Highway geological environment analysis method and system integrating GIS and intelligent algorithm

Abstract

The present application relates to the technical field of highway geological environment analysis, and more particularly to a highway geological environment analysis method and system integrating GIS and intelligent algorithms, which comprises acquiring geological state data, historical engineering measure data and historical monitoring data, constructing a causal diagram based on geomechanics causality, identifying road section groups with and without implemented engineering measures and estimating causal effect strength, calculating a predicted risk change value for a target road section by propagating effect with candidate engineering measures as intervention input, searching for a Pareto optimal solution set based on risk change and cost, receiving monitoring data after implementation and updating causal effect strength when deviation exceeds the standard, achieving adaptive optimization through causal inference and dynamic feedback, and improving the accuracy of geological risk prediction and engineering decision-making.

Description

Technical Field

[0001] This invention relates to the field of highway geological environment analysis technology, and in particular to a highway geological environment analysis method and system integrating GIS and intelligent algorithms. Background Technology

[0002] In the field of highway geological environment analysis, current conventional practices typically rely on Geographic Information Systems (GIS) for spatial data management and visualization. These systems combine geological exploration, field monitoring, and historical engineering records, employing empirical formulas, statistical regression, or numerical simulation methods to assess the impact of engineering measures on the geological conditions. These methods emphasize data collection and correlation analysis. For example, by comparing the changing trends of monitoring data from road sections with and without engineering measures, linear or nonlinear models are used to fit historical patterns, and thresholds or weights are set based on expert experience to determine the direction of risk evolution. The assessment results are often directly used for decision-making, such as selecting support or drainage schemes.

[0003] This conventional approach has a significant drawback: it struggles to accurately quantify the causal relationship between engineering measures and changes in geological conditions. Historical data is contaminated with multiple interfering factors, such as variations in geological conditions, climate fluctuations, and construction timing. Correlation-based analyses can easily confuse causal relationships with spurious ones, leading to biased assessments of measure effectiveness. For example, a reduction in risk on a road section might not stem from engineering measures but from natural improvements in geological conditions. However, traditional methods cannot effectively isolate these interfering factors, potentially resulting in resource misallocation or flawed decision-making.

[0004] Another drawback lies in the static nature of the decision-making process. Conventional analysis typically builds assessment models or rules based on historical data in a one-time process, lacking a dynamic feedback mechanism for new monitoring data. Once the construction environment or geological conditions change, the original model cannot self-adjust, causing the prediction accuracy to decay over time. After the engineering measures are implemented, if the actual risk changes do not match the expectations, the existing system cannot automatically correct the assessment parameters, and can only rely on manual remodeling, reducing the adaptability and long-term reliability of the method. Summary of the Invention

[0005] This invention provides a method and system for analyzing highway geological environment that integrates GIS and intelligent algorithms, which can solve the problems in the prior art.

[0006] A first aspect of this invention provides a method for analyzing the geological environment of highways that integrates GIS and intelligent algorithms, comprising:

[0007] Acquire geological condition data, historical engineering measures data, and historical monitoring data along the highway;

[0008] A causal graph containing geological state nodes and engineering measure nodes is constructed based on geomechanical causal relationships; a first road segment group with implemented engineering measures and a second road segment group without implemented engineering measures are identified from the historical engineering measure data; and the intensity of the causal effect from the engineering measure node to the geological state node in the causal graph is estimated based on the historical monitoring data of the two groups.

[0009] For the target road segment to be decided, candidate engineering measures are used as intervention inputs for the engineering measure nodes in the causal graph. The intervention effect is propagated along the directed edges of the causal graph. The predicted risk change value is calculated based on the intensity of the causal effect and the current geological state data of the target road segment.

[0010] Based on the predicted risk change value and cost data, a combination of engineering measures is searched with the goal of maximizing risk reduction and minimizing cost, and a Pareto optimal solution set is obtained.

[0011] Receive monitoring data after implementing the selected scheme in the Pareto optimal solution set, and update the causal effect strength when the deviation between the actual risk change value and the predicted risk change value exceeds a preset deviation threshold.

[0012] A causal graph containing geological state nodes and engineering measure nodes is constructed based on geomechanical causal relationships, including:

[0013] Based on geomechanics theory, a set of geological state variables and a set of engineering measures variables are determined. Each variable in the set of geological state variables is mapped to a geological state node in the causal graph, and each variable in the set of engineering measures variables is mapped to an engineering measures node in the causal graph.

[0014] Based on the mechanical action transmission path in geomechanics, the directed edge connection relationship between nodes is determined. When the variable corresponding to the first node directly affects the variable corresponding to the second node through the mechanical mechanism, a directed edge from the first node to the second node is established between the first node and the second node.

[0015] The time-series observations of node variables are extracted from the historical monitoring data. The conditional independence metric between node pairs connected by directed edges in the causal graph is calculated. If the conditional independence metric does not meet the independence determination condition, the directed edge is retained. If the conditional independence metric meets the independence determination condition, the directed edge is deleted, resulting in the optimized causal graph structure.

[0016] Identify the first road segment group with implemented engineering measures and the second road segment group without implemented engineering measures from the historical engineering measures data. Estimate the causal effect intensity from the engineering measures node to the geological state node in the causal diagram based on the historical monitoring data of the two groups, including:

[0017] The implementation records of engineering measures are extracted from the historical engineering measures data. Based on the implementation records, the road sections are classified into a first road section group where engineering measures were implemented and a second road section group where engineering measures were not implemented. The first road section group is used as the treatment group and the second road section group is used as the control group.

[0018] Extract the geological state observation sequence of the first road section group and the second road section group before and after the implementation time node of the engineering measures from the historical monitoring data, calculate the geological state change of the first road section group as the change of the treatment group, and calculate the geological state change of the second road section group as the change of the control group.

[0019] Construct a set of confounding variables, and perform covariate matching on the first road segment group and the second road segment group based on the set of confounding variables to eliminate the selection bias that affects the selection of engineering measures, and screen out road segment pairs whose confounding variable values ​​are less than a preset variable threshold;

[0020] The difference between the changes in the treatment group and the changes in the control group is calculated based on the selected road segments as the strength of the causal effect. The difference is used to separate the causal effect of engineering measures on the geological conditions through double difference analysis.

[0021] Construct a set of confounding variables, and perform covariate matching on the first road segment group and the second road segment group based on the set of confounding variables, including:

[0022] Extract confounding variables that influence the decision-making process for implementing engineering measures from the historical monitoring data and historical engineering measures data, including geological condition variables, geographical location variables, and traffic flow variables, and construct a set of confounding variables;

[0023] For each confounding variable in the set of confounding variables, a distance metric is selected based on the data type of the variable. For continuous variables, the standardized Euclidean distance is calculated, and for categorical variables, the Hamming distance is calculated. The distance metrics of each confounding variable are weighted and fused to obtain the comprehensive matching distance between road segments.

[0024] For each road segment in the first road segment group, the road segment with the smallest comprehensive matching distance is searched from the second road segment group as the matching road segment. When the minimum comprehensive matching distance exceeds the preset distance threshold, the road segment is marked as an invalid match and removed from subsequent analysis.

[0025] Calculate the distribution differences of the first and second road segment groups after matching on each confusion variable. When the distribution difference is greater than the preset distribution threshold, adjust the weight of the confusion variable and re-execute the matching. Iterate until the distribution difference converges to obtain a matching road segment pair that meets the balance requirements.

[0026] For the target road segment to be decided, candidate engineering measures are used as intervention inputs to the engineering measure nodes in the causal graph. The intervention effect is propagated along the directed edges of the causal graph. Based on the intensity of the causal effect and the current geological state data of the target road segment, the predicted risk change value is calculated, including:

[0027] The current geological state data of the target road segment is mapped to the corresponding geological state nodes in the causal graph as the initial state of the nodes, and the candidate engineering measures are mapped to the intervention amount of the engineering measure nodes;

[0028] Identify all propagation paths from engineering measure nodes to target geological state nodes. For each propagation path, multiply the causal effect intensities of each directed edge along the path to obtain the path causal propagation intensity. Construct an attenuation function based on the number of path nodes to perform nonlinear attenuation correction on the path causal propagation intensity, thus obtaining the path propagation coefficient.

[0029] For an intermediate node receiving multiple upstream propagation paths, the topological features of the upstream propagation paths are extracted and a path interaction correction factor is calculated. Based on the path propagation coefficient, the path interaction correction factor, and the intervention amount, the state increment of the intermediate node is calculated through weighted fusion.

[0030] The updated state value of the intermediate node is obtained by superimposing the state increment with the initial state of the node. The updated state value is used as the intervention input to propagate to downstream nodes, and the update is performed layer by layer until the target geological state node completes the state update.

[0031] The difference between the updated state value and the initial state value of the target geological state node is calculated as the predicted risk change value.

[0032] Based on the predicted risk changes and cost data, a combination of engineering measures is searched with the objective of maximizing risk reduction and minimizing cost, resulting in a Pareto optimal solution set, including:

[0033] Obtain cost data and corresponding predicted risk changes for each candidate engineering measure; extract the common downstream impact nodes for each engineering measure node in the causal graph; and analyze the propagation path topology to the common downstream impact nodes.

[0034] When two propagation paths pass through the same intermediate node before reaching a common downstream affected node, the synergistic enhancement coefficient is calculated; when the two propagation paths produce opposite state changes to the common downstream affected node, the adversarial suppression coefficient is calculated, and a measure interaction matrix is ​​constructed.

[0035] For each combination of engineering measures, the interaction coefficient between any two measures in the combination is obtained by querying the interaction matrix. The predicted risk change value of each engineering measure is weighted and calculated with the corresponding interaction coefficient to obtain the combined risk reduction considering the interaction effect of the measures. The total cost of the combination is obtained by summing the cost data of each engineering measure.

[0036] A dual-objective optimization framework is constructed, with the negative value of the combined risk reduction as the first objective function and the total combined cost as the second objective function. Candidate solutions are initialized and objective function values ​​are calculated. The dominance relationship of candidate solutions is compared to extract the non-dominated solution set. New candidate solutions are generated based on the distribution density of the non-dominated solution set in the objective function space and iteratively updated. The final non-dominated solution set is extracted as the Pareto optimal solution set.

[0037] Receive monitoring data after implementing the selected scheme in the Pareto optimal solution set, and update the causal effect strength when the deviation between the actual risk change value and the predicted risk change value exceeds a preset deviation threshold, including:

[0038] Receive monitoring data collected after implementing the selected scheme in the Pareto optimal solution set, extract geological state parameters from the monitoring data, and calculate the change in geological state parameters before and after implementation as the actual risk change value;

[0039] Calculate the deviation between the actual risk change value and the predicted risk change value corresponding to the selected scheme. When the deviation exceeds a preset deviation threshold, extract the engineering measure nodes and affected geological state nodes related to the selected scheme in the causal graph, and identify the set of directed edges connecting the nodes.

[0040] For each directed edge in the set of directed edges, the causal effect strength correction amount of the directed edge is calculated based on the deviation between the actual risk change value and the predicted risk change value. The causal effect strength correction amount is then superimposed with the current causal effect strength of the directed edge to obtain the updated causal effect strength, which is then stored in the causal graph.

[0041] A second aspect of this invention provides a highway geological environment analysis system integrating GIS and intelligent algorithms, comprising:

[0042] The data acquisition unit is used to acquire geological condition data, historical engineering measures data, and historical monitoring data along the highway.

[0043] Causal construction unit, used to construct a causal graph containing geological state nodes and engineering measure nodes based on geomechanical causal relationships;

[0044] The effect estimation unit is used to identify the first road segment group that implemented engineering measures and the second road segment group that did not implement engineering measures from the historical engineering measures data, and to estimate the intensity of causal effect from the engineering measures node to the geological state node in the causal diagram based on the historical monitoring data of the two groups.

[0045] The risk prediction unit is used to take candidate engineering measures as intervention inputs to the engineering measure nodes in the causal graph for the target road segment to be decided, propagate the intervention effect along the directed edges of the causal graph, and calculate the predicted risk change value based on the intensity of the causal effect and the current geological state data of the target road segment.

[0046] The solution search unit is used to search for a combination of engineering measures based on the predicted risk change value and cost data, with the goal of maximizing risk reduction and minimizing cost, to obtain a Pareto optimal solution set;

[0047] The feedback update unit is used to receive monitoring data after the implementation of the selected scheme in the Pareto optimal solution set, and update the causal effect strength when the deviation between the actual risk change value and the predicted risk change value exceeds a preset deviation threshold.

[0048] A third aspect of the present invention provides an electronic device, comprising:

[0049] processor;

[0050] Memory used to store processor-executable instructions;

[0051] The processor is configured to invoke instructions stored in the memory to execute the aforementioned method.

[0052] A fourth aspect of the present invention provides a computer-readable storage medium having stored thereon computer program instructions that, when executed by a processor, implement the aforementioned method.

[0053] This method, based on causal graphs and the propagation mechanism of intervention effects, significantly improves the accuracy of evaluating the effectiveness of engineering measures in highway geological environment analysis. By identifying road segment groups with and without intervention measures from historical monitoring data, it quantifies the strength of causal effects, avoiding the confounding bias problem in traditional correlation analysis. For the target road segment to be decided, the intervention effect is propagated along the directed edge of the causal graph, and the risk change value can be dynamically calculated and predicted in combination with the current geological state data. This makes the risk assessment not only rely on statistical regularities but also integrate geomechanical causal relationships, resulting in more interpretable and physically plausible results.

[0054] This method employs a Pareto optimal multi-objective optimization strategy to achieve a refined trade-off between risk reduction and cost input. By searching for the Pareto optimal solution set of engineering measure combinations, decision-makers can intuitively obtain a series of candidate solutions that are irreplaceable under both risk-benefit and cost constraints, without relying on single weights or compromises based on human experience. This method integrates GIS spatial data and intelligent algorithms, automatically adapting to the differentiated geological characteristics and governance needs of different road sections, significantly improving decision-making efficiency and the economic viability of solutions. It is particularly suitable for highway engineering scenarios with varied geological conditions, such as complex mountainous areas and high-altitude permafrost regions.

[0055] By introducing a closed-loop adaptive update mechanism, it possesses continuous learning and self-correction capabilities. As engineering practice data accumulates, the parameters of the causal graph gradually approach the actual physical process, and the prediction reliability increases over time, effectively avoiding the performance degradation risk of static models caused by environmental changes or new geological disturbances. Attached Figure Description

[0056] Figure 1 A flowchart illustrating a highway geological environment analysis method integrating GIS and intelligent algorithms;

[0057] Figure 2 This is a flowchart for estimating the causal effects of engineering measures based on difference-in-differences. Detailed Implementation

[0058] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0059] The technical solution of the present invention will be described in detail below with reference to specific embodiments. These specific embodiments can be combined with each other, and the same or similar concepts or processes will not be repeated in some embodiments.

[0060] Figure 1 This is a flowchart illustrating the highway geological environment analysis method integrating GIS and intelligent algorithms, as described in an embodiment of the present invention.

[0061] Highway geological environment analysis methods integrating GIS and intelligent algorithms include:

[0062] Acquire geological condition data, historical engineering measures data, and historical monitoring data along the highway;

[0063] A causal graph containing geological state nodes and engineering measure nodes is constructed based on geomechanical causal relationships;

[0064] Identify the first road segment group that implemented engineering measures and the second road segment group that did not implement engineering measures from the historical engineering measures data, and estimate the intensity of the causal effect from the engineering measures node to the geological state node in the causal diagram based on the historical monitoring data of the two groups;

[0065] For the target road segment to be decided, candidate engineering measures are used as intervention inputs for the engineering measure nodes in the causal graph. The intervention effect is propagated along the directed edges of the causal graph. The predicted risk change value is calculated based on the intensity of the causal effect and the current geological state data of the target road segment.

[0066] Based on the predicted risk change value and cost data, a combination of engineering measures is searched with the goal of maximizing risk reduction and minimizing cost, and a Pareto optimal solution set is obtained.

[0067] Receive monitoring data after implementing the selected scheme in the Pareto optimal solution set, and update the causal effect strength when the deviation between the actual risk change value and the predicted risk change value exceeds a preset deviation threshold.

[0068] In one optional implementation, a causal graph containing geological state nodes and engineering measure nodes is constructed based on geomechanical causal relationships, including:

[0069] Based on geomechanics theory, a set of geological state variables and a set of engineering measures variables are determined. Each variable in the set of geological state variables is mapped to a geological state node in the causal graph, and each variable in the set of engineering measures variables is mapped to an engineering measures node in the causal graph.

[0070] Based on the mechanical action transmission path in geomechanics, the directed edge connection relationship between nodes is determined. When the variable corresponding to the first node directly affects the variable corresponding to the second node through the mechanical mechanism, a directed edge from the first node to the second node is established between the first node and the second node.

[0071] The time-series observations of node variables are extracted from the historical monitoring data. The conditional independence metric between node pairs connected by directed edges in the causal graph is calculated. If the conditional independence metric does not meet the independence determination condition, the directed edge is retained. If the conditional independence metric meets the independence determination condition, the directed edge is deleted, resulting in the optimized causal graph structure.

[0072] For example, in constructing a causal graph, the determination of the geological state variable set needs to be based on geomechanical theory, systematically identifying the core physical quantities affecting the geological stability along the highway. Typical geological state variables include: slope displacement rate, pore water pressure, shear strength parameters of soil and rock, crack propagation width, groundwater depth, and slope erosion. These variables directly reflect the mechanical response state of the geological body under external loads and environmental conditions. Meanwhile, the engineering measure variable set corresponds to practically implementable intervention measures, including variables that quantitatively describe the intensity of engineering intervention, such as anchor bolt support density, drainage ditch layout length, retaining wall height, slope protection coverage area, and grouting reinforcement volume. Each variable in the geological state variable set is mapped to a geological state node in the causal graph, and each variable in the engineering measure variable set is mapped to an engineering measure node, thus forming a complete definition of the node set. The granularity of the nodes should be consistent with the road segment resolution of the GIS spatial analysis to ensure the accuracy of the spatial correspondence in subsequent calculations of the propagation of intervention effects.

[0073] The determination of directed edge connections relies on the physical analysis of the mechanical action transmission path in geomechanics. Taking slope stability analysis as an example, an increase in pore water pressure directly reduces the effective normal stress, thereby weakening the shear strength of the soil and rock mass, ultimately leading to an increase in the slope displacement rate. This mechanical transmission path corresponds to two directed edges in the causal graph, from the "pore water pressure node" to the "shear strength node" and then to the "slope displacement rate node." The directed edge connection method of engineering measure nodes also follows the mechanical mechanism: the drainage engineering node affects the pore water pressure node by lowering the groundwater level, and the anchor support node directly acts on the slope displacement rate node by increasing the constraint force. When the variable corresponding to the first node directly affects the variable corresponding to the second node through the mechanical mechanism, a directed edge from the first node to the second node is established between them; if the influence relationship between the two nodes needs to be transmitted through intermediate nodes, a direct directed edge is not established, but rather expressed through a chain connection of intermediate nodes. This principle of strictly following direct causal relationships in edge construction avoids redundant jump connections in the causal graph and ensures the physical interpretability of the graph structure.

[0074] After constructing the initial causal graph based on prior knowledge of geomechanics, data-driven verification and optimization of the graph structure are needed using historical monitoring data. Time-series observations of each node variable are extracted from the historical monitoring data to form a multivariate time series dataset. For each pair of nodes connected by an established directed edge in the causal graph, a conditional independence metric is calculated. The basic idea of ​​the conditional independence test is: given a common set of parent nodes for two nodes, test whether the observations of these two nodes are statistically independent. If the two nodes still show significant statistical dependence after conditionalizing their parent node sets, it indicates that there is a real direct causal relationship between them, and the corresponding directed edge should be retained; conversely, if the two nodes tend to be statistically independent after conditionalization, it indicates that the direct causal relationship represented by the directed edge is spurious or can be explained by other paths, and it should be deleted.

[0075] Conditional independence measures can be calculated using partial correlation coefficient tests or kernel-based conditional independence tests. For node pairs under the assumption of linear relationships, partial correlation coefficient tests are computationally efficient; for nonlinear geomechanical relationships, kernel-based conditional independence statistics can be used, as they have a stronger ability to capture nonlinear dependencies. Let the nodes... With nodes There exist candidate directed edges, and their condition set is as follows: (i.e., the set of common parent nodes of the two nodes), then calculate the conditional independence statistic. ,when Time (of which) (Assuming a preset independence threshold), if the conditional independence metric is satisfied, the directed edge is deleted; when... If the conditional independence measure value does not meet the independence criterion, the directed edge is retained. Independence criterion threshold. The value of needs to be determined by combining the sample size and significance level of historical monitoring data. When the sample size is small, the threshold should be appropriately relaxed to avoid excessive edge deletion that could lead to distortion of the causal graph structure.

[0076] In practical engineering scenarios, monitoring data of the geological environment along highways often contain missing and outlier values, requiring data quality preprocessing before extracting time-series observations. For short-term missing values ​​caused by sensor malfunctions, linear interpolation can be used to fill in the gaps; for abnormal peak values ​​caused by extreme geological events, the professional judgment of geological engineers is needed to determine whether to include them in the conditional independence calculation. Furthermore, due to the significant time lag effect of geological processes—for example, the increase in pore water pressure caused by rainfall infiltration typically lags by several hours to several days—an appropriate time lag window should be introduced when calculating conditional independence metrics to account for node... At any moment Observations and nodes At any moment The observed values ​​were correlated, among which These are time delay parameters estimated based on geomechanical mechanisms.

[0077] After edge deletion following conditional independence testing, an optimized causal graph structure is obtained. This structure retains the physical causal relationships supported by geomechanical theory while eliminating weakly correlated edges that lack statistical support in actual monitoring data through data validation, thus achieving a balance between physical interpretability and data fit. The optimized causal graph structure will serve as the basic topological framework for subsequently estimating the causal effect intensity from engineering measure nodes to geological state nodes. The accuracy of the graph structure directly affects the reliability of intervention effect propagation calculations and predicted risk change value estimations. After determining the causal graph structure, the directed acyclicity of the graph needs to be verified to ensure that there are no directed loops, thus meeting the basic requirements of causal inference theory for directed acyclic graphs. If a loop is detected, the edge directions in the loop need to be corrected or the weakest edge needs to be deleted to break the loop, based on geomechanical expert knowledge.

[0078] In one optional implementation, identifying a first road segment group with implemented engineering measures and a second road segment group without implemented engineering measures from the historical engineering measures data, and estimating the causal effect intensity from the engineering measures node to the geological state node in the causal graph based on the historical monitoring data of the two groups, including:

[0079] The implementation records of engineering measures are extracted from the historical engineering measures data. Based on the implementation records, the road sections are classified into a first road section group where engineering measures were implemented and a second road section group where engineering measures were not implemented. The first road section group is used as the treatment group and the second road section group is used as the control group.

[0080] Extract the geological state observation sequence of the first road section group and the second road section group before and after the implementation time node of the engineering measures from the historical monitoring data, calculate the geological state change of the first road section group as the change of the treatment group, and calculate the geological state change of the second road section group as the change of the control group.

[0081] Construct a set of confounding variables, and perform covariate matching on the first road segment group and the second road segment group based on the set of confounding variables to eliminate the selection bias that affects the selection of engineering measures, and screen out road segment pairs whose confounding variable values ​​are less than a preset variable threshold;

[0082] The difference between the changes in the treatment group and the changes in the control group is calculated based on the selected road segments as the strength of the causal effect. The difference is used to separate the causal effect of engineering measures on the geological conditions through double difference analysis.

[0083] For example, in combinationFigure 2 This paper describes the flowchart for estimating the causal effects of engineering measures based on the difference-in-differences method. When extracting implementation records from historical engineering measure data, each record needs to be structured and analyzed, including the type of engineering measure (e.g., anchoring reinforcement, drainage modification, slope protection, grouting reinforcement), implementation time, road segment number, and corresponding geographical coordinates. Based on these records, all road segments within the study area are divided into two groups: road segments with clear engineering measure implementation records within the analysis time window are assigned to the first group (treatment group); road segments without any engineering measure intervention records within the same time window are assigned to the second group (control group). During the division process, it is important to note that if an engineering measure is implemented on a road segment only for a portion of the analysis window, the first implementation time is used as the treatment time for that road segment, ensuring clear boundaries and no overlap between the treatment and control groups.

[0084] From historical monitoring data, geological condition observation sequences were extracted for each road segment, covering two consecutive observation periods before and after the implementation of engineering measures. These observation sequences encompassed various geological condition indicators, including slope displacement rate, crack width variation, groundwater level fluctuation, and soil moisture content. For each road segment in the first segment group, the mean values ​​of geological condition indicators were calculated for the observation periods before and after implementation, using the implementation time of the engineering measures as a dividing line. The difference between these two values ​​represents the treatment group change for that road segment, denoted as […]. Where positive values ​​indicate an increase in the indicator and negative values ​​indicate a decrease. For each road segment in the second road segment group, using the corresponding time node as a reference (usually the same time node aligned with the treatment group road segments in time), the mean difference between the two periods is calculated to obtain the change in the control group, denoted as . The changes in the control group reflect the natural evolution trend of the geological state under conditions without engineering intervention, including the combined effects of seasonal factors, regional rainfall events, and changes in the geological background over time.

[0085] Before estimating causal effects, a set of confounding variables needs to be constructed to eliminate selection bias. Selection bias refers to the tendency for engineering measures to be preferentially implemented on road sections with higher geological risks or obvious signs of deformation, leading to systematic differences in the baseline geological conditions of the treatment group and the control group before the implementation of engineering measures. If left uncontrolled, this can overestimate or underestimate the true effect of the engineering measures. The set of confounding variables typically includes the following categories: topographic features (such as slope, aspect, and catchment area), soil and rock properties (such as lithology, joint density, and soil cohesion), hydrological conditions (such as average annual rainfall and groundwater depth), and baseline geological state indicators before the implementation of engineering measures (such as the mean displacement rate in the early stages of implementation). These confounding variables are combined into a covariate vector, denoted as [vector name missing]. Its weight Indicates the first The values ​​of the confounding variables.

[0086] Based on covariate vectors Covariate matching was performed on the first and second road segment groups. The matching method employed a nearest neighbor matching strategy: for each road segment in the first group, the road segment in the second group with the closest covariate value was identified as its matching target. The matching distance was measured using Mahalanobis distance or propensity score distance. Defined as given covariates The conditional probability of a road segment being assigned to a processing group under certain conditions, i.e. The data was obtained by fitting historical data using a logistic regression model. After matching, for each road segment pair, the differences in the values ​​of its confounding variables were examined to see if they met the constraints of preset variable thresholds: if the differences in the values ​​of all confounding variables for a road segment pair were less than the preset variable thresholds, then the road segment pair was retained for subsequent analysis; otherwise, it was removed to ensure matching quality. After screening, the retained road segment pairs constituted a balanced sample, and the treatment group and the control group tended to be consistent in the distribution of covariates, thus effectively eliminating the selection bias caused by differences in geological conditions.

[0087] Based on the selected road segment pairs, the difference-in-differences method was used to calculate the intensity of causal effects. The core idea of ​​the difference-in-differences method is that, under the premise of controlling for common temporal trends, the difference between the changes in the treatment group and the control group can separate the net causal effect of engineering measures on the geological state, unaffected by external common temporal trends (such as the acceleration of overall displacement due to increased regional rainfall). For the first... For each matching road segment pair, its difference-in-differences estimator Defined as: ;in For the first The geological condition changes of each road segment were processed in the central section. This represents the change in geological conditions corresponding to the control group road sections. For all retained... The overall causal effect strength estimate is obtained by taking a weighted average of the values ​​from each road segment. : ; The physical meaning is: after excluding the effects of selection bias and common time trends, the average net effect of implementing engineering measures on changes in geological state indicators compared to not implementing engineering measures. When the value is negative, it indicates that the engineering measures have effectively reduced the corresponding geological condition indicators (such as slope displacement rate), and the causal effect is a positive inhibitory effect; when If the value is positive, it is necessary to further analyze the correspondence between the engineering measures type and the geological condition indicators to determine whether there is a short-term increase in indicators caused by construction disturbance.

[0088] For different types of engineering measures (such as anchoring, drainage, and grouting), corresponding road segment subsets are constructed, and the causal effect intensity of each type of engineering measure on various geological state nodes is independently estimated. Finally, a causal effect intensity matrix is ​​formed, where rows correspond to engineering measure nodes, columns correspond to geological state nodes, and matrix elements are the corresponding... The value. This matrix will serve as the weights of the directed edges in the causal graph, used for calculating the propagation of subsequent intervention effects and predicting changes in risk. In cases with a small sample size, further calculations are also needed. Bootstrap confidence interval estimation is performed to assess the statistical reliability of the causal effect strength. Only the causal effect strength whose confidence interval does not contain zero values ​​is included in subsequent analysis to avoid misjudging statistical noise as a real causal effect.

[0089] In one optional implementation, a set of confounding variables is constructed, and covariate matching is performed on the first road segment group and the second road segment group based on the set of confounding variables, including:

[0090] Extract confounding variables that influence the decision-making process for implementing engineering measures from the historical monitoring data and historical engineering measures data, including geological condition variables, geographical location variables, and traffic flow variables, and construct a set of confounding variables;

[0091] For each confounding variable in the set of confounding variables, a distance metric is selected based on the data type of the variable. For continuous variables, the standardized Euclidean distance is calculated, and for categorical variables, the Hamming distance is calculated. The distance metrics of each confounding variable are weighted and fused to obtain the comprehensive matching distance between road segments.

[0092] For each road segment in the first road segment group, the road segment with the smallest comprehensive matching distance is searched from the second road segment group as the matching road segment. When the minimum comprehensive matching distance exceeds the preset distance threshold, the road segment is marked as an invalid match and removed from subsequent analysis.

[0093] Calculate the distribution differences of the first and second road segment groups after matching on each confusion variable. When the distribution difference is greater than the preset distribution threshold, adjust the weight of the confusion variable and re-execute the matching. Iterate until the distribution difference converges to obtain a matching road segment pair that meets the balance requirements.

[0094] For example, before estimating the causal effects of engineering measures, it is necessary to systematically identify and control for confounding variables to ensure comparability between the first road segment group (the group with engineering measures implemented) and the second road segment group (the group without engineering measures implemented). Confounding variables are variables that simultaneously affect both the decision to implement engineering measures and the geological condition outcome; if not controlled, causal effect estimation will introduce systematic bias. Three types of confounding variables are extracted from historical monitoring data and historical engineering measure data: geological condition variables, geographical location variables, and traffic flow variables, which together constitute the set of confounding variables.

[0095] Geological condition variables include lithology classification, fault distance, stratum dip angle, uniaxial compressive strength of rock, and groundwater depth. These indicators directly reflect the geological vulnerability of the road section and are core references for decision-makers when choosing whether to implement engineering measures. Geographical location variables include elevation, slope, aspect, distance from major rivers, and topographic relief. These variables characterize the macroscopic topographic environment of the road section and have a significant impact on the development background of geological hazards. Traffic flow variables include average daily traffic volume, proportion of heavy-duty vehicles, and age of the road. These indicators reflect the dynamic load level borne by the road section and affect the fatigue damage of the soil and rock mass and the priority of engineering measures. After including the above three types of variables into the confounding variable set, it is necessary to select an appropriate distance measurement method for each variable's data type.

[0096] For continuous variables (such as fault distance, slope, and average daily traffic volume), standardized Euclidean distance is used for measurement. Standardization eliminates the influence of different dimensions and orders of magnitude, making the contribution scales of each continuous variable comparable in distance calculations. Let a certain road segment in the first road segment group be at the [missing information - likely a specific location or point]. The values ​​for each continuous confounding variable are... In the second road segment group, the value of a candidate matching road segment on the same variable is... The standard deviation of this variable across the entire road segment sample is Then the standardized Euclidean distance component of the two road segments on this variable is: For categorical variables (such as lithology classification, slope aspect category, etc.), Hamming distance is used for measurement: if two road segments have the same value for the categorical variable, the distance component is 0; if they have different values, the distance component is 1. This approach avoids the spurious size relationships introduced by unreasonable numerical coding of categorical data.

[0097] After obtaining the distance component of each confusion variable, the components are weighted and fused to obtain the comprehensive matching distance between the two road segments. Let the set of confusion variables contain a total of... The variable, the first The weights of the variables are The distance component corresponding to this variable is Then the overall matching distance The calculation method is as follows The weights satisfy and In the initial stage, the weights of each variable can be set to equal values, and then dynamically adjusted through iterative balance checks.

[0098] For each road segment in the first road segment group, traverse all candidate road segments in the second road segment group, calculate the comprehensive matching distance, and select the road segment with the smallest distance as its matching road segment, forming a one-to-one nearest neighbor matching pair. When the calculated minimum comprehensive matching distance exceeds a preset distance threshold... If the condition is met, it indicates that there is no sufficiently similar control segment in the second segment group to the processed segment. Forced matching would introduce significant confounding bias; therefore, the segment is marked as an invalid match and removed from subsequent causal effect analysis. (Preset distance threshold) The range can be set according to the size of the dataset and the requirements for matching quality. A reasonable range is usually determined by referring to the empirical distribution of the comprehensive matching distance (such as taking a certain percentile of the distance between all road segments).

[0099] After initial matching, the quality of the matching is assessed through a balance test. For each confounding variable, the distributional differences of the first and second road segment groups on that variable are calculated separately after matching. For continuous variables, the standardized mean difference (i.e., the difference between the two group means divided by the pooled standard deviation) is used as a quantitative indicator of distributional differences, denoted as […]. For categorical variables, the mean absolute value of the frequency differences between the two groups in each category is calculated as the distribution difference index. When the distribution difference index of a variable exceeds a preset distribution threshold... When the standardized mean difference is 0.1 (which is usually set as the acceptable upper limit), it is considered that the balance of the variable does not meet the requirements and its weight in the overall matching distance needs to be adjusted.

[0100] The weight adjustment strategy is as follows: for variables whose distribution differences exceed a threshold, their corresponding weights are increased. This makes the variable contribute more to the overall matching distance when re-performing the matching, thus driving the matching algorithm to prioritize finding road segment pairs that are more similar to that variable. For variables that have already met the balance requirements, their weights are appropriately reduced, tilting matching resources towards variables that are not yet balanced. The magnitude of the weight adjustment can be proportional to the degree to which the distribution difference exceeds the threshold, with the specific adjustment amount being... ,in This is the step size coefficient, controlling the adjustment range for each iteration. After adjustment, the weight vector is renormalized to ensure that the sum of all weights remains 1.

[0101] After adjusting the weights, the comprehensive matching distance for all road segment pairs is recalculated using the updated weights, and nearest neighbor matching and invalid match removal are re-executed. A balance test is then performed on the newly formed matching road segment pairs. If the distribution differences of all confounding variables are below a preset distribution threshold... If the distribution differences converge, the matching process is considered to have converged, and the matching process terminates; otherwise, the weights are adjusted and the process continues to iterate. To prevent the iteration process from failing to converge, a maximum number of iterations is set. When the maximum number of iterations is reached, the current optimal matching result (i.e., the matching scheme with the minimum sum of the distribution differences of each variable in each iteration) is output as the final set of matching road segment pairs.

[0102] Through the aforementioned iterative matching process, a set of matched road segment pairs that meets the balance requirements is finally obtained. In this set, the distributions of confounding variables such as geological conditions, geographical location, and traffic flow are highly similar in each pair of road segments. The systematic differences between the two groups mainly stem from the implementation of engineering measures rather than the interference of confounding factors. This lays a data foundation for subsequent estimation of the causal effect of engineering measures on geological conditions based on the difference-in-differences method, effectively improving the reliability and unbiasedness of the causal effect estimation results.

[0103] In one optional implementation, for the target road segment to be decided, candidate engineering measures are used as intervention inputs to the engineering measure nodes in the causal graph. The intervention effect is propagated along the directed edges of the causal graph. A predicted risk change value is calculated based on the intensity of the causal effect and the current geological state data of the target road segment, including:

[0104] The current geological state data of the target road segment is mapped to the corresponding geological state nodes in the causal graph as the initial state of the nodes, and the candidate engineering measures are mapped to the intervention amount of the engineering measure nodes;

[0105] Identify all propagation paths from engineering measure nodes to target geological state nodes. For each propagation path, multiply the causal effect intensities of each directed edge along the path to obtain the path causal propagation intensity. Construct an attenuation function based on the number of path nodes to perform nonlinear attenuation correction on the path causal propagation intensity, thus obtaining the path propagation coefficient.

[0106] For an intermediate node receiving multiple upstream propagation paths, the topological features of the upstream propagation paths are extracted and a path interaction correction factor is calculated. Based on the path propagation coefficient, the path interaction correction factor, and the intervention amount, the state increment of the intermediate node is calculated through weighted fusion.

[0107] The updated state value of the intermediate node is obtained by superimposing the state increment with the initial state of the node. The updated state value is used as the intervention input to propagate to downstream nodes, and the update is performed layer by layer until the target geological state node completes the state update.

[0108] The difference between the updated state value and the initial state value of the target geological state node is calculated as the predicted risk change value.

[0109] For example, after constructing the causal graph and estimating the intensity of causal effects, for the target road segment to be decided, candidate engineering measures need to be introduced into the causal graph and their impact on the geological state needs to be simulated to calculate the predicted risk change value. The current geological state data of the target road segment includes multi-dimensional geological monitoring indicators such as slope displacement rate, groundwater level, crack width, and soil and rock strength parameters. These indicators are mapped one by one to the corresponding geological state nodes in the causal graph as the initial state values ​​of each node. Candidate engineering measures—such as anchor reinforcement, drainage diversion, slope protection, grouting reinforcement, etc.—are mapped to the intervention quantities of the engineering measure nodes in the causal graph. The numerical value of the intervention quantity represents the quantitative description of the engineering measure in terms of strength, scale, or construction parameters, such as the magnitude of anchor prestress, drainage hole density, or grouting pressure.

[0110] When identifying all propagation paths from engineering measure nodes to target geological state nodes, a depth-first traversal is performed along the directed edges of the causal graph to enumerate the set of all paths satisfying directed connectivity. For each propagation path, the causal effect intensities corresponding to each directed edge on that path are multiplied sequentially to obtain the original path causal propagation intensity. Suppose a propagation path passes through... If there are 12 nodes (including the starting point engineering measures node and the ending point target geological state node), then the path contains 12 nodes. A directed edge, the causal effect strength of each edge is denoted as . Original path causal propagation strength for: ;in For path numbering, Let be the index of the directed edge on the path. Since causal signals gradually weaken as they propagate along longer paths during geological processes due to buffering and dissipation at intermediate nodes, an attenuation function needs to be introduced. Nonlinear attenuation correction is performed. The attenuation function is based on the number of path nodes. The path propagation coefficient is constructed using an exponential decay form and after decay correction. for: ;in The attenuation rate hyperparameter reflects the rate attenuation of geological causal signals as the number of propagation hops increases. This indicates the number of intermediate nodes in the path (i.e., the number of nodes after removing the start and end points). The value of can be calibrated based on the actual effects of different path lengths in historical engineering measures data, and is usually selected between 0.1 and 0.5. When The time path is a direct causal edge with no intermediate decay. It aligns with intuition.

[0111] When an intermediate node in a causal graph simultaneously receives intervention signals from multiple upstream propagation paths, it is necessary to address the signal superposition and interaction effects caused by the convergence of multiple paths. Topological features of the upstream propagation paths are extracted, including the number of upstream paths merging into the intermediate node, the propagation coefficient of each upstream path, and whether there is a common ancestor node among the upstream paths (i.e., path correlation). If multiple upstream paths share a directed edge or branch from the same upstream node, the intervention signals carried by these paths are correlated; directly accumulating them would lead to duplicate effect counting. Therefore, a path interaction correction factor is calculated. Its definition is: ;in This represents the number of path pairs in the upstream path that share edges or nodes. This represents the total number of upstream paths. The intensity coefficient is an interactive correction factor, with a value range of [value range missing]. This is used to control the strength of deduplication correction. When the upstream paths are completely independent (no shared nodes or edges), , No correction is made when the upstream paths highly overlap. Tend to This compresses the superposition effect.

[0112] intermediate nodes State increment It is obtained through weighted fusion calculation. Let the inflow node be... The upstream path set is Each path The corresponding path propagation coefficient is Intervention amount is (i.e., the intervention input quantity of the engineering measure node), then:

[0113]

[0114] in For nodes The corresponding path interaction correction factor. For intermediate nodes that only receive a single upstream path, , It degenerates into a simple product of single-path propagation. The physical meaning of weighted fusion is that engineering measures affect the same intermediate node through different paths in the causal graph, and the contributions of each path are superimposed with their propagation coefficients as weights. At the same time, overestimation caused by path correlation is avoided through correction factors.

[0115] Increment the state With nodes initial state value By superimposing the values, we obtain the updated state value of the intermediate node. : Updated status value As a node The new intervention input is propagated to its downstream nodes, replacing the original intervention. The process continues downstream along the directed edges. This layer-by-layer iterative update mechanism processes each intermediate node sequentially according to the topological order of the causal graph (i.e., starting from the engineering measure node and advancing layer by layer along the directed edges) until the target geological state node completes its state update. Topological ordering ensures that each node performs its own state calculation only after all upstream nodes have completed their updates, avoiding inconsistencies caused by incorrect calculation order.

[0116] After the target geological state node completes its iterative update, its updated state value is denoted as: The initial state value is denoted as Predicted risk change value Defined as the difference between the two: ; A negative value indicates that the engineering measures are expected to reduce the risk indicators represented by the target geological state node (such as the probability of slope instability or displacement rate), and the larger the absolute value, the more significant the risk reduction effect. A positive value indicates that the candidate engineering measure exacerbates the risk and needs to be excluded or adjusted during the scheme selection stage. For cases where multiple candidate engineering measures intervene in parallel, the intervention amount corresponding to each candidate measure is simultaneously injected into its respective engineering measure node. The comprehensive predicted risk change value of the target node is calculated according to the same propagation, attenuation, correction, and iteration process described above, providing an accurate quantitative basis for risk reduction in subsequent Pareto multi-objective optimization.

[0117] In one optional implementation, based on the predicted risk change value and cost data, a combination of engineering measures is searched with the objective of maximizing risk reduction and minimizing cost to obtain a Pareto optimal solution set, including:

[0118] Obtain cost data and corresponding predicted risk changes for each candidate engineering measure; extract the common downstream impact nodes for each engineering measure node in the causal graph; and analyze the propagation path topology to the common downstream impact nodes.

[0119] When two propagation paths pass through the same intermediate node before reaching a common downstream affected node, the synergistic enhancement coefficient is calculated; when the two propagation paths produce opposite state changes to the common downstream affected node, the adversarial suppression coefficient is calculated, and a measure interaction matrix is ​​constructed.

[0120] For each combination of engineering measures, the interaction coefficient between any two measures in the combination is obtained by querying the interaction matrix. The predicted risk change value of each engineering measure is weighted and calculated with the corresponding interaction coefficient to obtain the combined risk reduction considering the interaction effect of the measures. The total cost of the combination is obtained by summing the cost data of each engineering measure.

[0121] A dual-objective optimization framework is constructed, with the negative value of the combined risk reduction as the first objective function and the total combined cost as the second objective function. Candidate solutions are initialized and objective function values ​​are calculated. The dominance relationship of candidate solutions is compared to extract the non-dominated solution set. New candidate solutions are generated based on the distribution density of the non-dominated solution set in the objective function space and iteratively updated. The final non-dominated solution set is extracted as the Pareto optimal solution set.

[0122] For example, after obtaining the cost data and corresponding predicted risk changes for each candidate engineering measure, it is necessary to further analyze the interaction effects generated when multiple engineering measures are implemented simultaneously. This is because when different measures jointly influence the same downstream geological state node through directed edges in the causal graph, their combined effect is not simply equal to the linear sum of their individual effects. Therefore, common downstream influence nodes for each engineering measure node in the causal graph are extracted. That is, in the topological structure of the directed graph, a reachability traversal is performed along directed edges from each engineering measure node to find geological state nodes that are commonly reached by two or more engineering measure nodes. These common downstream influence nodes and their corresponding upstream propagation paths are recorded to form a path topology description.

[0123] For any two propagation paths leading to the same common downstream node, their topological relationships are analyzed to determine whether synergistic or antagonistic effects exist. When both propagation paths pass through the same intermediate node before reaching the common downstream node, it indicates that causal signals converge and superimpose at that intermediate node, and the two engineering measures exert a synergistic strengthening effect on the target node through the shared intermediate geological conditions. In this case, the synergistic enhancement coefficient is calculated. Its value is a normalized estimate based on the ratio of the product of the path propagation coefficients of the two paths at the shared intermediate node to the sum of their path propagation coefficients. Specifically, ,in and These are the path propagation coefficients of the two propagation paths after attenuation correction. The synergistic amplification intensity coefficient is used to control the amplification magnitude of the synergistic effect. Its value is determined by calibrating cases of joint implementation of multiple measures in historical engineering data.

[0124] When two propagation paths produce opposite state changes to a node with a common downstream influence—that is, one path increases the state value of the target geological state node while the other decreases it—a counteracting inhibition effect exists between the two measures. In this case, the counteracting inhibition coefficient is calculated. Its value is determined based on the difference in the sign and the ratio of the magnitude of the state changes caused by the two paths at the target node. ,in and These represent the predicted risk changes caused by each of the two paths. To counteract the suppression intensity coefficient, the value range is: This ensures that the resistance inhibition coefficient is always positive and does not exceed 1. For path pairs that neither share intermediate nodes nor produce opposite effects, their interaction coefficient is set to 1 by default, indicating no interaction effect. All pairwise interaction coefficients between engineering measure nodes are organized into a measure interaction matrix. The first in the matrix Line number Column elements That is, the first Project measures and the first The interaction coefficient between the engineering measures, when hour .

[0125] For each candidate engineering measure combination scheme, query the measure interaction matrix. Obtain the interaction coefficient between any two measures in the combination. For combinations containing The combined engineering measures scheme involves weighting the predicted risk changes of each measure with its corresponding interaction coefficient to obtain the combined risk reduction amount considering the interaction effects of the measures. The specific calculation method is as follows: based on the predicted risk change value when each engineering measure is implemented individually, the interaction coefficient of each pair of measures in the combination is cumulatively corrected, i.e. ,in For the first The predicted risk change value of the project measures For the first The measures and the first The interaction coefficients between the measures. Simultaneously, the total cost of the combination is obtained by summing the cost data of each engineering measure within the combination. ,in For the first The implementation cost of the project measures.

[0126] After obtaining the reduction in combined risk and the total combined cost for each candidate combination, a bi-objective optimization framework is constructed. The negative value of the reduction in combined risk is then used as the optimization target. As the first objective function Minimize This is equivalent to maximizing risk reduction; reducing the total cost of the portfolio. As the second objective function , minimize This corresponds to minimizing the total cost of project implementation. When initializing the candidate solution set, several initial candidate solutions are generated by randomly sampling from all engineering measure combination spaces. Each candidate solution corresponds to a specific engineering measure combination scheme, and its... and The objective function value, in the case of and The distribution of the initial solution is formed in the two-dimensional objective function space.

[0127] When comparing the dominance relationships between candidate solutions, a candidate solution is said to dominate the latter if it is not inferior to another candidate solution on both objective functions and is strictly superior to the other candidate solution on at least one objective function. The current set of candidate solutions is traversed, and all solutions not dominated by any other candidate solution are extracted to form the current non-dominated solution set, which is the approximation of the current Pareto front. When generating new candidate solutions based on the distribution density of the non-dominated solution set in the objective function space, the crowding distance between each non-dominated solution and its neighboring solutions in the objective function space is calculated. A larger crowding distance indicates a sparser solution distribution in that region. New candidate solutions are preferentially generated near regions with larger crowding distances to ensure the approximate uniform coverage of the Pareto front. The generation of new candidate solutions includes combining two randomly selected solutions from the non-dominated solutions and randomly perturbing the engineering measures selection states in existing solutions, thereby performing a directional search in the solution space.

[0128] The newly generated candidate solutions are merged with the current candidate solution set. The objective function values ​​of all solutions are recalculated, and the dominance comparison and non-dominated solution set extraction are performed again. This process is iteratively updated according to the above procedure until the number of iterations reaches a preset upper limit or the non-dominated solution set no longer changes significantly in a number of consecutive iterations. After the iteration is completed, the final non-dominated solution set is extracted as the Pareto optimal solution set. Each solution in this Pareto optimal solution set represents a combination of engineering measures that achieves a specific trade-off between risk reduction and cost control. Decision-makers can select the solution that best meets the actual engineering needs from the Pareto optimal solution set based on actual budget constraints and risk tolerance, thereby achieving effective control of geological risks of the target road section under limited resource conditions.

[0129] In one optional implementation, monitoring data is received after the implementation of the selected scheme in the Pareto optimal solution set. When the deviation between the actual risk change value and the predicted risk change value exceeds a preset deviation threshold, the causal effect strength is updated, including:

[0130] Receive monitoring data collected after implementing the selected scheme in the Pareto optimal solution set, extract geological state parameters from the monitoring data, and calculate the change in geological state parameters before and after implementation as the actual risk change value;

[0131] Calculate the deviation between the actual risk change value and the predicted risk change value corresponding to the selected scheme. When the deviation exceeds a preset deviation threshold, extract the engineering measure nodes and affected geological state nodes related to the selected scheme in the causal graph, and identify the set of directed edges connecting the nodes.

[0132] For each directed edge in the set of directed edges, the causal effect strength correction amount of the directed edge is calculated based on the deviation between the actual risk change value and the predicted risk change value. The causal effect strength correction amount is then superimposed with the current causal effect strength of the directed edge to obtain the updated causal effect strength, which is then stored in the causal graph.

[0133] For example, after the selected scheme is implemented and a certain observation period has elapsed, monitoring data is collected from the sensor network deployed on the target road section, manual inspection records, and remote sensing images. Geological state parameters are extracted, including key indicators such as slope displacement rate, crack width, groundwater level change, soil moisture content, and pore water pressure. The observed values ​​of each geological state parameter after the implementation of the engineering measures are subtracted from the baseline values ​​before implementation to obtain the actual changes in each parameter. Then, based on the same risk scoring rules used in calculating the predicted risk change value, the changes in multiple geological state parameters are weighted and merged into a unified actual risk change value. To make it consistent with the predicted risk change value Maintaining consistency in dimensions and calculation methods provides a comparable numerical basis for subsequent deviation calculations.

[0134] Calculate the actual risk change value Compared with the predicted risk change value Deviation between ,Right now .Will Deviation threshold from preset Comparison: If If the prediction accuracy of the causal graph for that road segment is considered to be within an acceptable range, then there is no need to trigger an update process; if This indicates a systematic bias in the current causal effect strength, requiring the initiation of a causal effect strength correction process. A preset bias threshold is set. The value of is usually determined based on the statistical distribution of historical prediction errors. For example, it is set to the mean of the absolute values ​​of historical prediction errors plus 1.5 times the standard deviation, in order to avoid frequent updates caused by random fluctuations, while ensuring timely response when significant deviations occur.

[0135] After an update is triggered, the engineering measure nodes directly related to the selected scheme are located in the cause-effect graph, and all geological state nodes affected by these engineering measure nodes are traced along the directed edges, forming a set of affected nodes. Within this set, all directed edges connecting the engineering measure nodes and geological state nodes are identified, including directly connected edges and indirect edges that transmit influence through intermediate geological state nodes, constituting the set of directed edges to be updated. For indirect paths that cross multiple intermediate nodes, the path needs to be broken down into directed edges and included separately. This allows for refined adjustments to the entire propagation chain, rather than simply adjusting the end-to-end aggregate effect.

[0136] against For each directed edge in the array, calculate its causal effect strength correction. Considering bias... It is the accumulation of errors from each directed edge along the entire propagation path, and the total deviation needs to be decomposed according to the contribution proportion of each directed edge. Specifically, for the set of directed edges... The Middle A directed edge, its correction amount Calculate using the following formula: ,in The strength of the causal effect of the directed edge before the update. The learning rate hyperparameter controls the step size for each update. This is the sum of the causal effect strengths of all edges in the directed edge set, used to normalize the contribution weight of each edge. The physical meaning of this decomposition method is that the directed edge with the greater original causal effect strength contributes more to the prediction result and therefore bears more responsibility for correction.

[0137] After calculating the correction amount, the correction amount is added to the current causal effect strength to obtain the updated causal effect strength. ,Right now To prevent unreasonable and significant jumps in the strength of causal effects due to a single abnormal monitoring data point, constraints need to be imposed on the updated values: if Exceeding the pre-set reasonable range If the value is too high, it will be truncated to the boundary value of the interval. The upper and lower bounds of a reasonable interval can be determined based on the knowledge of domain experts or historical statistical data. For example, for a directed edge between drainage engineering measures and groundwater level nodes, the reasonable range of the causal effect strength can be set with reference to the statistical results of a large number of engineering cases, so as to ensure that the updated causal graph is still physically reasonable.

[0138] Learning rate hyperparameter The settings have a significant impact on the stability and convergence speed of the update process. When When the value is too large, the single update amplitude is too large, causing the causal effect strength to oscillate repeatedly around the true value; when When the value is too small, the adaptive correction speed of the causal graph is too slow, making it difficult to respond promptly to substantial changes in the geological environment. Therefore, an adaptive learning rate strategy can be adopted: when the deviation direction of the same directed edge is consistent in multiple consecutive updates (i.e., ... When the symbols are the same, increase the value appropriately. Accelerate the correction speed; when the deviation direction reverses, appropriately reduce... This suppresses oscillations. The adaptive adjustment rule can be expressed as: or ,in and These are the proportionality coefficients for increasing and decreasing, typically taken as... , .

[0139] Updated causal effect strength Write the edge attribute storage structure into the causal graph, replacing the original one. Record the timestamp and trigger deviation value of this update. The system generates a historical update log of the causal effect intensity, including the corresponding road segment number. This log not only tracks the evolution of the causal graph but also, after accumulating sufficient update records, identifies the long-term drift trend of the causal effect intensity of specific directed edges through statistical analysis, thus providing data support for the study of macroscopic changes in the geological environment. When monitoring data from multiple road segments triggers updates to the same directed edge, the weighted average of the corrections calculated for each road segment is taken as the final correction. The weights are determined based on the geological similarity between each road segment and the target directed edge, avoiding excessive interference from data from individual abnormal road segments on the causal effect intensity of the common edge. After multiple rounds of monitoring and update iterations, the causal effect intensity of each directed edge in the causal graph will gradually approach the actual geomechanical response law, making the subsequent calculation of predicted risk changes for new target road segments more accurate and reliable.

[0140] A second aspect of this invention provides a highway geological environment analysis system integrating GIS and intelligent algorithms, comprising:

[0141] The data acquisition unit is used to acquire geological condition data, historical engineering measures data, and historical monitoring data along the highway.

[0142] Causal construction unit, used to construct a causal graph containing geological state nodes and engineering measure nodes based on geomechanical causal relationships;

[0143] The effect estimation unit is used to identify the first road segment group that implemented engineering measures and the second road segment group that did not implement engineering measures from the historical engineering measures data, and to estimate the intensity of causal effect from the engineering measures node to the geological state node in the causal diagram based on the historical monitoring data of the two groups.

[0144] The risk prediction unit is used to take candidate engineering measures as intervention inputs to the engineering measure nodes in the causal graph for the target road segment to be decided, propagate the intervention effect along the directed edges of the causal graph, and calculate the predicted risk change value based on the intensity of the causal effect and the current geological state data of the target road segment.

[0145] The solution search unit is used to search for a combination of engineering measures based on the predicted risk change value and cost data, with the goal of maximizing risk reduction and minimizing cost, to obtain a Pareto optimal solution set;

[0146] The feedback update unit is used to receive monitoring data after the implementation of the selected scheme in the Pareto optimal solution set, and update the causal effect strength when the deviation between the actual risk change value and the predicted risk change value exceeds a preset deviation threshold.

[0147] A third aspect of the present invention provides an electronic device, comprising:

[0148] processor;

[0149] Memory used to store processor-executable instructions;

[0150] The processor is configured to invoke instructions stored in the memory to execute the aforementioned method.

[0151] A fourth aspect of the present invention provides a computer-readable storage medium having stored thereon computer program instructions that, when executed by a processor, implement the aforementioned method.

[0152] This invention can be a method, apparatus, system, and / or computer program product. The computer program product may include a computer-readable storage medium having computer-readable program instructions loaded thereon for performing various aspects of the invention.

[0153] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A highway geological environment analysis method integrating GIS and intelligent algorithms, characterized in that, include: Acquire geological condition data, historical engineering measures data, and historical monitoring data along the highway; A causal graph containing geological state nodes and engineering measure nodes is constructed based on geomechanical causal relationships; a first road segment group with implemented engineering measures and a second road segment group without implemented engineering measures are identified from the historical engineering measure data; and the intensity of the causal effect from the engineering measure node to the geological state node in the causal graph is estimated based on the historical monitoring data of the two groups. For the target road segment to be decided, candidate engineering measures are used as intervention inputs for the engineering measure nodes in the causal graph. The intervention effect is propagated along the directed edges of the causal graph. The predicted risk change value is calculated based on the intensity of the causal effect and the current geological state data of the target road segment. Based on the predicted risk change value and cost data, a combination of engineering measures is searched with the goal of maximizing risk reduction and minimizing cost, and a Pareto optimal solution set is obtained. Receive monitoring data after implementing the selected scheme in the Pareto optimal solution set, and update the causal effect strength when the deviation between the actual risk change value and the predicted risk change value exceeds a preset deviation threshold.

2. The method according to claim 1, characterized in that, A causal graph containing geological state nodes and engineering measure nodes is constructed based on geomechanical causal relationships, including: Based on geomechanics theory, a set of geological state variables and a set of engineering measures variables are determined. Each variable in the set of geological state variables is mapped to a geological state node in the causal graph, and each variable in the set of engineering measures variables is mapped to an engineering measures node in the causal graph. Based on the mechanical action transmission path in geomechanics, the directed edge connection relationship between nodes is determined. When the variable corresponding to the first node directly affects the variable corresponding to the second node through the mechanical mechanism, a directed edge from the first node to the second node is established between the first node and the second node. The time-series observations of node variables are extracted from the historical monitoring data. The conditional independence metric between node pairs connected by directed edges in the causal graph is calculated. If the conditional independence metric does not meet the independence determination condition, the directed edge is retained. If the conditional independence metric meets the independence determination condition, the directed edge is deleted, resulting in the optimized causal graph structure.

3. The method according to claim 1, characterized in that, Identify the first road segment group with implemented engineering measures and the second road segment group without implemented engineering measures from the historical engineering measures data. Estimate the causal effect intensity from the engineering measures node to the geological state node in the causal diagram based on the historical monitoring data of the two groups, including: The implementation records of engineering measures are extracted from the historical engineering measures data. Based on the implementation records, the road sections are classified into a first road section group where engineering measures were implemented and a second road section group where engineering measures were not implemented. The first road section group is used as the treatment group and the second road section group is used as the control group. Extract the geological state observation sequence of the first road section group and the second road section group before and after the implementation time node of the engineering measures from the historical monitoring data, calculate the geological state change of the first road section group as the change of the treatment group, and calculate the geological state change of the second road section group as the change of the control group. Construct a set of confounding variables, and perform covariate matching on the first road segment group and the second road segment group based on the set of confounding variables to eliminate the selection bias that affects the selection of engineering measures, and screen out road segment pairs whose confounding variable values ​​are less than a preset variable threshold; The difference between the changes in the treatment group and the changes in the control group is calculated based on the selected road segments as the strength of the causal effect. The difference is used to separate the causal effect of engineering measures on the geological conditions through double difference analysis.

4. The method according to claim 3, characterized in that, Construct a set of confounding variables, and perform covariate matching on the first road segment group and the second road segment group based on the set of confounding variables, including: Extract confounding variables that influence the decision-making process for implementing engineering measures from the historical monitoring data and historical engineering measures data, including geological condition variables, geographical location variables, and traffic flow variables, and construct a set of confounding variables; For each confounding variable in the set of confounding variables, a distance metric is selected based on the data type of the variable. For continuous variables, the standardized Euclidean distance is calculated, and for categorical variables, the Hamming distance is calculated. The distance metrics of each confounding variable are weighted and fused to obtain the comprehensive matching distance between road segments. For each road segment in the first road segment group, the road segment with the smallest comprehensive matching distance is searched from the second road segment group as the matching road segment. When the minimum comprehensive matching distance exceeds the preset distance threshold, the road segment is marked as an invalid match and removed from subsequent analysis. Calculate the distribution differences of the first and second road segment groups after matching on each confusion variable. When the distribution difference is greater than the preset distribution threshold, adjust the weight of the confusion variable and re-execute the matching. Iterate until the distribution difference converges to obtain a matching road segment pair that meets the balance requirements.

5. The method according to claim 1, characterized in that, For the target road segment to be decided, candidate engineering measures are used as intervention inputs to the engineering measure nodes in the causal graph. The intervention effect is propagated along the directed edges of the causal graph. Based on the intensity of the causal effect and the current geological state data of the target road segment, the predicted risk change value is calculated, including: The current geological state data of the target road segment is mapped to the corresponding geological state nodes in the causal graph as the initial state of the nodes, and the candidate engineering measures are mapped to the intervention amount of the engineering measure nodes; Identify all propagation paths from engineering measure nodes to target geological state nodes. For each propagation path, multiply the causal effect intensities of each directed edge along the path to obtain the path causal propagation intensity. Construct an attenuation function based on the number of path nodes to perform nonlinear attenuation correction on the path causal propagation intensity, thus obtaining the path propagation coefficient. For an intermediate node receiving multiple upstream propagation paths, the topological features of the upstream propagation paths are extracted and a path interaction correction factor is calculated. Based on the path propagation coefficient, the path interaction correction factor, and the intervention amount, the state increment of the intermediate node is calculated through weighted fusion. The updated state value of the intermediate node is obtained by superimposing the state increment with the initial state of the node. The updated state value is used as the intervention input to propagate to downstream nodes, and the update is performed layer by layer until the target geological state node completes the state update. The difference between the updated state value and the initial state value of the target geological state node is calculated as the predicted risk change value.

6. The method according to claim 1, characterized in that, Based on the predicted risk changes and cost data, a combination of engineering measures is searched with the objective of maximizing risk reduction and minimizing cost, resulting in a Pareto optimal solution set, including: Obtain cost data and corresponding predicted risk changes for each candidate engineering measure; extract the common downstream impact nodes for each engineering measure node in the causal graph; and analyze the propagation path topology to the common downstream impact nodes. When two propagation paths pass through the same intermediate node before reaching a common downstream affected node, the synergistic enhancement coefficient is calculated; when the two propagation paths produce opposite state changes to the common downstream affected node, the adversarial suppression coefficient is calculated, and a measure interaction matrix is ​​constructed. For each combination of engineering measures, the interaction coefficient between any two measures in the combination is obtained by querying the interaction matrix. The predicted risk change value of each engineering measure is weighted and calculated with the corresponding interaction coefficient to obtain the combined risk reduction considering the interaction effect of the measures. The total cost of the combination is obtained by summing the cost data of each engineering measure. A dual-objective optimization framework is constructed, with the negative value of the combined risk reduction as the first objective function and the total combined cost as the second objective function. Candidate solutions are initialized and objective function values ​​are calculated. The dominance relationship of candidate solutions is compared to extract the non-dominated solution set. New candidate solutions are generated based on the distribution density of the non-dominated solution set in the objective function space and iteratively updated. The final non-dominated solution set is extracted as the Pareto optimal solution set.

7. The method according to claim 1, characterized in that, Receive monitoring data after implementing the selected scheme in the Pareto optimal solution set, and update the causal effect strength when the deviation between the actual risk change value and the predicted risk change value exceeds a preset deviation threshold, including: Receive monitoring data collected after implementing the selected scheme in the Pareto optimal solution set, extract geological state parameters from the monitoring data, and calculate the change in geological state parameters before and after implementation as the actual risk change value; Calculate the deviation between the actual risk change value and the predicted risk change value corresponding to the selected scheme. When the deviation exceeds a preset deviation threshold, extract the engineering measure nodes and affected geological state nodes related to the selected scheme in the causal graph, and identify the set of directed edges connecting the nodes. For each directed edge in the set of directed edges, the causal effect strength correction amount of the directed edge is calculated based on the deviation between the actual risk change value and the predicted risk change value. The causal effect strength correction amount is then superimposed with the current causal effect strength of the directed edge to obtain the updated causal effect strength, which is then stored in the causal graph.

8. A highway geological environment analysis system integrating GIS and intelligent algorithms, used to implement the method as described in any one of claims 1-7, characterized in that, include: The data acquisition unit is used to acquire geological condition data, historical engineering measures data, and historical monitoring data along the highway. Causal construction unit, used to construct a causal graph containing geological state nodes and engineering measure nodes based on geomechanical causal relationships; The effect estimation unit is used to identify the first road segment group that implemented engineering measures and the second road segment group that did not implement engineering measures from the historical engineering measures data, and to estimate the intensity of causal effect from the engineering measures node to the geological state node in the causal diagram based on the historical monitoring data of the two groups. The risk prediction unit is used to take candidate engineering measures as intervention inputs to the engineering measure nodes in the causal graph for the target road segment to be decided, propagate the intervention effect along the directed edges of the causal graph, and calculate the predicted risk change value based on the intensity of the causal effect and the current geological state data of the target road segment. The solution search unit is used to search for a combination of engineering measures based on the predicted risk change value and cost data, with the goal of maximizing risk reduction and minimizing cost, to obtain a Pareto optimal solution set; The feedback update unit is used to receive monitoring data after the implementation of the selected scheme in the Pareto optimal solution set, and update the causal effect strength when the deviation between the actual risk change value and the predicted risk change value exceeds a preset deviation threshold.

9. An electronic device, characterized in that, include: processor; Memory used to store processor-executable instructions; The processor is configured to invoke instructions stored in the memory to execute the method according to any one of claims 1 to 7.

10. A computer-readable storage medium having computer program instructions stored thereon, characterized in that, When the computer program instructions are executed by the processor, they implement the method described in any one of claims 1 to 7.

Images