result of the groundwater dynamic change trend;

[0074] S3. Based on the geological risk analysis results, a heterogeneous integrated model of a dynamic Bayesian network and an extreme gradient boosting tree is used to integrate the tunnel technical condition data, predict the tunnel technical condition, and generate maintenance decision parameters;

[0075] S4. Optimizing the maintenance decision parameters using a cuckoo search algorithm to minimize the maintenance cost under the influence of geological risks;

[0076] S5. The output includes the decision results of the total maintenance cost allocation based on geological risks, the expected improvement of technical conditions, the predicted value of the rock stability safety factor, and the groundwater seepage risk warning level.

[0077] First, static and dynamic tunnel data and surrounding geological environment data are obtained (S1). Then, the geological environment data are analyzed using a geostatistical simulation algorithm and a time series decomposition algorithm to obtain safety risk assessment results and groundwater dynamic change trend prediction results (S2). Based on the geological risk analysis results, a heterogeneous integration model of a dynamic Bayesian network and an extreme gradient boosting tree is used to integrate tunnel technical condition data, predict the tunnel technical condition, and generate maintenance decision parameters (S3). A cuckoo search algorithm is then used to globally optimize the maintenance decision parameters to minimize the maintenance cost under the influence of geological risks (S4). Finally, the decision results are output, including the total maintenance cost allocation based on geological risks, the expected improvement in technical conditions, the predicted value of the rock stability safety factor, and the groundwater seepage risk warning level (S5).

[0078] Through the integration of multiple algorithms, quantitative assessment and dynamic prediction of geological risks can be achieved, improving the scientific nature of maintenance decisions. Heterogeneous integrated models are used to combine geological risk and technical status data to improve the accuracy of tunnel technical status predictions. A global optimization algorithm is used to balance maintenance costs and geological risks to achieve rational resource allocation. The output of multi-dimensional decision results provides comprehensive data support for maintenance management, promoting the transformation of maintenance models to intelligent and proactive ones, ensuring tunnel operation safety and optimizing maintenance resource utilization.

[0079] Example 2

[0080] Further illustrate with reference to Example 1, Figure 1-2 As shown, in step S2, geological risk analysis is performed to obtain rock layer type, fracture development degree, groundwater level and seepage rate parameters;

[0081] Use Sequential Gaussian Simulation (SGS) to construct a spatial probability model of rock structure and groundwater distribution;

[0082] Assess the safety risk level of the tunnel structure affected by geological factors;

[0083] Use the seasonal adjustment method X-13 to predict groundwater dynamic trends and identify potential leakage or rock slip risks.

[0084] In the preferred solution, the geological risk analysis method is specifically as follows:

[0085] Through a distributed sensor network and drilling sampling, the rock type, fracture development, groundwater level, and seepage rate parameters around the tunnel are obtained and pre-processed to remove outliers, fill in missing values, and perform standardization.

[0086] Based on the pre-processed geological parameters, the sequential Gaussian simulation algorithm is used to transform the parameters into a normal distribution. The spatial autocorrelation is analyzed through the variogram and the theoretical model is fitted. The points to be estimated are traversed along a random path. The mean and variance are estimated through Kriging interpolation based on the conditional distribution of the simulated points. The simulated values ​​are randomly sampled to generate multiple three-dimensional spatial probabilistic realizations of rock structure and groundwater distribution.

[0087] Key parameters are extracted from simulation results to construct a hierarchical risk indicator system. Fuzzy comprehensive evaluation is used to assign weights to indicators and determine membership functions. Geological parameters are mapped to risk levels, and a three-dimensional risk heat map is generated in combination with tunnel structure design parameters.

[0088] Preprocessing groundwater level and seepage rate time series data, using seasonal adjustment to decompose them into trend, seasonal, and irregular components. An ARIMA model is fitted to the trend component, and the seasonal component is periodically extrapolated. Combined with the irregular component, multi-scenario forecast results are generated, and thresholds are set to identify potential leakage or rock slip risks.

[0089] The spatial probability model is integrated with the time series prediction results to establish a four-dimensional dynamic risk model. The uncertainty is quantified through Monte Carlo simulation, and an interactive interface is developed to achieve multi-dimensional visualization of risk results and output reports.

[0090] Based on the geological environmental data, a geological risk analysis was conducted using a geostatistical simulation algorithm and a time series decomposition algorithm to obtain safety risk assessment results of the tunnel structure affected by geological factors and prediction results of the dynamic change trend of groundwater. During the geological risk analysis, rock type, degree of fracture development, groundwater level and seepage rate parameters were obtained. Sequential Gaussian simulation (SGS) was used to construct a spatial probability model of rock structure and groundwater distribution. The safety risk level of the tunnel structure affected by geological factors was assessed. The seasonal adjustment method X-13 was used to predict the dynamic change trend of groundwater and identify potential leakage or rock slip risks.

[0091] First, geological parameters were collected and preprocessed. A distributed sensor network was used to monitor physical indicators such as acoustic reflection characteristics and resistivity changes in the rock formations surrounding the tunnel in real time. This data, combined with drilling sampling data, was used to establish a three-dimensional spatial sampling point set. Quality control was performed on the collected raw data, removing outliers and filling missing values. Rock formation type data was categorized and coded using expert knowledge combined with cluster analysis. Image processing techniques were used to quantify fracture density, length, and direction. Groundwater level and seepage rate data were standardized to eliminate dimensionality effects.

[0092] During the spatial probability model construction phase, the Sequential Gaussian Simulation (SGS) algorithm is applied. First, the preprocessed geological parameters are transformed to a normal distribution to ensure that the variables conform to the Gaussian assumption. Spatial autocorrelation is then analyzed based on the variogram, and variogram values ​​in different directions are calculated. Theoretical models, such as spherical and exponential models, are fitted to determine parameters such as the range and sill value. Unsampled points are traversed along a random path. At each point to be estimated, the mean and variance are estimated through kriging interpolation based on the conditional distribution of the simulated points. Simulated values ​​are then randomly sampled from this conditional distribution. This process is repeated to generate multiple equally probable realizations, forming a three-dimensional probability field of rock structure and groundwater distribution. Each realization reflects a possible spatial configuration of the geological parameters.

[0093] Safety risk level assessment is based on a comprehensive multi-indicator analysis. Key parameters, such as the distribution of rock compressive strength and the groundwater pressure gradient, are extracted from simulation results. A hierarchical risk indicator system is constructed, including factors such as formation deformation rate, seepage erosion index, and structural stress concentration factor. A fuzzy comprehensive evaluation method is used to assign weights to each indicator and determine membership functions, mapping geological parameters to risk levels. For example, when the groundwater seepage rate exceeds a critical value and the rock fracture density is above a threshold, the area is identified as high-risk. Combined with tunnel structural design parameters, the impact of geological factors on structural stability is assessed, generating a three-dimensional risk heat map.

[0094] The seasonal adjustment method (X-13) is used to predict the dynamic trend of groundwater changes. Long-term groundwater level and seepage rate data are preprocessed to detect and correct outliers and mutation points. The time series is decomposed into trend components, seasonal components, and irregular components. The trend component is estimated using the moving average method to extract long-term trends. Fourier analysis is used to identify periodic fluctuations and determine the length of the seasonal cycle. The remaining component is used as an irregular component to reflect random disturbances. An ARIMA model is fitted to the decomposed trend component to predict future trends. The seasonal component is periodically extrapolated and combined with the probability distribution of the irregular component to generate multi-scenario forecast results. A risk threshold is set. When the predicted water level or seepage rate exceeds the warning value, an early warning mechanism is triggered to identify potential leakage paths or rock slip risk areas.

[0095] Finally, risk results are integrated and visualized: spatial probability models are integrated with time series forecast results to establish a four-dimensional dynamic risk model. Monte Carlo simulations are used to quantify risk assessment uncertainty and calculate confidence intervals for each risk level. An interactive visualization interface is developed to support multi-dimensional display of risk results, such as risk slices at different depths of rock formations and the dynamic evolution of groundwater seepage paths. A risk assessment report is generated, including the location of key risk points, development trend forecasts, and risk management priority rankings, providing data support for maintenance decision-making.

[0096] Example 3

[0097] Further illustrating with reference to Example 1, predicting the tunnel technical condition and generating maintenance decision parameters in step S3 includes: updating the conditional probability distribution of the tunnel technical condition prediction using a dynamic Bayesian network based on the geological risk assessment result;

[0098] Integrate rock stability, seepage data and disease history records, and use extreme gradient boosting tree to construct a maintenance decision tree;

[0099] The dynamic Bayesian network deduces the probability of technical status transition over many years and generates a cross-cycle maintenance plan.

[0100] In the preferred solution, the geological risk analysis results are aligned with the tunnel technical status data in time and space to construct a multi-source heterogeneous feature set containing static, dynamic and historical features. After discretization, coding and feature screening, the input data set is formed;

[0101] Construct a dynamic Bayesian network, define hidden state variables and observable variables, build a conditional probability table based on geological risk factors, use the expectation maximization algorithm to estimate parameters, calculate the hidden state posterior probability through the forward-backward algorithm, and establish a technical status transition probability matrix;

[0102] An extreme gradient boosting tree heterogeneous integration model was constructed. This model takes geological risk data, historical technical status data, and maintenance measures records as inputs. A multi-objective loss function with a regularization term was used to iteratively generate decision trees through gradient boosting. A spatial weight matrix was introduced to adjust sample weights to address spatial correlation.

[0103] A stacking ensemble strategy was used to fuse the dynamic Bayesian network and the extreme gradient boosting tree model. The weight parameters were optimized through cross-validation. The expected utility was calculated based on the prediction results, and the ε-greedy strategy was used to select the optimal maintenance measure.

[0104] Based on the dynamic Bayesian network, the multi-period technical status transition probability is recursively calculated. A multi-stage decision optimization model is constructed and resource constraints are introduced. The optimal maintenance strategy sequence across the period is solved using strategy iteration and Lagrangian relaxation method.

[0105] The prediction uncertainty is quantified through Monte Carlo simulation, and the characteristic contribution is analyzed using Shapley value. The maintenance decision parameters and scenario analysis results containing probability distribution are output.

[0106] The specific steps of S3 are: based on the geological risk analysis results, a heterogeneous integrated model of a dynamic Bayesian network and an extreme gradient boosting tree is used to integrate the tunnel technical status data, predict the tunnel technical status, and generate maintenance decision parameters; in step S3, the tunnel technical status is predicted and maintenance decision parameters are generated, including: based on the geological risk assessment results, a dynamic Bayesian network is used to update the conditional probability distribution of the tunnel technical status prediction; the extreme gradient boosting tree is used to integrate rock formation stability, seepage data and disease history records; the dynamic Bayesian network deduces the probability of technical status transfer over many years to generate a cross-cycle maintenance plan.

[0107] First, data fusion and feature engineering are performed: the geological risk analysis results are spatially and temporally aligned with the tunnel technical condition data. A multi-source, heterogeneous feature set is constructed, including static, dynamic, and historical features. Continuous features are binned and discretized, and categorical features are one-hot encoded to generate high-dimensional feature vectors. The information gain ratio between features is calculated, and key features with high contribution to technical condition prediction are selected to form the final input dataset.

[0108] Dynamic Bayesian network modeling steps: define the network structure and divide the tunnel technical conditions into hidden state variables (lining status, seepage status) and observable variables (crack width, water seepage), the time slice interval is set as the maintenance period. Based on the geological risk assessment results, a conditional probability table is constructed. ,in G Represents the geological risk factor vector. The expectation maximization (EM) algorithm is used to estimate the model parameters, and the posterior probability of the hidden state is calculated by the forward-backward algorithm. . Establish the technology status transition probability matrix ,in , reflecting the probability that the technical status transfers from state i to state j under the influence of geological risk.

[0109] Extreme Gradient Boosting (XGBoost) heterogeneous ensemble modeling: Taking geological risk data, technical status historical data, and maintenance measures records as input, a multi-objective loss function is constructed. , where l is the logarithmic loss function, is a regularization term that controls the complexity of the tree. Using the gradient boosting framework, K decision trees are iteratively generated. , and the final prediction is At the splitting node of each tree, a greedy algorithm is used to find the optimal feature splitting point and calculate the gain , where G and H are gradient and Hessian matrices respectively. In order to deal with the spatial correlation of geological data, a spatial weight matrix W is introduced to adjust the sample weights ,in is the sample space distance, is the attenuation coefficient.

[0110] Model fusion and decision parameter generation: Using a stacking integration strategy, the technical status transition probability output by DBN is used as one of the input features of XGBoost. A meta-learner is constructed and weight parameters are optimized through cross-validation. The final prediction result is Defining the Conservation Decision Space , calculate expected utility based on the prediction results ,in Status s ′ is the utility function. The optimal maintenance measure is selected using the ε-greedy strategy , and at the same time introduce the exploration factor ε to balance development and exploration.

[0111] Generation of cross-cycle maintenance plans: Based on the Markov property of dynamic Bayesian networks, recursive calculation of multi-cycle technical condition transition probabilities . Construct a multi-stage decision optimization model ,in is the discount factor. The strategy iteration algorithm is used to solve the optimal maintenance strategy sequence , generate cross-cycle maintenance plans. Introduce resource constraints , where C is the maintenance cost function, For the budget of year t, the Lagrangian relaxation method is used to solve the constrained optimization problem.

[0112] Uncertainty quantification and sensitivity analysis: Generate random samples of geological risk parameters through Monte Carlo simulation and calculate the confidence interval of technical condition prediction. Shapley value is used to analyze the contribution of each input feature to maintenance decision. The formula is: , where f is the prediction model and N is the feature set. Scenario analysis is designed for highly sensitive features (such as groundwater seepage rate) to evaluate the robustness of maintenance plans under different geological conditions. The output includes maintenance decision parameters with probability distribution, such as the recommended maintenance time window. and its probability of occurrence , providing a quantitative basis for maintenance decisions.

[0113] Example 4

[0114] Further described in conjunction with Example 1, step S4 optimizes the maintenance decision parameters, including:

[0115] Determining the maintenance decision parameters including maintenance priorities and measure combinations;

[0116] A cuckoo search algorithm is used to globally optimize the maintenance priorities and measure combinations to minimize the maintenance costs under the influence of geological risks;

[0117] The maintenance decision parameters are modeled as a maintenance priority vector ( ) and the measure combination vector ( ) , construct a function with the weighted sum of maintenance cost and geological risk as the target , and set resource constraints , priority normalization constraint and mutually exclusive constraints on measures;

[0118] Initialize the cuckoo search algorithm population and update the solution vector through the Levy flight mechanism, where the Levy step size follows a stable distribution ( , ), the update formula is , perform S-shaped transformation on maintenance priority , the measures are updated using roulette wheel selection;

[0119] By probability Execute the nest elimination mechanism to generate new solutions through random perturbations;

[0120] Perform constraint repair on the updated solution and calculate the cost-benefit ratio when the budget is violated. Adjust the priority in descending order, passing when normalization is violated repair;

[0121] Calculate population diversity during the iteration process When the probability of discovery is lower than the threshold, the optimal solution is output after the algorithm terminates. , extract maintenance priority and combination of measures , calculate the total maintenance cost , risk-weighted cost and cost-effectiveness ratio and through sensitivity analysis and robustness indicators Evaluate the stability of the solution, For the The population diversity index at the iteration, is the population size, is the total number of simulated geological risk parameter fluctuation scenarios, For the Optimal measures for each segment maintenance costs.

[0122] S4. Optimizing the maintenance decision parameters using a cuckoo search algorithm to minimize the maintenance cost under the influence of geological risks; optimizing the maintenance decision parameters comprises: determining that the maintenance decision parameters include maintenance priorities and a combination of measures; and globally optimizing the maintenance priorities and the combination of measures using a cuckoo search algorithm to minimize the maintenance cost under the influence of geological risks.

[0123] The maintenance decision parameters are defined as multidimensional vectors ,in represents the maintenance priority of the i-th tunnel section, For the corresponding maintenance measures ( is the set of measures). Define the objective function ,in is the measure cost function, is the geological risk impact function ( Obtained by the risk level mapping output from step S2), Is the risk weight coefficient. Set constraints: resource constraints (B is the total budget), priority normalization , measures mutually exclusive constraint ( is the indicator function).

[0124] Cuckoo search algorithm initialization: randomly generate the initial population , each solution vector Represents a maintenance plan. Perform feasibility repairs on each solution. If the budget constraint is violated, adjust it in descending order of priority. Until the constraints are met. Calculate the initial fitness value , record the global optimal solution . Set algorithm parameters: Discovery probability , maximum number of iterations T, step size scaling factor .

[0125] Levi flight update mechanism: for each solution , generate step size according to Levy distribution ,in ( is a stable parameter, ). Update the solution vector , perform S-type transformation on the maintenance priority dimension To ensure the value range, the roulette wheel selection method is used to update the measure dimension . Calculate the fitness of the new solution, if Accept the update.

[0126] Nest elimination mechanism: for each solution , with probability Perform random perturbations. Generate random numbers ,like , then randomly select two dimensions for mutation, and : , is the variation intensity, Recalculate the fitness and update the global optimal solution.

[0127] Constraint processing and repair: For each updated solution, check whether the constraints are met. If the budget constraint is violated, adopt a greedy repair strategy: calculate the cost-benefit ratio of each segment ( is the risk reduction amount), sort them in descending order and reduce the priority until the budget is met. If the normalization constraint is violated, Perform normalization.

[0128] Algorithm iteration and termination: Repeat the Levy flight update and nest elimination operations, and calculate the population diversity after each iteration t times , if the diversity is below the threshold, increase To avoid premature convergence. When the maximum number of iterations T is reached or there is no significant improvement in the optimal solution for consecutive k generations, the algorithm is terminated and the global optimal solution is output. .

[0129] Result analysis and solution generation: from the optimal solution Extract the maintenance priority vector and measure combination vector , sort the tunnel sections in descending order of priority. For each section i, generate a maintenance work order containing the measure type , implementation time window (based on the technical status forecast of the S3 step), resource requirements and expected risk reduction Calculate total maintenance costs and risk-weighted costs , evaluate the cost-effectiveness of the program .

[0130] Sensitivity analysis and robustness assessment: Risk weight coefficient Perform sensitivity analysis to generate different The optimal solution set under the value of . The calculation solution is under the fluctuation of geological risk parameters. Robustness indicators within the range ,S is the number of scenarios, and is the parameter of the sth scenario, which ensures the stability of the scheme under uncertainty. The final output is a comprehensive maintenance plan that includes the optimal maintenance priority, measure combination, time window, and risk response strategy.

[0131] In actual deployment,

[0132] Rock formation parameters and groundwater data are collected through IoT sensors, and Python scripts (Pandas library) are used to clean and normalize the data and store it in a PostgreSQL database.

[0133] The sequential Gaussian simulation (SGS) was implemented based on the GSLIB library to construct a spatial probability model. The seasonal adjustment method (X-13) was implemented using the forecast package of R language to perform time series decomposition, and the risk assessment results were output to a MySQL database.

[0134] Use the PyMC3 library to build a dynamic Bayesian network (DBN), build a heterogeneous integration model through the XGBoost Python interface, integrate TensorFlow for model training and prediction, and store the results in MongoDB.

[0135] The Cuckoo Search (CS) algorithm is implemented based on the PySwarms library. The optimization logic is written in Jupyter Notebook, and the CPLEX solver is called to process the constraints. The generated optimal parameters are then written to the Redis cache.

[0136] Use Three.js to develop a three-dimensional geographic information interface, use ECharts to display data charts, use Python's ReportLab library to generate PDF reports, integrate the Flask framework to build Web services, and realize multi-terminal access and interaction.

[0137] We use Docker containerization technology to deploy various modules, implement cluster management through Kubernetes, use Nginx for load balancing, and build a production environment on cloud servers such as Alibaba Cloud and Tencent Cloud to ensure high availability and scalability.

[0138] Example 5

[0139] Further illustrating with reference to Example 1, in step S5, the total maintenance cost allocation, the expected improvement in technical conditions, the predicted value of the rock formation stability safety factor, and the groundwater seepage risk warning level are output;

[0140] Among them, construct a four-dimensional decision result matrix , generate and visualize maintenance decision parameters through the following steps: Based on the optimized maintenance priority and cost estimates of measures Calculate section maintenance costs and total cost , generate the total maintenance cost allocation matrix by geological risk level group ;

[0141] Using the dynamic Bayesian network state transition probability matrix Computing Technology Status Improvement Index and global improvement index , forming a matrix of expected improvements in technical conditions ;

[0142] Extract statistical parameters of rock compressive strength from sequential Gaussian simulation results and calculate safety factors based on tunnel loads , generate the predicted mean and confidence interval through Monte Carlo simulation, and construct the rock stability safety factor prediction value matrix ;

[0143] The predicted values ​​of groundwater level and seepage rate are used as inputs of the fuzzy logic system, and the risk level sequence is output through Gaussian membership function and rule base reasoning. , forming a groundwater seepage risk warning level matrix ;

[0144] Develop a 3D GIS interface to visualize the results of each dimension in the form of ring charts, heat maps, cloud maps, and dynamic arrows, and support interactive queries in the time dimension;

[0145] Calculating cost forecast error rate and risk level prediction accuracy , generate a PDF report containing uncertainty indicators and establish a feedback optimization mechanism.

[0146] : Cost forecast error rate, used to measure the degree of deviation between the predicted value of total maintenance cost and the actual value. :The total maintenance cost forecast value output by the system is Calculated. : The actual total cost after the implementation of maintenance measures, Accuracy of risk level prediction.

[0147] The specific implementation of S5 includes: outputting decision results including the allocation of total maintenance costs based on geological risks, expectations for improvement in technical conditions, predicted values ​​of rock stability safety factors, and the warning level of groundwater seepage risk; the output includes the allocation of total maintenance costs based on geological risks, expectations for improvement in technical conditions, predicted values ​​of rock stability safety factors, and the warning level of groundwater seepage risk.

[0148] First, build the decision result data framework: establish a four-dimensional output matrix , corresponding to maintenance cost allocation, technical condition improvement, rock formation stability and seepage risk warning respectively. The maintenance cost prediction value of each tunnel section is extracted from the XGBoost model output of step S3 , combined with the maintenance priority after optimization in step S4 , calculate section maintenance costs , total cost By geological risk level Group the segments and calculate the cost ratio of each group , generate a cost allocation matrix based on risk level .

[0149] The expected improvement of technical status is calculated by the state transition probability matrix T of the dynamic Bayesian network: For each segment i, let the current technical status level be , after maintenance measures After that, the state probability distribution of the next cycle is . Define technical condition improvement indicators (where s' is the status level, the larger the value, the worse the condition), if Indicates expected improvement. Calculate the global improvement index , generate the technology status improvement matrix .

[0150] The prediction of the safety factor of rock formation stability is based on the three-dimensional probability field of sequential Gaussian simulation: for each segment i, the mean compressive strength of the rock formation in the simulation is extracted and standard deviation , combined with the tunnel structure design load , calculate the safety factor ,k is the safety factor, usually 1.5~2.0. Generated by Monte Carlo simulation The probability distribution of the output prediction value is and 95% confidence interval , forming a rock stability matrix .

[0151] The groundwater seepage risk warning level is realized by fuzzy logic system: the groundwater level prediction value and the predicted seepage rate As input variables, the membership function adopts Gaussian Build a rule base such as "If the water level is high and the seepage rate is high, the risk level is extremely high." Output the risk level through fuzzy reasoning synthesis , corresponding to the warning color. For each segment i, output the risk level prediction sequence for the next 1 to 3 years , forming a seepage risk matrix .

[0152] Decision results visualization and interaction: Develop a three-dimensional geographic information system (3D-GIS) interface to Displayed as a risk level-cost ratio circular chart, Superimpose the tunnel model with the section-by-section thermal map, The stability distribution of rock formations is displayed using a safety factor cloud map. Dynamic flow arrows mark seepage risk paths. A time slider allows users to view projected technical improvements and seepage risk evolution over different years. Users can click on any section to view a detailed report, including cost breakdowns, safety factor probability density curves, and historical risk warning data.

[0153] Uncertainty quantification and reporting: Calculate uncertainty indicators for each output, such as the standard error of cost allocation , confidence level of improvement in technical status , is the standard normal distribution function. Generate a PDF decision report containing: ① Cover; ② Table of Contents; ③ Abstract; ④ Main Text; ⑤ Appendix. The report includes links to interactive charts and supports online dynamic updates.

[0154] Finally, conduct result verification and feedback: compare and verify the output results with historical maintenance data, and calculate the cost prediction error rate , Risk level prediction accuracy . A feedback mechanism is established to allow users to submit result evaluations through the interface. The system automatically collects feedback data and updates model parameters, forming a closed loop of "prediction-implementation-feedback-optimization".

[0155] : Cost forecast error rate, used to measure the degree of deviation between the predicted value of total maintenance cost and the actual value. :The total maintenance cost forecast value output by the system is Calculated. : The actual total cost after the implementation of maintenance measures, Accuracy of risk level prediction.

[0156] In the actual deployment process:

[0157] The Python Pandas library is used to extract the maintenance cost forecast and maintenance priority data generated in the S3 and S4 steps from the database (PostgreSQL) for data cleaning and calculation. The four-dimensional output matrix is ​​constructed using the NumPy library. , and calculate the data of each sub-matrix. For example, calculate the maintenance cost of the section through matrix operation Total cost and the proportion of expenses ,Finish Matrix construction; matrix for technical status improvement , using the relevant functions of the SciPy library according to the state transition probability matrix of the dynamic Bayesian network Computing Technology Status Improvement Index and global improvement index .

[0158] Based on Python's GSLIB library or similar geological simulation tools, extract the mean and standard deviation data of rock compressive strength generated by sequential Gaussian simulation. Use NumPy and SciPy libraries to calculate safety factors. Calculation and Monte Carlo simulation to generate the rock stability matrix For groundwater seepage risk warning, we use Python's Scikit-fuzzy library to build a fuzzy logic system, define fuzzy sets and membership functions, implement the fuzzy reasoning process, output the risk level prediction sequence, and form a seepage risk matrix. .

[0159] Use JavaScript's Three.js library to develop a three-dimensional geographic information system (3D-GIS) interface, and combine it with the D3.js library to achieve data visualization effects such as ring maps, heat maps, and cloud maps. A circular chart displays the cost allocation ratio, the ECharts library creates a heat map to illustrate the expected improvement in technical conditions, and Three.js creates a 3D scene to display a rock stability cloud map and dynamic arrows for seepage risk. HTML5 and CSS3 are used for interface layout and styling, implementing interactive features such as a time slider. AJAX technology is also used to enable users to click on a segment to access a detailed report. The charts in these detailed reports are generated using libraries such as Highcharts.

[0160] In Python, use the statistical functions of the SciPy library to calculate the standard error of the cost distribution , the confidence level of technical improvement is calculated using the normal distribution function of the Scipy.stats library Use Python's ReportLab library or the Jinja2 template engine in conjunction with LaTeX to generate a PDF decision report. This report embeds the calculated uncertainty indicators, analysis of each dimension's results, and relevant charts. To dynamically update the charts in the report, embed JavaScript-based interactive chart code in the report and provide access to it through a web server (Nginx).

[0161] Compare the output results with historical maintenance data stored in a database (such as MySQL) and use Python to write a script to calculate the cost forecast error rate and risk level prediction accuracy In the web interface, JavaScript and AJAX technologies are used to implement user feedback submission. The backend uses Python web frameworks such as Flask or Django to receive feedback data and store it in a database. Simultaneously, data processing scripts are written to automatically update model parameters based on feedback data, completing a closed-loop "prediction-implementation-feedback-optimization" process. Version control tools (Git) can be used to manage the model parameter update process. The entire system can be deployed on a cloud server, using Docker for containerized deployment and Kubernetes for cluster management and service orchestration, ensuring high availability and scalability.

[0162] Example 6

[0163] Further explanation is provided in conjunction with Example 1, which also includes maintenance measures management, and the maintenance measures management includes:

[0164] Manage maintenance measures based on disease name, maintenance type, maintenance measures and unit price;

[0165] By combining the non-parametric survival analysis algorithm with the rock formation life cycle data, the effectiveness attenuation curve of different maintenance measures in delaying the development of geological diseases is quantified;

[0166] Computational geometry algorithms are used to automatically optimize the repair length threshold according to the geological boundaries of the construction area.

[0167] In maintenance measures management, we first build a basic database for maintenance measures, enabling systematic management of disease names, maintenance types, maintenance measures, and unit prices. We create a structured data table, using disease name as the index key, to associate specific maintenance measures and their unit prices for different maintenance types. Through the database management system's add, delete, modify, and query interfaces, maintenance personnel can update and retrieve maintenance measures data in real time, ensuring the accuracy and timeliness of the information.

[0168] A non-parametric survival analysis algorithm is used in combination with rock formation life cycle data to quantify the benefit attenuation curve of different maintenance measures in delaying the development of geological diseases. Historical maintenance records are integrated with geological monitoring data to construct a data set containing fields such as sample individual ID, maintenance measure type, implementation time, disease occurrence time, rock formation characteristic parameters, etc. In the model construction stage, the risk function is defined. ,in is the individual's hazard rate at time t, is the benchmark risk function, is a covariate that affects disease development, is the regression coefficient of the covariate. The coefficient is solved by partial likelihood estimation method (PartialLikelihood Estimation) , we get the quantitative relationship between different maintenance measures and geological factors on the risk of disease development. Based on the model results, we simulate the survival function of disease occurrence under different maintenance measures. , draw a benefit attenuation curve to visually show the changing effects of maintenance measures on delaying disease development over time.

[0169] Computational geometry algorithms are used to automatically optimize the maintenance length threshold based on the geological boundaries of the construction area. First, the geological boundary data of the construction area is extracted using the spatial probability model of the rock structure generated by sequential Gaussian simulation, and it is abstracted into a set of polygons in two-dimensional or three-dimensional space. Using the Voronoi diagram partitioning algorithm, key monitoring points in the construction area (such as geological drilling locations and high-incidence points) are used as seed points to generate a Voronoi diagram. By analyzing the geometric characteristics of the Voronoi polygons, sub-areas with relatively uniform geological conditions are identified. For each sub-area, combined with the minimum construction length limit and tolerance requirements , the convex hull optimization algorithm is used to calculate the optimal repair length threshold. Specifically, the sub-region boundary point set is used as input, and its convex hull is calculated by the Graham scanning algorithm. Then, the convex hull boundary is smoothed and the length is corrected according to factors such as the construction machinery operation capacity and material loss. If the sub-region length L is less than , then merge it with the adjacent sub-region; if the length exceeds the tolerance range, adjust the boundary point position iteratively to make the repair length meet , and finally determine the maintenance length threshold of each area, providing accurate geometric parameter basis for the formulation of maintenance construction plan.

[0170] Example 7

[0171] Further described in conjunction with Example 1, the acquisition of tunnel static and dynamic data includes:

[0172] Data collection is performed through the APP, which includes inputting the pile number, selecting the inspection item, selecting the disease, recording the disease attributes, and recording the disease assessment information;

[0173] The data collected by the data is encrypted for transmission using the elliptic curve encryption algorithm ECC;

[0174] Implementing fuzzy query of the disease records through suffix automaton algorithm;

[0175] Achieve incremental real-time synchronization between the APP and the Web through a two-way differential synchronization algorithm;

[0176] The spatial distribution map of hazards integrating geological risk distribution is drawn using the spline interpolation algorithm.

[0177] In step S1, the specific process of obtaining the static and dynamic data of the tunnel and the geological environment data around the tunnel is as follows:

[0178] In the data collection phase, the APP serves as the main tool for on-site data entry, and adopts a simple and efficient interactive interface design. When maintenance personnel enter the data collection interface, they must first enter the pile number. The pile number information is used to accurately locate the tunnel inspection position, and its format follows industry standards. Then, select the inspection item from the preset inspection item list, which covers core items such as tunnel lining inspection, road condition inspection, and electromechanical facility inspection. When selecting the type of disease, the system provides a disease classification tree based on historical data and industry standards. For example, lining diseases are subdivided into cracks, water leakage, peeling, etc. Maintenance personnel can quickly locate specific diseases by clicking on the level. When recording disease attributes, for crack diseases, the crack length l, width w, and strike angle need to be entered. For water leakage, the location and flow rate q are recorded. The damage assessment information is based on the scoring criteria set out in the Technical Specifications for Highway Tunnel Maintenance. Maintenance personnel select the corresponding score level on the app based on the severity of the damage.

[0179] During data transmission, the elliptic curve cryptography (ECC) algorithm is used to ensure data security. First, the appropriate elliptic curve parameters are selected and the prime number domain is used. The elliptic curve equation is defined on ,in and satisfy When generating a public-private key pair, a base point G on the curve is selected, and the private key d is in the interval A random integer in, n is the order of the base point G, the public key During the encryption process, for the data m to be transmitted, first map it to a point M on the elliptic curve, then select a random number k and calculate the ciphertext At the receiving end, the private key d is used for decryption, and the calculation Restore the original data to effectively prevent data from being stolen or tampered with during transmission.

[0180] In terms of data query, a suffix automaton algorithm is used to achieve efficient fuzzy query of disease records. When constructing the suffix automaton, all disease record texts are concatenated into a long string S, and the suffix automaton is gradually constructed using an online algorithm (Ukkonen algorithm). The Ukkonen algorithm is based on three invariants. Each time a character is added to the string, the state transition and suffix link adjustment are adjusted to achieve near linear time complexity. After building the suffix automaton, when a user enters a keyword for a fuzzy query, the algorithm begins at the initial state of the suffix automaton and traverses along the transition edges that match the keyword characters. If a terminal state is reached, it indicates that a defective record containing the keyword exists. All matching records can be retrieved by backtracking the path. Compared to traditional string matching algorithms, suffix automata have significant efficiency advantages when processing large-scale text and fuzzy queries.

[0181] In order to achieve incremental real-time synchronization between the APP and Web terminals, a two-way differential synchronization algorithm is adopted. The data version number is maintained on the APP and Web terminals respectively. and , and data modification logs and When the data on the APP side changes, record the change operation to and will Add 1. When syncing, the APP will Sent to the Web side, the Web side compares its own version number ,like , then request the APP to send middle to The operation logs between versions, the Web side performs corresponding operations based on the logs to update local data, and Updated to Similarly, when data on the web changes, the updates are synchronized to the app according to this process. This approach avoids full data transmission, greatly reduces network traffic, and enables fast, real-time synchronization.

[0182] When drawing the spatial distribution map of the disease that integrates the geological risk distribution, the spline interpolation algorithm is used. First, the tunnel is divided into multiple nodes along the pile number direction. Each node corresponds to a detection location. The node records the disease information and geological risk level, which are obtained through the geological risk analysis in step S2. For the disease data, the cubic spline interpolation function is used. Interpolation is performed in the interval superior, ,in , n is the number of nodes. By solving the linear equations including boundary conditions, the coefficients , making the interpolation curve continuous at the nodes and its first- and second-order derivatives continuous, thus obtaining a smooth disease distribution curve. Interpolation is also performed for geological risk levels, converting discrete risk level data into a continuous risk field. Finally, the disease distribution curve is overlaid and visualized with the geological risk field. Different colors and patterns are used on the distribution map to distinguish disease types and risk levels. For example, red indicates high-risk areas with severe diseases, providing an intuitive spatial distribution reference for maintenance decisions.

[0183] When obtaining geological environment data around the tunnel, a variety of professional monitoring equipment is used. The geological radar is used to detect the rock structure. The geological radar emits high-frequency electromagnetic waves f and reflects the time t, amplitude A and phase of the reflected wave. And other information, through the formula The depth of the rock interface is calculated, where v is the propagation velocity of electromagnetic waves in the medium. The rock formation type and fracture development are determined by combining the reflected wave characteristics. Groundwater monitoring wells are deployed, and pressure sensors measure the water level h in real time. Electromagnetic flowmeters measure the groundwater seepage rate v. The sensors convert physical quantities into electrical signals, which are then transmitted to the data acquisition terminal after A / D conversion. All geological environmental data is collected at a set sampling frequency and transmitted to a data center via wireless networks (4G and 5G). It is integrated with static and dynamic tunnel data to provide a comprehensive data foundation for subsequent geological risk analysis and maintenance decisions.

[0184] Example 8

[0185] Further explained in conjunction with Example 1, the system should include: a cockpit, maintenance decision-making, road property management, system management, and structure inspection modules;

[0186] Road property management module, used to maintain basic tunnel data;

[0187] Structural inspection module, used to obtain tunnel site data;

[0188] A maintenance decision module, configured to generate tunnel maintenance decision parameters based on the tunnel basic data and the tunnel site data;

[0189] A cockpit module, for visually displaying the tunnel basic data, the tunnel site data, and the tunnel maintenance decision parameters;

[0190] System management module, used for system management;

[0191] Among them, the road property management module, the structure inspection module, the maintenance decision module, the cockpit module and the system management module operate in coordination to form a data-driven tunnel maintenance management closed loop.

[0192] The road property management module maintains basic tunnel data. The structural inspection module's app collects tunnel field data, transmits the data using the elliptic curve encryption algorithm, implements fuzzy queries on disease records with the help of the suffix automaton algorithm, synchronizes with the web client in real time using the bidirectional differential synchronization algorithm, and draws a spatial distribution map of the disease using the spline interpolation algorithm. After acquiring basic and field data, the maintenance decision-making module uses a geostatistical simulation algorithm, a time series decomposition algorithm, a dynamic Bayesian network, an extreme gradient boosting tree model, and a cuckoo search algorithm to perform geological risk analysis, technical condition prediction, and maintenance decision parameter optimization, generating decision results that include cost allocation and risk warning. The cockpit module visualizes basic data, field data, and decision parameters using a 3D GIS interface, pie charts, heat maps, and other formats, supporting multi-dimensional screening and interactive queries. The system management module is responsible for unit organization, user permissions, and role management to ensure the operational specifications of each module. The modules operate collaboratively to form a data-driven closed loop of "data maintenance - field data collection - decision generation - visualization - system management."

[0193] The road property management module implements standardized maintenance of basic data, providing a reliable data source for decision-making; the structural inspection module improves the efficiency and security of on-site data acquisition through intelligent collection and synchronization technology, and combines interpolation algorithms to intuitively present the spatial distribution of defects; the maintenance decision-making module integrates multiple algorithms to deeply analyze geological risks and technical conditions, generate scientific and quantified maintenance parameters and cross-cycle plans, and balance costs and risks; the cockpit module uses visualization technology to lower the threshold for data comprehension, assisting managers to quickly locate high-risk areas and optimize resource allocation; the system management module ensures data security and operational compliance through permission control; the closed-loop collaboration of all modules realizes the automated flow from data collection to decision optimization, promoting tunnel maintenance from experience-oriented to data-driven, improving management efficiency, reducing safety hazards and extending the service life of tunnels.

[0194] The above embodiments are merely preferred technical solutions of the present invention and should not be construed as limiting the present invention. The scope of protection of the present invention shall be the technical solutions set forth in the claims, including equivalent alternatives to the technical features of the technical solutions set forth in the claims. In other words, equivalent alternatives and improvements within this scope are also within the scope of protection of the present invention.

Claims

1. A highway tunnel decision-making and maintenance method, characterized by: The method includes: S1. Obtain static and dynamic data of the tunnel and geological environment data around the tunnel; S2. Based on the geological environment data, a geological statistical simulation algorithm and a time series decomposition algorithm are used to perform a geological risk analysis to obtain a safety risk assessment result of the tunnel structure affected by geological factors and a prediction result of the groundwater dynamic change trend; In step S2, geological risk analysis is performed to obtain rock layer type, fracture development degree, groundwater level and seepage rate parameters; Use Sequential Gaussian Simulation (SGS) to construct a spatial probability model of rock structure and groundwater distribution; Assess the safety risk level of the tunnel structure affected by geological factors; Use the seasonal adjustment method X-13 to predict groundwater dynamic trends and identify potential leakage or rock slip risks; The geological risk analysis method is: Through a distributed sensor network and drilling sampling, the rock type, fracture development, groundwater level, and seepage rate parameters around the tunnel are obtained and pre-processed to remove outliers, fill in missing values, and perform standardization. Based on the pre-processed geological parameters, the sequential Gaussian simulation algorithm is used to transform the parameters into a normal distribution. The spatial autocorrelation is analyzed through the variogram and the theoretical model is fitted. The points to be estimated are traversed along a random path. The mean and variance are estimated through Kriging interpolation based on the conditional distribution of the simulated points. The simulated values ​​are randomly sampled to generate multiple three-dimensional spatial probabilistic realizations of rock structure and groundwater distribution. Key parameters are extracted from simulation results to construct a hierarchical risk indicator system. Fuzzy comprehensive evaluation is used to assign weights to indicators and determine membership functions. Geological parameters are mapped to risk levels, and a three-dimensional risk heat map is generated in combination with tunnel structure design parameters. Preprocessing groundwater level and seepage rate time series data, using seasonal adjustment to decompose them into trend, seasonal, and irregular components. An ARIMA model is fitted to the trend component, and the seasonal component is periodically extrapolated. Combined with the irregular component, multi-scenario forecast results are generated, and thresholds are set to identify potential leakage or rock slip risks. The spatial probability model is integrated with the time series forecast results to establish a four-dimensional dynamic risk model. Uncertainty is quantified through Monte Carlo simulation. An interactive interface is developed to achieve multi-dimensional visualization of risk results and output reports. S3. Based on the geological risk analysis results, a heterogeneous integrated model of dynamic Bayesian networks and extreme gradient boosting trees is used to integrate tunnel technical condition data, predict the tunnel technical condition, and generate maintenance decision parameters; S4. Optimizing the maintenance decision parameters using a cuckoo search algorithm to minimize the maintenance cost under the influence of geological risks; S5. The output includes the decision results of the total maintenance cost allocation based on geological risks, the expected improvement of technical conditions, the predicted value of the rock stability safety factor, and the groundwater seepage risk warning level.

2. The highway tunnel decision-making and maintenance method according to claim 1, characterized in that: Step S3 predicts the tunnel technical condition and generates maintenance decision parameters, including: updating the conditional probability distribution of the tunnel technical condition prediction using a dynamic Bayesian network based on the geological risk analysis results; Integrate rock stability, seepage data and disease history records, and use extreme gradient boosting tree to construct a maintenance decision tree; The dynamic Bayesian network deduces the probability of technical status transition over many years and generates a cross-cycle maintenance plan.

3. The highway tunnel decision-making and maintenance method according to claim 2, characterized in that: The geological risk analysis results are aligned with the tunnel technical status data in time and space to construct a multi-source heterogeneous feature set containing static, dynamic, and historical features. The input data set is formed through discretization, coding, and feature screening. Construct a dynamic Bayesian network, define hidden state variables and observable variables, build a conditional probability table based on geological risk factors, use the expectation maximization algorithm to estimate parameters, calculate the hidden state posterior probability through the forward-backward algorithm, and establish a technical status transition probability matrix; An extreme gradient boosting tree heterogeneous integration model was constructed. This model takes geological risk data, historical technical status data, and maintenance measures records as inputs. A multi-objective loss function with a regularization term was used to iteratively generate decision trees through gradient boosting. A spatial weight matrix was introduced to adjust sample weights to address spatial correlation. A stacking ensemble strategy was used to fuse the dynamic Bayesian network and the extreme gradient boosting tree model. The weight parameters were optimized through cross-validation. The expected utility was calculated based on the prediction results, and the ε-greedy strategy was used to select the optimal maintenance measure. Based on the dynamic Bayesian network, the multi-period technical status transition probability is recursively calculated. A multi-stage decision optimization model is constructed and resource constraints are introduced. The optimal maintenance strategy sequence across the period is solved using strategy iteration and Lagrangian relaxation method. The prediction uncertainty is quantified through Monte Carlo simulation, and the characteristic contribution is analyzed using Shapley value. The maintenance decision parameters and scenario analysis results containing probability distribution are output.

4. The highway tunnel decision-making and maintenance method according to claim 1, characterized in that: Step S4 optimizes the maintenance decision parameters, including: Determining the maintenance decision parameters including maintenance priorities and measure combinations; A cuckoo search algorithm is used to globally optimize the maintenance priorities and measure combinations to minimize the maintenance costs under the influence of geological risks; The maintenance decision parameters are modeled as a maintenance priority vector and measure combination vector Multidimensional vector of , construct a function with the weighted sum of maintenance cost and geological risk as the target , and set resource constraints , priority normalization constraint and the mutually exclusive constraint of measures, where B is the total budget; in, is the measure cost function, is the geological risk impact function, Obtained from the risk level mapping output by step S2, is the risk weight coefficient; Initialize the cuckoo search algorithm population and update the solution vector through the Levy flight mechanism, where the Levy step size follows a stable distribution , the update formula is , perform S-shaped transformation on maintenance priority , the measures are updated using roulette wheel selection, is the step size scaling factor, where is a stable parameter; where the solution vector Indicates a maintenance plan. is the global optimal solution, By probability Execute the nest elimination mechanism to generate new solutions through random perturbations; Perform constraint repair on the updated solution and calculate the cost-benefit ratio when the budget is violated. Adjust the priority in descending order, passing when normalization is violated Repair, among others, is the cost function of the measure, which means that the maintenance measures are taken for the i-th section The cost of for the amount of risk reduction; Calculate population diversity during the iteration process When the probability of discovery is lower than the threshold, the optimal solution is output after the algorithm terminates. , extract maintenance priority and combination of measures , calculate the total maintenance cost , risk-weighted cost and cost-effectiveness ratio and through sensitivity analysis and robustness indicators Evaluate the stability of the solution, For the The population diversity index at the iteration, is the population size, is the total number of simulated geological risk parameter fluctuation scenarios, For the Optimal measures for each segment maintenance costs.

5. The highway tunnel decision-making and maintenance method according to claim 1, characterized in that: In step S5, the total maintenance cost distribution, the expected improvement of the technical condition, the predicted value of the rock formation stability safety factor, and the groundwater seepage risk warning level are output; Among them, construct a four-dimensional decision result matrix , generate and visualize maintenance decision parameters through the following steps: Based on the optimized maintenance priority and cost estimates of measures Calculate section maintenance costs and total cost , generate the total maintenance cost allocation matrix by geological risk level group ; Using the dynamic Bayesian network state transition probability matrix Computing Technology Status Improvement Index and global improvement index , forming a matrix of expected improvements in technical conditions ; in, is the status level; Extract statistical parameters of rock compressive strength from sequential Gaussian simulation results and calculate safety factors based on tunnel loads , k is the safety factor, and the predicted mean and confidence interval are generated through Monte Carlo simulation to construct the rock formation stability safety factor prediction value matrix ; The predicted values ​​of groundwater level and seepage rate are used as inputs of the fuzzy logic system, and the risk level sequence is output through Gaussian membership function and rule base reasoning. , forming a groundwater seepage risk warning level matrix ; Develop a 3D GIS interface to visualize the results of each dimension in the form of ring charts, heat maps, cloud maps, and dynamic arrows, and support interactive queries in the time dimension; Calculating cost forecast error rate and risk level prediction accuracy , generate a PDF report including uncertainty indicators and establish a feedback optimization mechanism; is the cost forecast error rate, is the total maintenance cost forecast value output by the system, is the actual total cost after the implementation of the maintenance measures, Accuracy of risk level prediction.

6. The highway tunnel decision-making and maintenance method according to claim 1, characterized in that: It also includes maintenance measures management, which includes: Manage maintenance measures based on disease name, maintenance type, maintenance measures and unit price; By combining the non-parametric survival analysis algorithm with the rock formation life cycle data, the effectiveness attenuation curve of different maintenance measures in delaying the development of geological diseases is quantified; Computational geometry algorithms are used to automatically optimize the repair length threshold according to the geological boundaries of the construction area.

7. The highway tunnel decision-making and maintenance method according to claim 1, characterized in that: The obtaining of tunnel static and dynamic data includes: Data collection is performed through the APP, which includes inputting the pile number, selecting the inspection item, selecting the disease, recording the disease attributes, and recording the disease assessment information; The data collected by the data is encrypted for transmission using the elliptic curve encryption algorithm ECC; Fuzzy query of disease records is realized through suffix automaton algorithm; Realize incremental real-time synchronization between the APP and Web terminals through a two-way differential synchronization algorithm; The spatial distribution map of hazards integrating geological risk distribution is drawn using the spline interpolation algorithm.

Images