Method for predicting stability of collapsible loess stratum compaction-pouring composite foundation

Abstract

The invention relates to the technical field of foundation stability prediction, and discloses a stability prediction method for a collapsible loess stratum compaction-pouring composite foundation. The method is implemented in a computer aided design environment, environment parameters and distribution characteristics are obtained, a stability evaluation framework is established, and relative coordinates of a key monitoring area are calibrated to form stability configuration. Core monitoring points and layout information in stability configuration are identified, a peripheral point displacement prediction network is constructed, and a stability correction strategy is formulated. The foundation deformation dynamic state is monitored in real time in combination with stability configuration and a correction strategy, an environment evolution law is analyzed based on geological exploration data, and a risk identification mode is set in combination with the foundation deformation dynamic state and the evolution law. And detecting the local variation condition of the foundation by adopting a risk identification mode, and drawing up a local disposal scheme according to the variation condition and the prediction network. And executing stability prediction analysis based on the risk identification mode, the local disposal scheme and the correction strategy, and outputting a stability prediction value.

Description

Technical Field

[0001] This invention relates to the field of foundation stability prediction technology, specifically a method for predicting the stability of a compacted-irrigated composite foundation in collapsible loess strata. Background Technology

[0002] Current stability assessments of collapsible loess foundations primarily employ static analysis methods based on traditional soil mechanics theory and limited monitoring data. Existing techniques typically rely on limited geological exploration data, extrapolating the distribution of geological parameters across the entire site through interpolation or empirical formulas. This approach struggles to accurately reflect the spatial heterogeneity and anisotropy of collapsible loess. Foundation stability assessments often utilize the safety factor method or limit equilibrium theory. These methods, based on simplification and homogeneity assumptions, cannot effectively simulate the complex stress-strain response of compacted-grouted composite foundations during construction and use. Monitoring system layouts often follow uniform grid principles or rely on engineer experience, lacking targeted identification of key strata variation areas, resulting in monitoring data that fails to comprehensively capture the actual deformation behavior of the foundation. Stability analysis results typically present a single safety factor, failing to reflect local stability differences and potential failure mechanisms within the foundation.

[0003] Traditional prediction methods fail to adequately consider the environmental sensitivity of collapsible loess, particularly the impact mechanisms of environmental factors such as water migration and wet-dry cycles on foundation stability. Most existing models treat the foundation as a homogeneous continuum, neglecting the spatial variability of composite foundation material properties caused by compaction-grouting processes. There is a disconnect between monitoring data analysis and stability assessment; monitoring data is used only to validate design assumptions rather than to revise prediction models in real time. Risk identification is largely based on threshold alarm mechanisms, lacking the ability to proactively predict foundation deformation trends. Local variation detection typically relies on manual experience and lacks systematic, automated identification methods.

[0004] Existing technologies need to address the bottlenecks across the entire process, from geological parameter identification and monitoring network optimization to dynamic stability prediction. Specifically, there is a need to establish refined prediction models capable of integrating multi-source geological data and reflecting the spatial variability of foundation materials, as well as dynamic risk assessment methods based on real-time monitoring data. Collapsible loess foundation engineering urgently requires a stability prediction method that comprehensively considers geological uncertainties, construction process influences, and environmental factors. This method should possess multi-level analytical capabilities, ranging from macroscopic stability assessment to precise identification of local risks. Summary of the Invention

[0005] The purpose of this invention is to provide a method for predicting the stability of compaction-grouting composite foundations in collapsible loess strata, so as to solve the problems mentioned in the background art.

[0006] To achieve the above objectives, this invention provides a method for predicting the stability of a compaction-grouting composite foundation in collapsible loess strata, the method comprising:

[0007] Implemented in a computer-aided design environment, the environmental parameters and distribution characteristics of collapsible loess strata are identified by acquiring geological exploration data;

[0008] Based on the environmental parameters and distribution characteristics, a stability assessment framework is established, in which the relative coordinates of key monitoring areas are calibrated to form a stability configuration.

[0009] Identify the core monitoring points and layout information in the stability configuration, construct a surrounding point displacement prediction network, and formulate a stability correction strategy based on the prediction network and layout information;

[0010] Real-time monitoring of foundation deformation dynamics by combining stability configuration and correction strategies;

[0011] Based on the analysis of environmental evolution patterns using geological exploration data, a risk identification model is established by combining the dynamics and evolution patterns of foundation deformation.

[0012] The aforementioned risk identification model is used to detect local variations in the foundation, and local treatment plans are formulated based on the variations and the prediction network.

[0013] Based on risk identification patterns, local mitigation plans, and correction strategies, stability prediction analysis is performed, and stability prediction values ​​are output.

[0014] Preferably, establishing a stability assessment framework based on the environmental parameters and distribution characteristics includes:

[0015] Stratigraphic sequence was extracted based on the geological exploration data;

[0016] Based on the stratigraphic sequence, the soil property data and stability model of the foundation are retrieved;

[0017] Based on the soil property data and the relative coordinates, the stability mode is dynamically optimized to obtain an optimized mode;

[0018] The soil property data and the optimization mode are integrated to form a stable configuration.

[0019] Preferably, the construction of the peripheral point displacement prediction network includes:

[0020] Core monitoring points are monitoring points within the critical monitoring area specified in the stability configuration, while non-core monitoring points are monitoring points outside this area but within the adjacent distance range of the core monitoring points.

[0021] Calculate the relative distances between adjacent points of the core monitoring point based on the layout information;

[0022] Identify non-core monitoring points in the relative distance between the adjacent points;

[0023] Analyze the motion behavior of the non-core monitoring points relative to the core monitoring points;

[0024] Extract the response features of the core monitoring points, and evaluate the impact of the response features on the displacement of the non-core monitoring points based on the motion behavior;

[0025] Based on the motion behavior and displacement influence, a displacement prediction network for surrounding points is constructed.

[0026] Preferably, the step of formulating a stability correction strategy based on the predicted network and layout information includes:

[0027] Continuously track the set of monitoring points on the foundation and identify deviation points in the set of monitoring points;

[0028] The degree of interference of the deviation point on the stability configuration is analyzed based on the prediction network.

[0029] The data transmission records between the foundation and the deviation point are retrieved, and the cause of the deviation is determined based on the data transmission records.

[0030] Based on the degree of interference and the causes of deviation, a compensation mechanism for the deviation point is established;

[0031] Identify the deformation trajectory of the foundation, and based on the deformation trajectory and compensation mechanism, plan the regression path of the deviation point;

[0032] By integrating layout information and regression paths, a stability correction strategy is set.

[0033] Preferably, the analysis of environmental evolution patterns based on geological exploration data includes:

[0034] The geological exploration data is classified by attributes to obtain classified data;

[0035] Key indicators are extracted from the categorized data, and environmental historical records are collected based on the key indicators.

[0036] Identify the hydrological and climate data in the historical records;

[0037] Analyze groundwater level fluctuations and seepage patterns based on the aforementioned hydrological data;

[0038] Assess environmental humidity changes based on the aforementioned climate data;

[0039] By combining groundwater level fluctuations, infiltration patterns, and humidity changes, the environmental evolution patterns can be analyzed.

[0040] Preferably, the risk identification mode based on the dynamics and evolution of foundation deformation includes:

[0041] Based on deformation dynamics, the geological structure surrounding the foundation is identified;

[0042] Analyze the interaction between the geological structure and the foundation;

[0043] Based on the laws of environmental evolution and interaction, potential risk factors of the foundation are detected, and the threat level of the potential risk factors is assessed. The threat level is determined by combining the probability of occurrence of the potential risk factors and the severity of the consequences.

[0044] Based on deformation dynamics and threat levels, configure risk response measures for the foundation;

[0045] A rule system is formed by combining threat levels and risk response measures, and a risk identification model is established.

[0046] Preferably, the step of using the risk identification mode to detect local variations in the foundation includes:

[0047] Historical foundation monitoring data was collected based on risk identification models;

[0048] Determine the normal deformation parameters of the foundation based on the historical monitoring data;

[0049] Identify the current deformation state of the foundation, and calculate the current deviation by combining the current deformation state with historical monitoring data;

[0050] Set an abnormal threshold based on the fluctuation range of normal deformation parameters;

[0051] Based on the anomaly threshold and the current deviation, detect the local variation of the foundation.

[0052] Preferably, the step of formulating a local treatment plan based on the mutation status and the prediction network includes:

[0053] Locate the abnormal areas of the foundation based on the variation;

[0054] Analyze the damage types in abnormal areas and set up emergency response layers for abnormal areas based on the damage types;

[0055] Based on the prediction network, the movement trajectories of adjacent points in abnormal regions are identified;

[0056] Design repair paths for abnormal areas based on the motion trajectories of adjacent points;

[0057] By integrating emergency response layers and repair pathways, a localized treatment plan is formulated.

[0058] Preferably, the stability prediction analysis based on risk identification patterns, local treatment plans, and correction strategies includes:

[0059] Assess the stability threats to the foundation based on risk identification models;

[0060] Triggering conditions for activating local response plans based on stability threats, and setting risk avoidance strategies for the foundation;

[0061] Based on the triggering conditions and risk avoidance strategies, the foundation is treated to obtain the treatment results;

[0062] Based on the treatment results and correction strategies, a stability prediction analysis is performed, and the predicted stability values ​​are output.

[0063] Preferably, the method further includes:

[0064] Integrate geological data management functions into the computer-aided design platform to store and update the geological exploration data;

[0065] The platform enables visualized adjustment of stability configurations and dynamic optimization of the prediction network based on real-time monitoring data.

[0066] The platform outputs a stability prediction report, including predicted values ​​and confidence intervals.

[0067] Compared with the prior art, the beneficial effects of the present invention are:

[0068] By identifying environmental parameters and distribution characteristics of collapsible loess strata using geological exploration data, a stability assessment framework was established, and the relative coordinates of key monitoring areas were determined. This monitoring network configuration method based on geological characteristics allows for targeted monitoring point placement tailored to the specific engineering properties of collapsible loess. A displacement prediction network for surrounding points was constructed, and the intrinsic link between local deformation and overall stability was established through displacement correlation analysis between monitoring points. A risk identification model was set by combining real-time monitoring data with geological evolution patterns, enabling the system to dynamically capture the development trend of foundation deformation. Based on the prediction network and layout information, a stability correction strategy was formulated, achieving a direct correlation between monitoring data and treatment measures.

[0069] Employing a risk identification model to detect local variations in the foundation allows for the timely detection of development trends in locally weak areas. Based on these variations and the prediction network, local remediation plans are formulated to implement targeted reinforcement measures. Integrating the risk identification model, local remediation plans, and correction strategies for stability prediction analysis forms a complete technical chain from monitoring to early warning to remediation. A dynamic prediction model based on real-time monitoring data accurately reflects the consolidation and deformation process of the foundation after treatment. The local variation identification mechanism can precisely locate potential slip surfaces or collapsible areas, providing a basis for local reinforcement. This prediction method, combining global and local approaches, significantly improves the accuracy and reliability of stability prediction for collapsible loess foundations. Attached Figure Description

[0070] Figure 1This is a schematic diagram illustrating the working principle of the stability prediction method for the compaction-grouting composite foundation of collapsible loess strata described in this invention.

[0071] Figure 2 A flowchart for establishing a stability assessment framework based on environmental parameters and distribution characteristics;

[0072] Figure 3 A flowchart for constructing a peripheral point displacement prediction network;

[0073] Figure 4 A comprehensive analysis diagram of the environmental evolution of collapsible loess foundations;

[0074] Figure 5 This is a monitoring map of local variations in the foundation. Detailed Implementation

[0075] 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.

[0076] Please see Figure 1 This invention provides a stability prediction method for compaction-grouting composite foundations in collapsible loess strata. The method includes: identifying environmental parameters and distribution characteristics of the collapsible loess strata by acquiring geological exploration data, and establishing a stability assessment framework based on these parameters and characteristics. Within the assessment framework, the relative coordinates of key monitoring areas are calibrated to form a stability configuration. Core monitoring points and their layout information within the stability configuration are identified, a displacement prediction network for surrounding points is constructed, and a stability correction strategy is formulated based on this prediction network and layout information. The deformation dynamics of the foundation are monitored in real time, combining the stability configuration and correction strategy. Simultaneously, environmental evolution patterns are analyzed based on geological exploration data, and a risk identification model is set based on the foundation deformation dynamics and evolution patterns. The risk identification model is used to detect local variations in the foundation, and local treatment plans are formulated based on the variations and the prediction network. Stability prediction analysis is performed based on the risk identification model, local treatment plans, and correction strategies, and the predicted stability value is output.

[0077] Example 1: See Figure 2In the stability prediction method for compaction-grouting composite foundations in collapsible loess strata, establishing a stability assessment framework based on environmental parameters and distribution characteristics is a core step. The construction of this framework begins with in-depth processing of geological exploration data, including borehole records, soil sample test results, and geophysical exploration information. Specifically, the stability assessment framework is established by extracting stratigraphic sequences from the geological exploration data, clarifying the interface depth, thickness, material composition, and spatial continuity of different soil layers, and clearly presenting the distribution of key strata such as collapsible loess layers and potentially weak interlayers. Based on this stratigraphic sequence, soil property data corresponding to the foundation, including indicators such as water content, void ratio, and compression modulus, as well as a stability model adapted to the collapsible loess foundation, are retrieved from the database. Combining the soil property data and the relative coordinates of key monitoring areas marked in the stability configuration, the retrieved stability model is dynamically optimized. Parameters such as the safety factor threshold and allowable deformation value in the model are iteratively adjusted to ensure it conforms to the local geological conditions of specific monitoring points. The optimized stability model is integrated with soil property data to clarify the safety thresholds, correlation rules, and early warning conditions for each monitoring point. Simultaneously, the interaction relationships and influence weights among the monitoring points are defined, forming a complete stability assessment framework adapted to actual geological conditions. In specific implementation, a stratigraphic sequence is extracted based on geological exploration data. This extraction is accomplished by analyzing the hierarchical information in the exploration data. For example, stratigraphic interpretation software is used to identify the interface depth and thickness of different soil layers, thereby generating a vertical profile from the surface to the bedrock. This profile records in detail the material, density, and spatial continuity of each soil layer. The stratigraphic sequence not only reflects the sedimentary history of the soil but also reveals the distribution of potential weak interlayers or collapsible loess layers, providing a foundation for subsequent analysis. In some embodiments, when extracting the stratigraphic sequence, digital modeling tools are used to interpolate discrete exploration point data into a continuous three-dimensional geological model. The three-dimensional geological model can visualize the spatial variability of the strata and distinguish soil units with different engineering properties through color coding, enabling engineers to intuitively grasp the overall stratigraphic framework.

[0078] Based on the stratigraphic sequence, the system retrieves soil property data and stability models for the foundation. The soil property data, derived from laboratory geotechnical tests, includes indicators such as water content, void ratio, compression modulus, cohesion, and internal friction angle. These indicators are obtained through standardized testing procedures and stored in a database for easy retrieval. The stability model is a mathematical model or set of empirical formulas describing the foundation's response mechanism under external loads or environmental changes. For example, it includes unsaturated soil mechanics models suitable for collapsible loess. These models are based on classical soil mechanics theory and the characteristics of collapsible loess, and include pre-loaded default parameter values ​​for common loess regions. In practice, the retrieval process is implemented through the database interface of the computer-aided design platform. The platform automatically matches the corresponding soil property data and stability models based on the stratigraphic sequence. The matching logic is based on the consistency of soil layer type and depth. For example, when a collapsible loess layer is identified in the sequence, the system prioritizes retrieval of a dedicated stability model for collapsible soil. After retrieval, the soil property data and stability models are loaded into memory, forming a preliminary analysis environment to prepare for dynamic optimization.

[0079] Based on soil property data and the relative coordinates of key monitoring areas, local geological condition differences are identified. Using the initial parameters of the stability model as a foundation, parameters are iteratively adjusted based on real-time monitoring data of the coordinate areas—for example, the load factor is reduced in areas with high collapsibility risk, while the safety factor is relaxed in areas with strong stability. The model output is continuously compared with the actual response, correcting deviations until the model accurately adapts to local conditions, ultimately resulting in an optimized model. The relative coordinates of key monitoring areas are pre-calibrated within the stability assessment framework. These coordinates define the spatial location of monitoring points, typically arranged in a grid to cover potential deformation areas of the foundation. The layout of monitoring points and sensor configuration in key monitoring areas must be tailored to the characteristics of collapsible loess foundations and potential deformation risks. For sensor types, displacement sensors, earth pressure sensors, and pore water pressure sensors are primarily used; some areas may also be equipped with humidity sensors. In terms of layout, in addition to uniformly distributing foundation monitoring points in a grid pattern, monitoring points are densely distributed in potential deformation areas. The grid spacing is determined based on the foundation size and risk level, with a higher monitoring point density in potential deformation areas than in ordinary areas. Simultaneously, non-core monitoring points are deployed around the core monitoring point at a preset radius, forming a local monitoring network centered on the core point to ensure the capture of displacement correlations between the core point and surrounding points. Regarding sampling frequency, data is collected at a fixed cycle during normal operation, with the cycle set according to the foundation stability. When monitoring data shows that foundation deformation is approaching the warning threshold, or when environmental conditions change significantly, the sampling frequency is automatically increased, and the data collection interval is shortened to capture foundation deformation dynamics more promptly.

[0080] The dynamic optimization process involves adjusting parameters in the stability model to reflect the local conditions at specific monitoring points. For example, iterative algorithms can be used to correct safety factor thresholds or allowable deformation values ​​within the model. The optimization goal is to make the model output more closely match actual monitoring data. In practice, dynamic optimization employs a feedback mechanism, using real-time monitoring data or simulation results as input to continuously fine-tune model parameters. For instance, when a monitoring point shows excessive displacement, the system automatically reduces the load factor in the stability model or increases the soil strength reduction factor. The optimized model not only inherits the theoretical framework of the original stability model but also incorporates local soil property data and spatial coordinate information, thereby improving prediction accuracy. Dynamic optimization is a continuous process that updates periodically with the input of new data, ensuring the model always adapts to the current foundation condition.

[0081] A stability configuration is formed by integrating soil property data and optimization models. This configuration is a structured dataset that defines the safety threshold, association rules, and early warning conditions for each monitoring point. The integration process is implemented through a data fusion algorithm, combining the static indicators of soil property data with the dynamic output of the optimization model. In practice, the integration operation is performed on a computer-aided design platform. The platform generates a configuration file, stored in XML or JSON format, containing monitoring point IDs, coordinates, soil parameters, optimization model parameters, and calculation logic. Core monitoring points are those within the critical monitoring area defined in the stability configuration, while non-core monitoring points are those outside this area but within the adjacent distance range of the core monitoring points. The stability configuration also clarifies the interaction relationships between monitoring points, for example, by using weight coefficients to express the influence of core monitoring points on surrounding points. This allows the configuration to not only assess the stability of individual points but also analyze the collaborative behavior of the overall system. In some embodiments, after forming the stability configuration, verification tests are performed, such as inputting historical monitoring data to check whether the configuration output matches the actual deformation, thereby ensuring the reliability of the configuration. Finally, the stability configuration, as the output of the stability assessment framework, provides a benchmark for subsequent monitoring and prediction.

[0082] In practice, when extracting stratigraphic sequences, if geological exploration data is missing or noisy, data interpolation or machine learning methods are used to fill in the gaps. For example, Kriging interpolation is used to estimate soil properties in unexplored areas, or neural networks are trained to predict stratigraphic interface depths. This improves the completeness and accuracy of the sequence. When using Kriging interpolation, complete data from known exploration points surrounding the missing area are first selected, and the spatial correlation of these points is analyzed to determine their influence weights, with higher weights for points closer to the missing area. Based on the known point data and weights, interpolation is used to calculate estimated soil properties in the missing area, ensuring consistency with the distribution patterns of surrounding data. When using machine learning to fill in the gaps, known point coordinates and depths are used as input features, and soil properties are used as output labels to train a regression model. The missing area features are input into the model to obtain estimated values, while simultaneously verifying whether the results are consistent with geological patterns.

[0083] The extraction of stratigraphic sequences considers not only vertical profiles but also lateral variability. Contour maps or 3D surfaces are generated to identify regional differences in soil properties. Contour maps specifically refer to soil interface contour maps, which visually present the horizontal distribution, undulation, and thickness variations of soil layers by marking the elevations of different soil layers (such as collapsible loess layers and underlying impermeable layers). For example, dense contour lines on the top surface of collapsible loess in a certain area indicate significant lateral variations in soil layer thickness. 3D surfaces specifically refer to soil property 3D surfaces, which correlate key soil properties such as water content, density, and collapsibility coefficient with spatial coordinates to construct a continuous 3D visualization model, clearly showing the lateral gradual or abrupt changes in attribute values. For example, areas with a large thickness of collapsible loess may require a denser deployment of monitoring points. When calling soil property data and stability models, if no completely matching data is available in the database, the system will activate a similarity search function to find case data with similar geological conditions as substitutes, or prompt the user to manually enter missing parameters, ensuring flexibility in data retrieval. It is understandable that accessing soil property data may involve a large amount of data exchange, so the platform will use compression transmission or caching technology to optimize performance and avoid processing delays.

[0084] When performing dynamic optimization based on soil property data and relative coordinates, optimization algorithms may include gradient descent or genetic algorithms. These algorithms can efficiently search the parameter space to find the optimal solution that minimizes the stability mode error. The optimization process will set termination conditions, such as stopping when the parameter change is less than a threshold or when the maximum number of iterations is reached. The parameter update formula for gradient descent is:

[0085]

[0086] in: For the first The parameter values ​​after the next iteration. For the first The parameter values ​​after the next iteration. The learning rate (step size) controls the magnitude of each parameter update. For loss function exist The gradient at a given point reflects the direction and rate of change of the loss function with respect to the parameters.

[0087] Crossover probability in genetic algorithms and mutation probability The adaptive adjustment formula (piecewise adaptive strategy) is as follows:

[0088] Phase 1 (Generations 1-0.4M): , , ,

[0089] Second stage (0.4M~0.8M generation): , , ,

[0090] Phase 3 (0.8M~M generation): , , ,

[0091] in: The maximum number of generations. For crossover probability, , These represent the upper and lower limits of the crossover probability for the corresponding stage. The mutation probability, , These represent the upper and lower limits of the mutation probability for the corresponding stage.

[0092] The formation of optimization patterns relies not only on mathematical optimization but also incorporates expert knowledge rules. For example, when soil property data shows high water content, the permeability coefficient in the pattern is automatically adjusted to reflect the increased risk of collapse, which increases the practicality of the optimization. When integrating soil property data and optimization patterns, the generation of stability configurations includes multiple version management, such as saving configuration snapshots at different time points, facilitating retrospective analysis or comparison of the effects of different optimization strategies. Optionally, stability configurations can be exported to standard engineering formats, compatible with third-party analysis software, expanding the applicability of the method. In some embodiments, the integration process also adds metadata, such as creation time, operator information, and modification history, enhancing the traceability and auditability of the configuration.

[0093] Example 2: See Figure 3In the stability prediction method for compaction-grouting composite foundations in collapsible loess strata, constructing a displacement prediction network for surrounding points and formulating a stability correction strategy based on the prediction network and layout information are core steps. Constructing the displacement prediction network for surrounding points begins with the layout information of the core monitoring points, which includes the coordinate set and density parameters of the monitoring points in three-dimensional space. In practice, the relative distances between adjacent points of the core monitoring point are calculated based on the layout information. This calculation process is implemented using spatial geometric algorithms. For example, for each core monitoring point, all other monitoring points are searched within a preset radius centered on its coordinates, and the Euclidean distances between these points and the core point are accurately calculated. The formula for calculating the Euclidean distance in three-dimensional space is:

[0094]

[0095] in: As the core monitoring point Other monitoring points The Euclidean distance between them As the core monitoring point The three-dimensional coordinates For other monitoring points The three-dimensional coordinates.

[0096] This preset radius is typically determined based on empirical values ​​of the foundation scale and soil properties. The calculated relative distances between adjacent points form a distance matrix, which is stored in computer memory for subsequent analysis. Non-core monitoring points are identified based on the relative distances between adjacent points. These non-core monitoring points are auxiliary monitoring points located within the influence range of core monitoring points but not marked as core. This identification is accomplished by traversing the distance matrix. The system filters out all monitoring points with distances less than or equal to a preset threshold and categorizes them as target non-core monitoring points. These points constitute the direct objects of displacement prediction.

[0097] Analyzing the motion behavior of non-core monitoring points relative to core monitoring points requires continuous input of time-series data from monitoring sensors, including displacement, settlement rate, and horizontal movement vector. In practice, the analysis process employs correlation analysis and regression models. For example, a regression model is used, determining the input and output variables: time-series data (displacement, settlement rate, and horizontal movement vector) from the core monitoring points as independent variables, and the corresponding displacement, settlement rate, and horizontal movement vector from the non-core monitoring points as dependent variables during the same period. This ensures consistency in the time dimension of the variables and eliminates the interference of time lag on the analysis results. The appropriate regression model type is selected based on the characteristics of the monitoring data. If the motion data of core and non-core monitoring points show a linear correlation, a multiple linear regression model is used, with the following expression:

[0098]

[0099] in: These are the predicted displacement values ​​for non-core monitoring points; For the regression constant term; Here, represents the regression coefficients, and represents the core monitoring points, respectively. The degree of influence of unit displacement on the displacement of non-core monitoring points; As the core monitoring point The actual displacement value.

[0100] If a nonlinear correlation exists between the two, a nonlinear regression model is selected. Historical monitoring data is divided into training and validation sets. The regression model is fitted using the training set data. For a multiple linear regression model, the model coefficients are determined iteratively using the least squares method. The formula for calculating the regression coefficients in the least squares method is:

[0101]

[0102] in: The regression coefficient vector ; To design the matrix, the first column is a vector of all 1s (corresponding to the constant term). The remaining columns contain displacement sample data for each core monitoring point; This represents the displacement sample vector of non-core monitoring points; For designing a matrix transpose; For designing a matrix The inverse matrix.

[0103] The above calculations clarify the influence of each motion parameter of the core monitoring point on the motion behavior of the non-core monitoring points. The model fitting effect is tested using validation set data. If the deviation between the model's predicted values ​​and the actual motion data of the non-core monitoring points is within a preset range, it indicates that the model can effectively characterize the motion relationship between the two. If the deviation exceeds the range, the model type is readjusted or more historical data is added for a second fitting until the model accurately reflects the motion pattern of the non-core monitoring points relative to the core monitoring points, providing a quantitative basis for subsequent evaluation of the impact of the core monitoring point's response characteristics on the displacement of the non-core monitoring points.

[0104] This coefficient quantifies the synchronicity and dependence of the two movements. The motion behavior is further decomposed into the consistency of displacement direction, the proportional relationship of displacement amplitude, and the phase difference of the motion occurrence. Response characteristics of the core monitoring points are extracted. Response characteristics refer to the specific deformation patterns exhibited by the core monitoring points under changes in external loads or environmental factors, such as rapid settlement, lateral extrusion, or rebound. Extraction methods include feature engineering on historical data of the core monitoring points to extract statistical features such as mean, variance, and peak value, as well as time-domain features such as trend and periodic terms. The impact of response characteristics on the displacement of non-core monitoring points is assessed based on the motion behavior. The assessment process establishes a mathematical model, such as using a transfer function or influence factor matrix to describe how much displacement change in non-core monitoring points is caused by a unit deformation of the core monitoring point. If a multiple linear regression model is used to describe the displacement impact, its expression is:

[0105] in: These are the predicted displacement values ​​for non-core monitoring points. For the regression constant term, The regression coefficients (i.e., influence factors) represent the core monitoring points. The degree of influence of unit displacement on the displacement of non-core monitoring points. As the core monitoring point The actual displacement value.

[0106] The magnitude of the displacement effect is determined by fitting the observed data using the least squares method. The formula for solving the regression coefficients in the least squares method is as follows:

[0107]

[0108] in: The regression coefficient vector To design the matrix, the first column is a vector of all 1s (corresponding to the constant term). The remaining columns contain displacement sample data for each core monitoring point. These are displacement sample vectors for non-core monitoring points. For designing a matrix transpose, For designing a matrix The inverse matrix.

[0109] Based on motion behavior and displacement influence, a peripheral point displacement prediction network is constructed. Essentially, the peripheral point displacement prediction network is a graph neural network or multiple linear regression model that takes core monitoring points as input and non-core monitoring point predicted displacements as outputs. The network topology is defined by the relative positions between monitoring points and the calculated displacement influence weights. The network is trained using historical monitoring data, and the network parameters are optimized through backpropagation algorithm, enabling the network to accurately predict the displacement of peripheral points based on real-time data from core monitoring points.

[0110] A stability correction strategy is formulated based on the prediction network and layout information. This strategy formulation is a dynamic decision-making process. The monitoring point set of the foundation, containing all sensor nodes deployed in the foundation, is continuously tracked. Readings from each monitoring point are acquired in real time through a data acquisition system and updated to a central database. Deviation points in the monitoring point set are identified based on the deviation between the real-time reading of the monitoring point and its normal value range defined in the stability configuration or the predicted value output by the prediction network. When the deviation exceeds a set tolerance threshold, the monitoring point is marked as a deviation point. The degree of interference of deviation points on the stability configuration is analyzed based on the prediction network. The analysis method involves inputting the abnormal data from the deviation points into the prediction network and simulating how this abnormal state will be transmitted through the network and affect the predicted displacement of other monitoring points. The degree of interference is quantified by calculating the number of affected monitoring points and the weighted sum of their displacement deviations. The data transmission records between the foundation and the deviation point are retrieved. These records contain historical readings from the deviation point sensors, communication status logs, and potential error records. Based on these records, the cause of the deviation is determined. The determination logic includes checking the continuity of the data sequence, the presence of jumps, and consistency with readings from other relevant sensors, thus distinguishing between genuine soil deformation, sensor malfunction, and data transmission interference. Combining the degree of interference and the cause of the deviation, a compensation mechanism is established for the deviation point. This mechanism may include algorithmic compensation and physical compensation. Algorithmic compensation might involve assigning lower weights to the deviation point data or replacing it with predicted values ​​during data fusion. Physical compensation might trigger temporary reinforcement measures in the area where the deviation point is located. The deformation trajectory of the foundation is identified. This trajectory is formed by connecting the displacement vectors of all monitoring points over a period of time, creating one or more paths. Based on the deformation trajectory and the compensation mechanism, a regression path for the deviation point is planned. This regression path is a theoretical path that gradually restores the monitored value of the deviation point to the normal range through a series of correction actions. The planning method may involve optimal control theory. By integrating layout information and regression paths, a stability correction strategy is set. The stability correction strategy is ultimately reflected in a complete set of action rules, which clearly stipulates under what conditions, for which monitoring point, and what correction action to take.

[0111] In practical implementation, when calculating the relative distance between adjacent points, if the monitoring points are not evenly distributed, an adaptive radius algorithm will be used. This algorithm dynamically adjusts the search radius based on the local point density to ensure that each core monitoring point has a reasonable number of adjacent points included in the analysis. The formula for calculating the adaptive radius is:

[0112]

[0113] in: For the first The adaptive search radius of each core monitoring point The initial reference radius (preset according to the foundation scale, such as 2~5 meters). This represents the expected average number of neighboring points for all core monitoring points (an empirical value, typically 5-8). For the first The core monitoring points are at the initial baseline radius. The number of neighboring points actually detected within the range. When hour, Expand the search radius to increase the number of neighboring points, when hour, This narrows the search radius to filter key neighboring points.

[0114] When analyzing motion behavior, in addition to displacement data, stress or pore water pressure data are introduced as auxiliary variables to more comprehensively understand the mechanical interaction between core and non-core monitoring points. If pore water pressure is introduced as an auxiliary variable, its correlation with displacement can be quantified using the Pearson correlation coefficient. The mechanical interaction between core and non-core monitoring points is manifested in stress transmission and the influence of pore water pressure. Stress changes in the core area are transmitted to the non-core area through soil particle contact forces, leading to stress redistribution in the non-core area and triggering compressive deformation or stress concentration. Changes in pore water pressure in the core area allow water to infiltrate into the non-core area, altering the soil moisture content and strength, thereby exacerbating or mitigating deformation. This cascading effect can be clearly identified by combining stress and pore water pressure data.

[0115] When constructing a network for predicting displacement of surrounding points, the complexity of the network needs to match the amount of actual data. If historical data is insufficient, a simple linear model will be used first, and then gradually upgraded to a nonlinear model as data accumulates. The core objective function of the support vector regression model is:

[0116]

[0117]

[0118]

[0119]

[0120] in: For the model weight vector, For bias terms, , Slack variables (allowing the model to exist) (prediction error within the range) The penalty coefficient (for the control model to exceed) (Penalty level for error samples) For insensitive loss parameters (preset allowable error range, such as 0.1~0.5mm), For the first Input features of each sample (such as displacement of core monitoring points and pore water pressure). For the first Output labels for each sample (displacement of non-core monitoring points). This represents the number of samples.

[0121] It is understandable that the accuracy of the prediction network is highly dependent on the accuracy of the layout information, so precise measurement and positioning are required when setting up monitoring points.

[0122] In some embodiments, diagnostic algorithms are employed to identify the causes of deviations. For example, by analyzing the temporal correlation between deviation data and neighboring data, isolated deviations are tended to be attributed to instrument malfunction, while regional deviations are tended to be attributed to actual soil deformation. When establishing a compensation mechanism, for deviations caused by sensor malfunction, the mechanism activates a backup sensor or switches to the predicted data stream; for actual soil deformation, the mechanism adjusts the parameters in the stability assessment model. When planning the regression path, engineering constraints such as construction feasibility, cost, and time factors are considered to ensure that the planned path is not only theoretically correct but also practically feasible. It can be understood that setting the stability correction strategy is an iterative process; the effect data generated after strategy execution is fed back to the system to optimize the parameters of the strategy itself.

[0123] Optionally, when constructing the displacement prediction network for surrounding points, a time delay effect can be introduced. This means considering the time required for the response characteristics of core monitoring points to propagate to non-core monitoring points; the network model will include a time delay parameter, which can be estimated through time series analysis. When formulating stability correction strategies, a risk assessment module can be introduced to conduct risk-benefit analyses of different correction schemes and select the scheme with the lowest risk for implementation.

[0124] Example 3: In the stability prediction method for compaction-grouting composite foundations in collapsible loess strata, analyzing environmental evolution patterns based on geological exploration data and further combining this with the dynamics and evolution of foundation deformation to set a risk identification model is a crucial step in ensuring long-term stability. Analyzing environmental evolution patterns based on geological exploration data begins with classifying the attributes of the data. Geological exploration data contains a large amount of raw observation values ​​and test results. Attribute classification is based on the physical meaning and engineering application of the data, such as classifying data into different attribute groups like basic soil layer information, hydrogeological parameters, and physical and mechanical indices. In specific implementation, the attribute classification process is achieved through the data management module of a computer-aided design platform. This module has a pre-set classification rule base. After reading the raw geological exploration data, the system automatically identifies the meaning of data fields and assigns them to the corresponding attribute categories, forming structured classified data. Key indicators are extracted from the classified data. These key indicators are parameters that are sensitive to environmental changes and have a significant impact on foundation stability, such as the collapsibility coefficient, saturated permeability coefficient, and capillary rise height. The extraction operation is based on the importance weights of indicators defined in the expert knowledge base. Environmental historical records are collected based on key indicators. These records are derived from regional meteorological stations, long-term groundwater monitoring wells, and historical engineering archives. Collection is accomplished through data interfaces accessing external databases or manual input, forming a time-series dataset covering several years or even decades. Hydrological and climate data are identified from the historical records. Hydrological data primarily includes groundwater level depth, water level fluctuations, and water mineralization. Climate data mainly includes annual average precipitation, evaporation, annual temperature range, and freeze-thaw cycles. Groundwater level fluctuations and infiltration patterns are analyzed based on the hydrological data. Time-series analysis is used to analyze groundwater level fluctuations, calculating the periodicity and trend components of the water level and identifying abnormally high or low water level events. Infiltration pattern analysis involves establishing a hydrogeological conceptual model to simulate the flow path and rate of groundwater in the soil and its impact on soil moisture content. Environmental humidity changes are assessed based on climate data. Assessing environmental humidity changes requires integrating precipitation and evaporation data to calculate a humidity index or drought index, and analyzing its long-term trends and its driving effect on changes in shallow soil moisture content. By combining groundwater level fluctuations, infiltration patterns, and humidity changes, the laws governing environmental evolution are analyzed. These laws are a comprehensive description of the mechanisms by which the external environment acts. The analysis process employs a combination of statistical induction and mechanistic models, ultimately leading to qualitative judgments and quantitative predictions of future environmental trends.

[0125] A risk identification model is established by combining the dynamics and evolution of foundation deformation. Foundation deformation dynamics are obtained in real-time through a monitoring network, including information on foundation displacement, settlement, and tilt. Based on deformation dynamics, the geological structures surrounding the foundation are identified, including known faults, fracture zones, ancient landslides, and weak interlayers. Identification is accomplished by overlaying and analyzing deformation dynamic cloud maps with digital maps of geological structures to pinpoint the spatial correlation between deformation anomalies and geological structures. The interaction between geological structures and the foundation is analyzed, focusing on how geological structures alter the stress and displacement fields of the foundation. For example, fault activity generates additional stress on the foundation, or weak interlayers become potential sliding surfaces. Potential risk factors of the foundation are detected based on environmental evolution patterns and interaction relationships. Potential risk factors refer to adverse phenomena that may be induced by the combined effects of the environment and geological structures, such as uneven foundation settlement, localized slippage, and loess collapse. The detection method employs logic tree or fault tree analysis, using environmental evolution patterns as triggering conditions and interaction relationships as amplification factors to systematically deduce possible risk scenarios. The threat level of potential risk factors is assessed, combining the likelihood of occurrence and the severity of consequences. The threat level is a comprehensive measure of the probability of a risk event occurring and the severity of its consequences, using a risk matrix method. The severity of consequences is quantified using a dimensionless severity index, which is mapped and assigned values ​​based on predefined level standards according to the potential consequences. The quantitative value of the threat level is calculated using the following formula:

[0126]

[0127] in: A dimensionless quantized value representing the threat level. The dimensionless probability representing the occurrence of potential risk factors. A dimensionless index representing the severity of consequences. Based on deformation dynamics and threat levels, risk response measures for the foundation are configured. These measures are a pre-set library of response plans for different threat levels; for example, for low-threat levels, simply increasing monitoring frequency might be sufficient, while for high-threat levels, early warning systems might be activated and engineering intervention prepared. A rule system is formed by linking threat levels and risk response measures, establishing risk identification models. These models are ultimately manifested as an automated decision-making rule base, defining a complete logical chain from data input and risk analysis to recommended measures.

[0128] In some embodiments, when classifying geological exploration data by attribute, if the data format is inconsistent, data cleaning and standardization are performed first to ensure that all data conforms to the preset format specifications before classification. When extracting key indicators, statistical methods such as principal component analysis are used to screen out the few indicators with the highest contribution from a large number of parameters to reduce data dimensionality and computational complexity. When collecting historical environmental records, special attention is paid to the reliability of data sources and the completeness of time series; missing or anomalous data segments are marked or imputed using appropriate methods. It is understandable that the accuracy of analyzing environmental evolution patterns directly depends on the length and quality of historical records; therefore, long-term, continuous, and reliable monitoring data is a crucial foundation.

[0129] In some embodiments, when analyzing the interaction between geological structures and the foundation, numerical simulation techniques, such as the finite element method, are employed to establish a refined model that includes the foundation and surrounding geological structures, simulating their mechanical responses under different loads and environmental conditions. When detecting potential risk factors, in addition to qualitative reasoning, machine learning algorithms, such as classification models, are introduced to identify new risk patterns by training on historical case data. When assessing threat levels, probability is used... The determination of the severity index can be achieved using a Bayesian update method, which continuously adjusts the prior probability as new monitoring data is acquired, thereby improving the accuracy of the assessment. Assigning values ​​requires collaboration across multiple disciplines, integrating judgments from geotechnical engineering, structural engineering, and other perspectives to establish a unified standard for risk classification. It's understandable that risk identification models are not static but rather a dynamic system that needs regular review and updates to adapt to new understandings and data changes.

[0130] See Figure 4 This paper presents a comprehensive analysis of the environmental evolution of collapsible loess foundations. The charts clearly show the trends of groundwater depth and collapsibility coefficient over time using a dual-axis system, reflecting the long-term impact of environmental factors on foundation stability. The solid blue line represents the temporal variation of groundwater depth, showing the fluctuation characteristics of water level in different seasons and years. The dashed red line indicates a clear correlation between the collapsibility coefficient and water level fluctuations, reflecting the direct influence of hydrogeological conditions on the collapsibility characteristics of loess. The shaded areas in the charts mark the normal range of water level fluctuations, helping to identify abnormal fluctuations. The marked key change points indicate important turning points in the environment; these nodes are often associated with extreme climate events or major engineering activities. By analyzing this environmental evolution pattern, the response behavior of the foundation under future environmental changes can be predicted, providing a scientific basis for risk identification and the formulation of prevention and control measures. The accumulation of long-term environmental monitoring data is crucial for accurately interpreting the evolution pattern.

[0131] Example 4: In the stability prediction method for compaction-grouting composite foundations in collapsible loess strata, the key stage for implementing precise intervention is to detect local variations in the foundation using a risk identification model and formulate local treatment plans based on these variations and the prediction network. Detecting local variations in the foundation using a risk identification model begins with collecting historical monitoring data of the foundation based on this model. The risk identification model is a pre-defined analytical framework that includes risk factors and early warning rules. In practice, historical monitoring data is collected by accessing the historical database in the data storage system. This historical monitoring data covers various parameters recorded under normal operating conditions, including but not limited to time-series data of physical quantities such as vertical displacement, horizontal displacement, earth pressure, and pore water pressure. The data collection process must ensure the continuity of timestamps and the integrity of data values. Based on the historical monitoring data, normal deformation parameters of the foundation are determined. These parameters characterize the deformation features of the foundation when it is in a stable state. The determination method involves statistical analysis of the historical monitoring data, calculating the average value, standard deviation, and normal fluctuation range of each monitoring parameter over long-term observation, thereby establishing a set of benchmark indicators for the health status of the foundation. The current deformation state of the foundation is identified through a sensor network deployed on the foundation in real time. This real-time data stream is continuously input into the processing system and compared with historical benchmarks. The current deviation is calculated by combining the current deformation state and historical monitoring data. The current deviation is a crucial indicator quantifying the degree to which the current state deviates from historical normality. The calculation method involves comparing the real-time value with the historical benchmark value (e.g., the average value) for the same parameter at the same monitoring point. This difference can be expressed as an absolute difference or a relative percentage. An anomaly threshold is set based on the fluctuation range of the normal deformation parameter. This dynamic threshold determines whether a deviation constitutes an anomaly. The anomaly threshold is typically determined by multiplying the standard deviation of historical data by a safety factor. For example, a range of plus or minus three times the standard deviation of the average value is set as the normal threshold; values ​​exceeding this range are considered anomalies. Based on the anomaly threshold and the current deviation, local variations in the foundation are detected. This detection process involves systematically scanning the current deviation of all monitoring points and comparing it with the corresponding anomaly threshold. If the deviation of one or more monitoring points consistently exceeds the anomaly threshold, a local variation is identified in that area. The type and location of the variation are recorded and marked.

[0132] Based on the variation status and prediction network, a local treatment plan is formulated, which is a comprehensive decision-making process. The abnormal areas of the foundation are located according to the variation status. The location operation involves mapping the coordinates of monitoring points showing variation onto the digital model of the foundation using spatial analysis technology. The planar extent and depth of the affected area are delineated using contour lines or spatial interpolation methods. The damage type of the abnormal area is analyzed. The damage type needs to be determined based on the specific manifestation of the variation. For example, variation dominated by vertical settlement may indicate compressive deformation, while variation dominated by lateral displacement may indicate shear deformation. The analysis process combines geological survey data and mechanical models for comprehensive diagnosis. An emergency treatment layer is set up for the abnormal area based on the damage type. The emergency treatment layer is a series of pre-designed rapid response measures for different damage types. For example, for softening deformation caused by seepage, the emergency treatment layer may include immediately lowering the surrounding water level or providing temporary covering; for plastic deformation caused by excessive load, the emergency treatment layer may include unloading or applying counterpressure. Based on the predictive network, the motion trajectories of adjacent points in the abnormal area are identified. The predictive network is a model describing the displacement transfer relationship between monitoring points. By inputting deformation data of the abnormal area, the predictive network can simulate and calculate the future motion trends and trajectories of surrounding adjacent points. A repair path for the abnormal area is designed based on the motion trajectories of adjacent points. The repair path refers to the technical route guiding the abnormal area and its affected areas back to a stable state. Designing the repair path requires considering the direction and rate of the motion trajectory and formulating step-by-step reinforcement or adjustment measures, such as installing anti-slide piles or grouting reinforcement along the displacement expansion direction. Integrating the emergency treatment layer and the repair path, a local treatment plan is formulated. The local treatment plan is a complete action plan that clarifies the priority of emergency treatment, the specific steps of the repair project, the required materials and equipment, and the timing and cycle of implementation.

[0133] In practical implementation, referring to Table 1, when collecting historical monitoring data, if the data volume is huge, data downsampling technology will be used to reduce the data volume while maintaining the data trend characteristics, thereby improving processing efficiency. When determining normal deformation parameters, seasonal fluctuations or periodic changes of the parameters will be considered, for example, differentiating between different benchmark values ​​for dry and rainy seasons to avoid misjudgment. When calculating the current deviation, the system will use the sliding window averaging method to process real-time data to eliminate the influence of instantaneous interference noise and obtain a more realistic deformation trend. The setting of abnormal thresholds is not static; the system will periodically review and adjust the setting of abnormal thresholds based on the increase in the foundation's operating years and changes in the external environment to make them more consistent with the current state of the foundation.

[0134] Table 1: Analysis of Deviation at Monitoring Points

[0135]

[0136] It is understandable that the detection accuracy of local variations directly depends on the density of the monitoring network and the accuracy of the sensors. Therefore, deploying a sufficient density of monitoring points in key areas is a prerequisite for the effective implementation of this method. When analyzing damage types, it is often necessary to combine on-site investigation and in-situ testing results for verification to improve diagnostic accuracy. When designing remediation paths, a technical and economic comparison of multiple options is required to select the optimal solution. When detecting local variations, the system can automatically trigger different levels of early warning signals, such as a yellow warning to indicate attention and a red warning to require immediate action. When formulating local treatment plans, an expert system can be introduced to incorporate the treatment experience knowledge base of experts in the field, assisting in the generation of more reasonable solutions.

[0137] See Figure 5 This chart showcases the monitoring and identification results of local ground variability. By comparing the historical average displacement with the current actual displacement of each monitoring point, and combining this with preset anomaly thresholds, key areas with potential variability risks are systematically identified. The chart uses a dual-bar graph to clearly present the displacement status of each monitoring point. The light blue bars represent the historical average displacement baseline value obtained from long-term observations, while the light red bars display the current real-time monitored displacement value. The anomaly threshold range indicated by the red dashed line constitutes an important basis for judging the variability status. Monitoring points specifically marked as "abnormal" indicate that their current displacement has exceeded the normal fluctuation range. These points may experience stability problems due to changes in local geological conditions, uneven load distribution, or environmental factors. The statistical information box in the upper right corner of the chart provides a quantitative assessment of the overall monitoring status. Based on this local variability identification result, engineering technicians can accurately locate areas requiring key attention and intervention, and combine this with predictive network analysis of the movement trajectories of adjacent points to provide data support for formulating precise local intervention plans, realizing a shift from passive monitoring to proactive prevention and control.

[0138] Example 5: In the stability prediction method for compaction-grouting composite foundations in collapsible loess strata, the final execution stage of the entire method is to perform stability prediction analysis and output results based on risk identification models, local treatment plans, and correction strategies. This stage integrates data management, visualization adjustment, and report output functions on a computer-aided design platform. The stability prediction analysis based on risk identification models, local treatment plans, and correction strategies begins with judging the stability threat to the foundation according to the risk identification model. The risk identification model is a pre-established automated analysis module containing a risk element library and judgment logic. In a specific example, the risk identification model, through continuous analysis of settlement and tilt data at monitoring points MP-103 and MP-105, identified that the settlement rate in this area had exceeded the warning threshold of 5 mm per day for 12 consecutive hours, and the tilt angle change rate exceeded the allowable range. Based on the prediction of recent concentrated rainfall leading to a rise in groundwater levels in the environmental evolution law, the risk identification model determined that there was a significant stability threat in the southeast quadrant of the foundation caused by increased collapsibility, and the threat level was rated as "high". The system activates local response plans based on the triggering conditions for stability threats. These triggering conditions are pre-defined activation rules within the local response plans. For example, when the threat level in a specific area reaches "high" and the deformation trend continues to worsen, the system automatically activates the corresponding response plan. In this example, for the stability threat in the southeast quadrant, the system retrieves the contingency plan database and activates the local response plan numbered "RP-2024-05". This plan includes explicit instructions for emergency pressure grouting and setting up temporary drainage facilities in the abnormal area. The system also sets up foundation risk mitigation strategies, which are auxiliary measures taken to prevent further deterioration or ensure safety. In this example, the automatically generated risk mitigation strategies include: temporarily restricting heavy vehicle traffic above the area, increasing the data collection frequency of surrounding monitoring points to once per minute, and sending early warning notifications to relevant management personnel.

[0139] Based on the triggering conditions and risk avoidance strategies, the foundation is treated. This treatment process translates the local treatment plan and risk avoidance strategies into actual action, resulting in a treatment outcome. Continuing the example above, the system issues the activated local treatment plan "RP-2024-05" to the intelligent construction management subsystem. The subsystem dispatches grouting equipment and personnel to the designated coordinates in the southeast quadrant, executing emergency grouting according to the grouting pressure, grout ratio, and grouting volume specified in the plan. Simultaneously, the drainage team deploys lightweight well points for forced drainage according to instructions. The risk avoidance strategy is implemented by the monitoring center, restricting area access by setting up roadblocks and adjusting the sampling frequency of the automated monitoring system. The treatment results are fed back through monitoring sensors. For example, within 24 hours after grouting, the settlement rate at monitoring point MP-103 drops to 1.5 mm per day, and groundwater level monitoring data shows a decrease in water level. These data collectively constitute the result of this treatment operation. Based on the treatment results and correction strategies, a stability prediction analysis is completed. Stability prediction analysis is a comprehensive evaluation process. The correction strategy provides rules for adjusting the prediction model after treatment. In this example, the system substitutes post-treatment monitoring data (such as reduced settlement rate and tilt stability) into the stability assessment model. Simultaneously, it applies the correction rules from the correction strategy for short-term strengthening of soil parameters in the grouting-reinforced area, recalculating the overall safety factor of the foundation and the deformation trend over a future period. The system outputs a stability prediction value, a quantitative evaluation result. In this example, the system ultimately outputs a predicted safety factor of 1.25 for the foundation over the next 30 days and a predicted maximum additional settlement of 8 mm. This prediction value is attached to a detailed prediction report.

[0140] A geological data management function is integrated into the computer-aided design platform for storing and updating geological exploration data. This function is one of the platform's core modules. In implementation, this function provides a structured database with tables containing fields such as borehole number, coordinates, soil layer depth, geotechnical test parameters, and groundwater level. It supports batch data import, conditional queries, version control, and historical data tracing. When geological conditions change or new exploration data is obtained, authorized engineers can update the corresponding records in the database through the platform's data maintenance interface, ensuring the timeliness of the basic data. The platform also enables visual adjustments to stability configurations. This visual adjustment function allows users to interactively modify monitoring network parameters through a graphical interface and dynamically optimize the prediction network based on real-time monitoring data. In practice, the platform interface displays the foundation stability configuration in the form of a two-dimensional plan view or a three-dimensional model, such as the location of monitoring points and initial safety thresholds. Users can adjust the placement of monitoring points by dragging and dropping icons, or directly modify the safety threshold values ​​on the legend. The platform backend records these adjustments and updates the stability configuration. Simultaneously, the platform's data analysis engine continuously receives real-time monitoring data streams and uses this data to periodically retrain the surrounding point displacement prediction network using machine learning algorithms, adjusting the connection weights within the network to enable it to adapt to changes in the foundation state. The platform outputs a stability prediction report, the final presentation of the methodology's results. The report includes predicted values ​​and confidence intervals. In practice, the report generation module integrates all key data and analysis conclusions obtained during the stability prediction analysis phase, automatically generating a richly illustrated document. This document not only includes the final overall stability prediction value but also displays the curves of the predicted value over time in chart form, clearly providing the confidence intervals for the predicted values. The confidence intervals are calculated using statistical methods, taking into account the uncertainties of the input parameters and the model itself. The report supports export in PDF, Word, and other formats.

[0141] In some embodiments, when setting up risk mitigation strategies, the strategy library includes a tiered response mechanism to initiate risk mitigation measures of varying intensities for different levels of stability threats. For example, for "medium" level threats, only the monitoring frequency might be increased, while for "high" level threats, area isolation and personnel evacuation preparations might be implemented. When performing predictive analysis based on the handling results and correction strategies, if the handling results do not achieve the expected results, the system will activate the backup plan selection logic in the correction strategy, automatically recommending and potentially activating another, more robust local handling plan for iterative handling and prediction. It is understood that the stability and computational efficiency of the computer-aided design platform are fundamental to ensuring the smooth implementation of the entire method; therefore, the platform is typically deployed on high-performance servers and equipped with data backup and disaster recovery mechanisms.

[0142] Example 6: In a compaction-grouting composite foundation project in collapsible loess strata, geological exploration data of the area was retrieved in a computer-aided design environment to identify the distribution range, water content, and void ratio of the collapsible loess layer. Based on this, a stability assessment framework was established, and the central and peripheral areas of the foundation were designated as key monitoring areas with relative coordinates to form a stability configuration. From the stability configuration, 12 core monitoring points were identified within the key monitoring area, and 24 non-core monitoring points were identified within a 5-meter radius of the core points. After obtaining the layout information of each monitoring point, the adjacent distances between the core and non-core monitoring points were calculated. The settlement and horizontal movement patterns of non-core points relative to the core points were analyzed, and the settlement response characteristics of the core points under load were extracted. The influence of these characteristics on the displacement of non-core points was assessed, and a displacement prediction network for the surrounding points was constructed. A stability correction strategy was formulated based on the layout information—continuously tracking the data of all monitoring points. If the displacement of a monitoring point deviates from the output value of the prediction network, the data transmission record of that point is retrieved to determine the cause of the deviation. If it is a sensor malfunction, backup data is used; if it is soil deformation, a regression path is planned.

[0143] By combining stability configuration and correction strategies, data from various monitoring points are collected in real time using displacement sensors and pore water pressure sensors to obtain the dynamics of foundation deformation. Simultaneously, based on historical hydrological and climatic records from geological exploration data, the seasonal fluctuation patterns of groundwater levels and humidity trends in the area are analyzed to form environmental evolution patterns. By combining foundation deformation dynamics with environmental evolution patterns, the risk of subsidence due to rising groundwater levels at the foundation edges is identified. The threat level of this risk is assessed, and measures to increase the monitoring frequency in the edge areas are configured, establishing a risk identification model.

[0144] Historical foundation monitoring data for the past six months was collected using a risk identification model to determine normal deformation parameters such as normal settlement rate and horizontal displacement range, and anomaly thresholds were set. Real-time monitoring revealed that the settlement rate of three non-core monitoring points in the edge area exceeded the anomaly threshold, indicating that this area was a local variation zone. After locating the coordinates of the anomaly zone, its damage type was analyzed as collapsible compressive deformation. Based on the damage type, a temporary drainage emergency treatment layer was set up. The settlement trajectory of adjacent points in the anomaly zone was identified using a displacement prediction network, and a grouting reinforcement repair path was designed along the trajectory direction. A local treatment plan was formulated by integrating the emergency treatment layer and the repair path.

[0145] Stability prediction analysis was performed based on risk identification models, local treatment plans, and correction strategies. This involved assessing the impact of edge area subsidence risk on overall foundation stability, activating the trigger conditions for local treatment plans, executing temporary drainage and grouting operations, and obtaining post-treatment monitoring data—the settlement rate in the abnormal area returned to the normal range, and the groundwater level stabilized. Based on these treatment results and the stability correction strategy, a predicted value for the composite foundation stability was output, determining that the current overall foundation stability meets the requirements.

[0146] On the computer-aided design platform, the monitoring data and geological parameters of the abnormal area after this incident were updated to the geological data management module. The layout of monitoring points in the edge area was adjusted through the platform's visual interface, adding two core monitoring points. The displacement prediction network parameters were dynamically optimized based on the new monitoring data. Simultaneously, combining monitoring data from different periods over the past three months, a spatiotemporal evolution curve of foundation stability was plotted—showing a pattern of slightly decreased stability in the edge area during the rainy season and an overall increase in stability during the dry season, thus completing the stability prediction of the loess composite foundation. Subsequent real-time monitoring data collection will continue. If a deviation in displacement is detected again at a monitoring point, the above risk identification, handling, and prediction process will be repeated, dynamically updating the stability assessment data and prediction model to achieve continuous control over the stability of the composite foundation.

[0147] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0148] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for predicting the stability of a compaction-grouting composite foundation in collapsible loess strata, characterized in that, The method includes: Implemented in a computer-aided design environment, the environmental parameters and distribution characteristics of collapsible loess strata are identified by acquiring geological exploration data; Based on the environmental parameters and distribution characteristics, a stability assessment framework is established, in which the relative coordinates of key monitoring areas are calibrated to form a stability configuration. Identify the core monitoring points and layout information in the stability configuration, construct a surrounding point displacement prediction network, and formulate a stability correction strategy based on the prediction network and layout information; Real-time monitoring of foundation deformation dynamics by combining stability configuration and correction strategies; Based on the analysis of environmental evolution patterns using geological exploration data, a risk identification model is established by combining the dynamics and evolution patterns of foundation deformation. The risk identification model is used to detect local variations in the foundation, and a local treatment plan is formulated based on the variations and the prediction network. Based on risk identification patterns, local mitigation plans, and correction strategies, stability prediction analysis is performed, and stability prediction values ​​are output.

2. The stability prediction method for a compaction-grouting composite foundation in collapsible loess strata as described in claim 1, characterized in that, The establishment of a stability assessment framework based on the environmental parameters and distribution characteristics includes: Stratigraphic sequence was extracted based on the geological exploration data; Based on the stratigraphic sequence, the soil property data and stability model of the foundation are retrieved; Based on the soil property data and the relative coordinates, the stability mode is dynamically optimized to obtain an optimized mode; The soil property data and the optimization mode are integrated to form a stability configuration.

3. The stability prediction method for a compaction-grouting composite foundation in collapsible loess strata as described in claim 1, characterized in that, The construction of the surrounding point displacement prediction network includes: Core monitoring points are monitoring points within the critical monitoring area specified in the stability configuration, while non-core monitoring points are monitoring points outside this area but within the adjacent distance range of the core monitoring points. Calculate the relative distances between adjacent points of the core monitoring point based on the layout information; Identify non-core monitoring points in the relative distance between the adjacent points; Analyze the motion behavior of the non-core monitoring points relative to the core monitoring points; Extract the response features of the core monitoring points, and evaluate the impact of the response features on the displacement of the non-core monitoring points based on the motion behavior; Based on the motion behavior and displacement influence, a displacement prediction network for surrounding points is constructed.

4. The stability prediction method for a compaction-grouting composite foundation in collapsible loess strata as described in claim 1, characterized in that, The step of formulating a stability correction strategy based on the predicted network and layout information includes: Continuously track the set of monitoring points on the foundation and identify deviation points in the set of monitoring points; The degree of interference of the deviation point on the stability configuration is analyzed based on the prediction network. The data transmission records between the foundation and the deviation point are retrieved, and the cause of the deviation is determined based on the data transmission records. Based on the degree of interference and the causes of deviation, a compensation mechanism for the deviation point is established; Identify the deformation trajectory of the foundation, and based on the deformation trajectory and compensation mechanism, plan the regression path of the deviation point; By integrating layout information and regression paths, a stability correction strategy is set.

5. The stability prediction method for a compaction-grouting composite foundation in collapsible loess strata as described in claim 1, characterized in that, The environmental evolution patterns analyzed based on geological exploration data include: The geological exploration data is classified by attributes to obtain classified data; Key indicators are extracted from the categorized data, and environmental historical records are collected based on the key indicators. Identify the hydrological and climate data in the historical records; Analyze groundwater level fluctuations and seepage patterns based on the aforementioned hydrological data; Assess environmental humidity changes based on the aforementioned climate data; By combining groundwater level fluctuations, infiltration patterns, and humidity changes, the environmental evolution patterns can be analyzed.

6. The stability prediction method for a compaction-grouting composite foundation in collapsible loess strata as described in claim 1, characterized in that, The risk identification model, which combines the dynamics and evolution of foundation deformation, includes: Based on deformation dynamics, the geological structure surrounding the foundation is identified; Analyze the interaction between the geological structure and the foundation; Based on the laws of environmental evolution and interaction, potential risk factors of the foundation are detected, and the threat level of the potential risk factors is assessed. The threat level is determined by combining the probability of occurrence of the potential risk factors and the severity of the consequences. Based on deformation dynamics and threat levels, configure risk response measures for the foundation; A rule system is formed by combining threat levels and risk response measures, and a risk identification model is established.

7. The stability prediction method for a compaction-grouting composite foundation in collapsible loess strata as described in claim 1, characterized in that, The method of using the risk identification mode to detect local variations in the foundation includes: Historical foundation monitoring data was collected based on risk identification models; Determine the normal deformation parameters of the foundation based on the historical monitoring data; Identify the current deformation state of the foundation, and calculate the current deviation by combining the current deformation state with historical monitoring data; Set an abnormal threshold based on the fluctuation range of normal deformation parameters; Based on the anomaly threshold and the current deviation, detect the local variation of the foundation.

8. The stability prediction method for a compaction-grouting composite foundation in collapsible loess strata as described in claim 1, characterized in that, The local treatment plan formulated based on the mutation status and prediction network includes: Locate the abnormal areas of the foundation based on the variation; Analyze the damage types in abnormal areas and set up emergency response layers for abnormal areas based on the damage types; Based on the prediction network, the movement trajectories of adjacent points in abnormal regions are identified; Design repair paths for abnormal areas based on the motion trajectories of adjacent points; By integrating emergency response layers and repair pathways, a localized treatment plan is formulated.

9. The stability prediction method for a compaction-grouting composite foundation in collapsible loess strata as described in claim 1, characterized in that, The stability prediction analysis based on risk identification patterns, local treatment plans, and correction strategies includes: Assess the stability threats to the foundation based on risk identification models; Triggering conditions for activating local response plans based on stability threats, and setting risk avoidance strategies for the foundation; Based on the triggering conditions and risk avoidance strategies, the foundation is treated to obtain the treatment results; Based on the treatment results and correction strategies, a stability prediction analysis is performed, and the predicted stability values ​​are output.

10. The stability prediction method for a compaction-grouting composite foundation in collapsible loess strata as described in claim 1, characterized in that, The method further includes: Integrate geological data management functions into the computer-aided design platform to store and update the geological exploration data; The platform enables visualized adjustment of stability configurations and dynamic optimization of the prediction network based on real-time monitoring data. The platform outputs a stability prediction report, including predicted values ​​and confidence intervals.

Images