A method and system for safety assessment of subway foundation pit construction

Abstract

This invention provides a method and system for safety assessment of construction of foundation pits adjacent to subway lines, belonging to the field of construction technology for foundation pits adjacent to subway lines. The technical solution includes the following steps: S1, collecting multi-source monitoring data and subway operation load data to construct a foundation pit-subway digital twin model; S2, calculating the cumulative fatigue damage degree based on the foundation pit-subway digital twin model; S3, calculating the construction disturbance entropy index; S4, outputting the real-time occurrence probability of various engineering failure modes; S5, generating graded early warning information and corresponding handling suggestions. The beneficial effects of this invention are: constructing a consistent virtual-real foundation pit-subway digital twin model by assimilating multi-source monitoring, construction disturbance, and subway load data; extracting the dynamic response of key components according to train events and calculating the cumulative fatigue damage degree; fusing and generating the construction disturbance entropy index; driving a dynamic Bayesian network to output the real-time probability of failure modes; and generating graded early warning and handling suggestions in a coordinated manner, achieving full-cycle risk quantitative assessment.

Description

Technical Field

[0001] This invention relates to the field of construction technology for foundation pits near subway lines, and in particular to a method and system for safety assessment of construction of foundation pits near subway lines. Background Technology

[0002] In recent years, the integrated development of urban rail transit and underground space has accelerated, leading to a significant increase in the number of subway-adjacent foundation pit projects. Retaining structure systems, diaphragm walls and support / anchoring technologies, and dewatering and waterproofing methods have become increasingly sophisticated. Engineering monitoring methods have evolved from traditional measurement to automation and networking, with multi-source monitoring of displacement, stress, pore water pressure, and vibration acceleration gradually forming a comprehensive online monitoring system. Simultaneously, BIM and numerical simulation are widely used in foundation pit design and construction organization. Risk control has shifted from experience-based assessment to model-driven assessment, and some projects have begun to introduce data assimilation, digital twins, and intelligent early warning concepts to achieve dynamic characterization of the construction process and assessment of risk trends.

[0003] However, existing safety assessments for subway foundation pit construction still have systemic shortcomings.

[0004] Firstly, most methods rely on static calculations and threshold warnings before or in stages of construction, making it difficult to incorporate real-time changes in construction disturbance parameters, subway operating loads, and structure-soil response into a unified framework. This results in assessment results that are not sensitive to sudden changes in operating conditions and stage transitions.

[0005] Secondly, monitoring data and numerical models are often disconnected: monitoring focuses on alarms based on "phenomenal quantity," while models focus on verification of "design conditions." There is a lack of closed-loop updates that achieve consistency between virtual and real data through spatiotemporal alignment and dynamic data assimilation, making it difficult to correct deviations when model predictions deviate from the actual situation.

[0006] Third, the cyclic vibration load caused by subway trains is event-driven and cumulative. Existing assessments focus on the instantaneous exceedance of displacement or stress, making it difficult to establish a quantitative chain of "dynamic response - stress cycle - damage accumulation" at the level of key components, and making it difficult to characterize the long-term constraints of material degradation on risk.

[0007] Fourth, the risk output under multi-factor uncertainty is mostly a deterministic conclusion or a single indicator level. There is a lack of a mechanism for collaborative reasoning and joint decision-making on the construction disturbance entropy, cumulative fatigue damage degree and failure mode probability. It is difficult to balance the suppression of false alarms and the prevention of false alarms, and it is also difficult to form an executable and traceable graded disposal recommendation. Summary of the Invention

[0008] The purpose of this invention is to provide a method and system for safety assessment of subway foundation pit construction by constructing a virtual-real consistent digital twin model of the foundation pit and subway through multi-source monitoring, assimilation of construction disturbance and subway load data, extraction of dynamic response of key components according to train events and calculation of cumulative fatigue damage, fusion to generate construction disturbance entropy index, driving dynamic Bayesian network to output real-time probability of failure mode, and linkage to generate graded early warning and disposal suggestions, thereby realizing full-cycle risk quantitative assessment.

[0009] This invention is achieved through the following measures: A method for safety assessment of construction near subway foundation pits, characterized by the following steps: S1. Collect construction disturbance parameters, multi-source monitoring data of the foundation pit and adjacent subway structures, and subway operation load data. Construct and update a foundation pit-subway digital twin model consistent with the physical state using spatiotemporal alignment and dynamic data assimilation methods. S2. Extract the dynamic response of key components based on the foundation pit-subway digital twin model and calculate the cumulative fatigue damage degree of key components under cyclic loading. S3. Integrate the cumulative fatigue damage degree and the construction disturbance parameters to calculate a construction disturbance entropy index that comprehensively reflects system risk. S4. Input the construction disturbance entropy index and the cumulative fatigue damage degree as observation nodes into a dynamic Bayesian network, update network parameters online, and output the real-time occurrence probability of various engineering failure modes. S5. Perform multi-dimensional collaborative judgment based on the real-time occurrence probability, the cumulative fatigue damage degree, and the construction disturbance entropy index to generate graded early warning information and corresponding handling suggestions.

[0010] The invention also has the following specific features: The collection of construction disturbance parameters, multi-source monitoring data of the foundation pit and adjacent subway structures, and subway operation load data specifically includes: simultaneous collection and preprocessing of the three types of data. The first category is construction disturbance parameters, which continuously acquire real-time engineering parameters reflecting the intensity of excavation, support, and dewatering activities by connecting the automated control system for foundation pit construction with monitoring instruments. The second category is multi-source monitoring data of the foundation pit and adjacent subway structure. It involves deploying sensor arrays at key parts of the engineering structure and collecting physical quantities that reflect the mechanical state of the structure and the response of the soil layer according to a preset sampling strategy. These physical quantities include, but are not limited to, deformation, stress, pore water pressure and vibration acceleration. The vibration data is then sliced ​​into time slices according to the train passing events. The third category is subway operation load data, which is obtained through data interfaces or dedicated monitoring equipment to acquire train operation status parameters in the affected sections; All collected raw data streams are subject to uniform time synchronization and spatial coordinate association, and undergo standardized preprocessing including outlier identification, noise filtering, and confidence assessment to finally generate a multi-source fusion dataset that is time-synchronized, coordinate-associated, and includes quality / confidence labels.

[0011] The construction and updating of the foundation pit-subway digital twin model, consistent with the physical state, through spatiotemporal alignment and dynamic data assimilation methods specifically includes: First, a spatiotemporal alignment operation is performed on the multi-source fusion dataset. The spatiotemporal alignment operation is to uniformly transform the spatial coordinates of all monitoring data to an engineering coordinate system or a preset reference coordinate system based on the subway structure axis, and to synchronously calibrate the timestamps of all data. Then, based on this aligned dataset, a parameterized foundation pit-subway coupled mechanical model is iteratively updated using a dynamic data assimilation method. The dynamic data assimilation method uses key soil parameters and structural parameters in the coupled mechanical model as state variables to be corrected, and spatiotemporally aligned monitoring data as observation constraints. A sequential assimilation algorithm is used to compare the observation data with the model prediction values ​​in a rolling manner, and the state variables and model states are adjusted in reverse according to the comparison results so that the prediction residuals meet the preset consistency criteria. By iteratively executing the above assimilation process, the output of the coupled mechanical model is made consistent with the real-time monitoring under the consistency criterion, thereby obtaining and continuously updating the foundation pit-subway digital twin model that is consistent with the physical state.

[0012] The extraction of the dynamic response of key components based on the foundation pit-subway digital twin model includes: under the condition that the foundation pit-subway digital twin model satisfies the preset consistency criterion, and based on the preset engineering safety concerns, defining a set of key components from the engineering structure represented by the foundation pit-subway digital twin model. Simultaneously, the impact time window of the train passing through the adjacent subway tunnel is analyzed and generated in real time from the subway operation load data, and a correspondence is established between the impact time window and the train passing event; Then, the foundation pit-subway digital twin model is driven to perform transient dynamic analysis for each of the impact time windows, and the complete dynamic time history data of each component in the key component set within the corresponding impact time window is located and extracted from the analysis results. Finally, a dynamic response dataset is generated and output to characterize the dynamic behavior of each key component. The dynamic response dataset is indexed by the key component identifier and the sequence of train passing events, and serves as the direct input for calculating the cumulative fatigue damage.

[0013] The calculation of the cumulative fatigue damage of the key components under cyclic loading includes: taking the dynamic response dataset output in the previous step as input, for each key component in the dataset; First, the relevant physical quantities in its dynamic time history data are converted into the stress time history of the key component, and the stress time history is then correlated with the train passing event and the construction stage. Then, the stress time history is processed using a cycle counting algorithm to identify and count the independent stress cycles contained therein, and the amplitude and mean of each stress cycle are recorded. Next, the fatigue performance parameters of the materials corresponding to the key components, which are pre-stored and associated with the key component identifier, are queried, and each identified stress cycle is converted into a damage increment according to the stress-life relationship and the mean stress correction rule. Finally, all the damage increments are accumulated in chronological order to output the cumulative fatigue damage of the key component from the start of construction to the current calculation time.

[0014] The calculation of the construction disturbance entropy index, which comprehensively reflects the system risk, by integrating the cumulative fatigue damage degree and the construction disturbance parameters includes: normalizing and weighting the construction disturbance parameters and the cumulative fatigue damage degree to form a multidimensional disturbance state feature vector. A probability distribution vector characterizing the perturbation intensity distribution is constructed based on each component in the perturbation state feature vector. The basic perturbation entropy value is obtained by calculating the probability distribution vector using the information entropy function; The basic disturbance entropy value is smoothed in the time dimension and weighted and aggregated in the spatial dimension to generate a comprehensive construction disturbance entropy index. The dynamic risk level is determined based on the historical statistical distribution of the construction disturbance entropy index, and the construction disturbance entropy index and the dynamic risk level are output.

[0015] The process of using the construction disturbance entropy index and the cumulative fatigue damage degree as observation nodes to input into a dynamic Bayesian network, updating network parameters online, and outputting the real-time occurrence probability of various engineering failure modes includes: Construct a dynamic Bayesian network comprising observation layer nodes, intermediate mechanism layer nodes, and top-level failure layer nodes, wherein the observation layer nodes include at least the construction disturbance entropy index and the cumulative fatigue damage degree; The construction disturbance entropy index and the cumulative fatigue damage value output in real time from the previous steps are used as evidence of the observation layer node at the current moment and input into the dynamic Bayesian network. Based on the input evidence, the posterior probability distribution of all nodes in the dynamic Bayesian network is updated online using a probabilistic inference algorithm. The update process includes both the correction of the state probability of the intermediate mechanism layer nodes and the recalculation of the occurrence probability of the top-level failure layer nodes. The top-level failure layer nodes correspond to various predefined engineering failure modes. After updating and calculation, the real-time occurrence probability of each type of engineering failure mode under the current evidence is output.

[0016] The step of generating graded early warning information and corresponding handling suggestions by performing multi-dimensional collaborative judgment based on the real-time occurrence probability, the cumulative fatigue damage degree, and the construction disturbance entropy index includes: The system continuously receives and integrates the real-time occurrence probability, the cumulative fatigue damage degree, and the construction disturbance entropy index. Based on the preset multi-dimensional linkage early warning rules, it performs collaborative comparison and logical judgment on the values ​​or levels of the above indicators and the corresponding thresholds to determine the current comprehensive early warning level and identify the dominant risk factor. According to the comprehensive early warning level and the dominant risk factor, it matches the corresponding disposal suggestions from the preset disposal measures library. Finally, a graded early warning information containing the comprehensive early warning level and the proposed action measures is generated and output.

[0017] A system employing the aforementioned safety assessment method for construction of subway foundation pits, characterized in that it comprises: The system comprises the following modules: a multi-source acquisition and assimilation module, a multi-source monitoring data module for the foundation pit and adjacent subway structures, and subway operation load data, which collects construction disturbance parameters, multi-source monitoring data, and subway operation load data. It constructs and updates a foundation pit-subway digital twin model consistent with the physical state using spatiotemporal alignment and dynamic data assimilation methods. A dynamic response extraction module extracts the dynamic response of key components based on the foundation pit-subway digital twin model and calculates the cumulative fatigue damage of key components under cyclic loading. A disturbance entropy index calculation module integrates the cumulative fatigue damage and construction disturbance parameters to calculate a construction disturbance entropy index that comprehensively reflects system risk. A Bayesian probabilistic reasoning module uses the construction disturbance entropy index and the cumulative fatigue damage as observation nodes input to a dynamic Bayesian network, updates network parameters online, and outputs the real-time occurrence probability of various engineering failure modes. A coordinated early warning decision-making module performs multi-dimensional collaborative judgment based on the real-time occurrence probability, the cumulative fatigue damage, and the construction disturbance entropy index to generate tiered early warning information and corresponding handling suggestions.

[0018] The beneficial effects of this invention are as follows: a virtual-real consistent digital twin model of the foundation pit and subway is constructed by assimilating multi-source monitoring, construction disturbance and subway load data; the dynamic response of key components is extracted according to train events and the cumulative fatigue damage is calculated; the construction disturbance entropy index is generated by fusion; the dynamic Bayesian network is driven to output the real-time probability of failure mode; and graded early warning and disposal suggestions are generated in linkage to achieve full-cycle risk quantitative assessment. Attached Figure Description

[0019] Figure 1 The flowchart illustrates the overall process of the safety assessment method for construction of subway foundation pits provided in this embodiment of the invention. Detailed Implementation

[0020] To clearly illustrate the technical features of this solution, the following detailed implementation method will be used to explain the solution.

[0021] Example 1 See Figure 1 A method for safety assessment of construction near subway foundation pits, characterized by the following steps: S1. The collection of construction disturbance parameters, multi-source monitoring data of the foundation pit and adjacent subway structures, and subway operation load data specifically includes: simultaneous collection and preprocessing of the three types of data. The first category is construction disturbance parameters, which continuously acquire real-time engineering parameters reflecting the intensity of excavation, support, and dewatering activities by connecting the automated control system for foundation pit construction with monitoring instruments. The second category is multi-source monitoring data of the foundation pit and adjacent subway structure. It involves deploying sensor arrays at key parts of the engineering structure and collecting physical quantities that reflect the mechanical state of the structure and the response of the soil layer according to a preset sampling strategy. These physical quantities include, but are not limited to, deformation, stress, pore water pressure and vibration acceleration. The vibration data is then sliced ​​into time slices according to the train passing events. The third category is subway operation load data, which is obtained through data interfaces or dedicated monitoring equipment to acquire train operation status parameters in the affected sections; All collected raw data streams are subject to uniform time synchronization and spatial coordinate association, and undergo standardized preprocessing including outlier identification, noise filtering, and confidence assessment to finally generate a multi-source fusion dataset that is time-synchronized, coordinate-associated, and includes quality / confidence labels.

[0022] The construction and updating of the foundation pit-subway digital twin model, consistent with the physical state, through spatiotemporal alignment and dynamic data assimilation methods specifically includes: First, a spatiotemporal alignment operation is performed on the multi-source fusion dataset. The spatiotemporal alignment operation is to uniformly transform the spatial coordinates of all monitoring data to an engineering coordinate system or a preset reference coordinate system based on the subway structure axis, and to synchronously calibrate the timestamps of all data. Then, based on this aligned dataset, a parameterized foundation pit-subway coupled mechanical model is iteratively updated using a dynamic data assimilation method. The dynamic data assimilation method uses key soil parameters and structural parameters in the coupled mechanical model as state variables to be corrected, and spatiotemporally aligned monitoring data as observation constraints. A sequential assimilation algorithm is used to compare the observation data with the model prediction values ​​in a rolling manner, and the state variables and model states are adjusted in reverse according to the comparison results so that the prediction residuals meet the preset consistency criteria. By iteratively executing the above assimilation process, the output of the coupled mechanical model is made consistent with the real-time monitoring under the consistency criterion, thereby obtaining and continuously updating the foundation pit-subway digital twin model that is consistent with the physical state.

[0023] Step S1 specifically includes: In this embodiment, step S1 is used to form a continuously updated "foundation pit-subway digital twin model" during the construction of the subway foundation pit. This digital twin model is not a static modeling result, but rather a model that is generated under the combined effects of construction disturbance and train operation load. Through the synchronous acquisition of multi-source data, standardized preprocessing, spatiotemporal alignment and dynamic data assimilation, the model output and monitoring observation are kept consistent under a preset consistency criterion. This provides a reliable model state and data benchmark for the subsequent extraction of dynamic response of key components and calculation of fatigue damage.

[0024] First, simultaneous acquisition and preprocessing of three types of data are performed: The first type of data is construction disturbance parameters: In this embodiment, by connecting the foundation pit construction automation control system and monitoring instruments, real-time engineering parameters reflecting the intensity of excavation, support, and dewatering activities are continuously read and recorded. To ensure that construction disturbance parameters can establish a traceable mapping relationship with structural response, this embodiment archives construction disturbance parameters according to construction stages and solidifies the construction stage switching time in the parameter stream, so that the subsequent assimilation process can establish a consistent time index relationship among the three "stage-parameter-response"; The second type of data consists of multi-source monitoring data of the foundation pit and adjacent subway structures. In this embodiment, sensor arrays are deployed at key parts of the engineering structure, and physical quantities such as deformation, stress, pore water pressure, and vibration acceleration are collected synchronously according to a preset sampling strategy. Among them, the vibration acceleration data is sliced ​​according to the time slice of train passing events. This "train passing event time slice" is a key step of the present invention in the context of the subway: since the vibration during construction and the vibration during train operation may overlap in time and partially intersect in frequency band, if the full-time vibration sequence is directly statistically analyzed or assimilated, it is easy to form observation bias caused by non-stationary superposition, which makes the model's response to train load "diluted" by construction disturbance or "amplified" by abnormal peaks. By generating slices with train passing events as boundaries, this embodiment can constrain the dynamic impact of each train passing within a comparable time window, so that the assimilation algorithm can repeatedly correct the model parameters for similar events, avoiding misjudging short-term strong excitation as long-term trend drift.

[0025] The third type of data is subway operation load data: In this embodiment, train operation status parameters are obtained through data interfaces or dedicated monitoring equipment, and these parameters are associated with train passing events as the input source of external boundary conditions for dynamic analysis and assimilation.

[0026] Secondly, a unified time synchronization marker and spatial coordinate association are applied to all collected raw data streams, and standardized preprocessing is performed. This embodiment uniformly performs outlier identification, noise filtering, and confidence assessment on three types of data: outlier identification is used to eliminate non-physical abrupt changes caused by sensor saturation, momentary chain breaks, or construction mishaps; noise filtering is used to suppress the contribution of high-frequency random noise to the assimilation residuals; confidence assessment is used to generate a quality / confidence label for each data stream and output it synchronously with the data. This "quality / confidence label" has a specific purpose in this invention: the core of dynamic data assimilation lies in correcting the model state using observation constraints. If the observation quality participates in assimilation indiscriminately, a small amount of low-quality data can introduce systematic errors during iteration, causing model parameter divergence or false convergence. By outputting the quality / confidence label, this embodiment provides an executable observation weight basis for the subsequent assimilation process, enabling the assimilation update to remain stable and controllable even in engineering sites with uneven data quality. This is an important metrological foundation for the "strong disturbance, high noise, and strong non-stationary" working conditions of subway construction. After completing the above preprocessing, a multi-source fusion dataset with time synchronization, coordinate association, and quality / confidence labels is generated.

[0027] Then, a spatiotemporal alignment operation is performed on the multi-source fusion dataset. In this embodiment, the spatial coordinates of all monitoring data are uniformly transformed to an engineering coordinate system or a preset reference coordinate system based on the subway structure axis, and all data timestamps are synchronously calibrated. The purpose of unifying the spatial coordinates is not formal processing, but to provide a single coordinate reference for the geometric mapping of the "foundation pit-subway coupled mechanical model", so that a verifiable consistency relationship is formed between the monitoring points, model mesh nodes, and subway structure axis. This step can directly solve the common problem in the prior art of "point coordinates coming from different measurement references, resulting in the position offset of the observation constraint application": in the near-subway scenario, the offset of the observation point will cause the assimilation to mistakenly apply local deformation to the wrong component or the wrong soil layer unit, thus causing a systematic deviation in the direction of model parameter correction. By uniformly transforming to an engineering coordinate system based on the subway structure axis, this embodiment uses the subway structure as a global reference, improves the repeatability of the relative positional relationship between the foundation pit and the subway, and makes the subsequent assimilation update have a clear spatial orientation.

[0028] Furthermore, based on the spatiotemporally aligned dataset, the parameterized foundation pit-subway coupled mechanical model is iteratively updated using a dynamic data assimilation method. In this embodiment, key soil and structural parameters in the coupled mechanical model are used as state variables to be corrected, and spatiotemporally aligned monitoring data are used as observation constraints. A sequential assimilation algorithm is employed to continuously compare the observed data with the model's predicted values, and the state variables and model states are adjusted in reverse based on the comparison results to ensure the predicted residuals meet a preset consistency criterion. To make the "consistency criterion" feasible, this embodiment can use a weighted residual consistency index to quantify the difference between the model output and the observations, and use a threshold as a trigger condition for accepting or continuing iterations, for example:

[0029] in, As a consistency indicator; For observation channel index; This represents the total number of observation channels; For the first The weights of each observation channel, wherein the weights are determined by a quality / confidence identifier; For the coupled mechanical model in the first... The prediction vector for the location and time corresponding to the observation channel; For the multi-source fusion dataset, the first... The observation vector of the observation channel; The threshold for the preset consistency criterion is set.

[0030] In the aforementioned assimilation process, the "sequential assimilation algorithm rolling update" and the "quality / confidence label weighting" together constitute the key innovation of step S1 of this invention: sequential assimilation can continuously absorb new observations as construction progresses, avoiding the model lag caused by traditional offline calibration only updating at a few stage points; the weighting mechanism enables assimilation to maintain the stability of parameter updates even when facing sensor drift, local missing measurements, and abnormal impacts, thereby reducing the probability of model miscalibration under strong disturbances near the subway. By iteratively executing the assimilation process, the output of the coupled mechanics model is kept consistent with real-time monitoring under the consistency criterion, ultimately obtaining and continuously updating a pit-subway digital twin model consistent with the physical state. The purpose of this digital twin model is twofold: on the one hand, it provides the key parameter states and the overall response field that evolve over time, serving as a reliable computational basis for extracting the dynamic response of key components in subsequent steps; on the other hand, through the repeated calibration mechanism formed by train passing event slices, the model can maintain the consistency of dynamic response under the action of train cyclic loads, thereby establishing a stable data and model benchmark for subsequent fatigue analysis related to cyclic loads.

[0031] In summary, step S1 of this embodiment, through a combined process of "multi-source synchronous acquisition and preprocessing - train passing event slicing - spatiotemporal alignment - confidence-weighted sequential assimilation - consistency criterion constraint", addresses technical issues such as data heterogeneity, time asynchrony, inconsistent spatial benchmarks, and model drift caused by strong noise interference in the construction of subway foundation pits. It provides an executable digital twin construction and update process and provides an input foundation with a unified benchmark, traceable index, and confidence constraints for subsequent steps.

[0032] S2. The extraction of the dynamic response of key components based on the foundation pit-subway digital twin model includes: under the condition that the foundation pit-subway digital twin model satisfies the preset consistency criterion, and based on the preset engineering safety concerns, defining a set of key components from the engineering structure represented by the foundation pit-subway digital twin model. Simultaneously, the impact time window of the train passing through the adjacent subway tunnel is analyzed and generated in real time from the subway operation load data, and a correspondence is established between the impact time window and the train passing event; Then, the foundation pit-subway digital twin model is driven to perform transient dynamic analysis for each of the impact time windows, and the complete dynamic time history data of each component in the key component set within the corresponding impact time window is located and extracted from the analysis results. Finally, a dynamic response dataset is generated and output to characterize the dynamic behavior of each key component. The dynamic response dataset is indexed by the key component identifier and the sequence of train passing events, and serves as the direct input for calculating the cumulative fatigue damage.

[0033] The calculation of the cumulative fatigue damage of the key components under cyclic loading includes: taking the dynamic response dataset output in the previous step as input, for each key component in the dataset; First, the relevant physical quantities in its dynamic time history data are converted into the stress time history of the key component, and the stress time history is then correlated with the train passing event and the construction stage. Then, the stress time history is processed using a cycle counting algorithm to identify and count the independent stress cycles contained therein, and the amplitude and mean of each stress cycle are recorded. Next, the fatigue performance parameters of the materials corresponding to the key components, which are pre-stored and associated with the key component identifier, are queried, and each identified stress cycle is converted into a damage increment according to the stress-life relationship and the mean stress correction rule. Finally, all the damage increments are accumulated in chronological order to output the cumulative fatigue damage of the key component from the start of construction to the current calculation time.

[0034] Step S2 specifically includes: In this embodiment, step S2 is used to extract the dynamic response of key components and calculate the cumulative fatigue damage of key components under cyclic loading based on the digital twin model of the foundation pit-subway that has been built and continuously updated in step S1. The core of step S2 lies in two aspects: First, the train passage event is analyzed into an impact time window, driving the digital twin model to output the dynamic time history of key components at a repeatable and comparable dynamic event scale; second, this dynamic time history is further converted into a stress time history for which fatigue assessment can be performed, completing cycle counting and damage accumulation, thereby forming a cumulative fatigue damage degree that is bound to the key component identifier and can be updated over time. It should be noted that the "adjacent subway tunnel" mentioned in this embodiment is one implementation of the "adjacent subway structure," and the method is also applicable to scenarios where the adjacent subway structure is underground or non-underground and is affected by operational loads. First, under the condition that the foundation pit-subway digital twin model satisfies the preset consistency criterion, a set of key components is defined from the engineering structure represented by the foundation pit-subway digital twin model based on preset engineering safety concerns. In this embodiment, the engineering safety concerns are used to constrain the scope and granularity of the set of key components. The objects of concern are usually components that have controllable significance for the stability of the foundation pit and the safety of the subway structure and are sensitive to dynamic response, such as key stress sections of the retaining structure, internal support nodes, connectors, and sensitive parts of the adjacent subway structure. The purpose of this step is to converge the calculation objects of subsequent dynamic analysis and fatigue analysis from the entire field to a manageable set of components, so that dynamic response extraction and damage calculation can establish a stable data index system with "key component identifier" as the primary key. Compared with the common method of response evaluation based only on monitoring points, this invention defines components as objects in the digital twin model, which can ensure that the response results are consistent with the component topology, material diameter, and boundary conditions, providing a structured support for subsequent fatigue performance parameter queries and damage accumulation.

[0035] Simultaneously, the impact time window of the train passing through the adjacent subway tunnel is analyzed and generated in real time from the subway operation load data, and a correspondence is established between the impact time window and the train passing event. This embodiment follows the organization of vibration data according to the time slice of train passing events in step S1, and further uses the train passing events for time domain organization in dynamic analysis. The impact time window is determined based on the time information of entering, passing through and leaving the impact section reflected by the train operation status parameters in the subway operation load data, in order to cover the entire process of the train load having a significant dynamic impact on the subway structure-surrounding soil-foundation pit support system. The key purpose of this step is to transform the dynamic input in the adjacent subway working condition from a continuous random process into a discrete event sequence, so that the subsequent dynamic response extraction can achieve repeated comparison and consistency verification at the event level, thereby reducing the interference of construction disturbance and environmental noise on the identification of train cyclic load. Even though the cycle counting algorithm itself is a conventional method, when it is directly applied to the stress sequence in the whole time domain without windowing, it is often difficult to distinguish between the low-frequency trend caused by train event cycles and the drift during the construction phase. This invention, by influencing the correspondence between the time window and the train passing event, enables the subsequent cycle identification to obtain a stable statistical caliber at the event scale, thereby providing a traceable event index basis for the cumulative fatigue damage.

[0036] Then, the foundation pit-subway digital twin model is driven to perform transient dynamic analysis for each of the impact time windows, and the complete dynamic time history data of each component in the key component set within the corresponding impact time window is located and extracted from the analysis results. In this embodiment, the transient dynamic analysis is performed on the model state after assimilation and update in step S1. The model parameters have been corrected by observation constraints and meet the preset consistency criteria. Therefore, the dynamic calculation is based on the premise of a reliable model state, avoiding systematic bias caused by direct extrapolation using an uncalibrated model. For each impact time window, the transient dynamic analysis outputs dynamic time history data corresponding to the location of the key component. The dynamic time history data may include one or more of displacement, acceleration, strain, or stress, and each set of time history data is associated with the key component identifier and the train passing event number. Finally, a dynamic response dataset is generated and output to characterize the dynamic behavior of each key component. The dynamic response dataset is indexed by the key component identifier and the order of train passing events, serving as the direct input for calculating the cumulative fatigue damage. This allows subsequent damage calculations to be completed stepwise accumulation and rolling updates under a unified structure of "same key component - multiple train events - multiple construction stages".

[0037] Subsequently, using the dynamic response dataset output in the previous step as input, for each key component in the dataset, this embodiment converts the relevant physical quantities in the dynamic time history data into the stress time history of the key component, and establishes a correspondence between the stress time history and the train passage event and construction stage. The purpose of this conversion process is to uniformly map the dynamic response quantities into the stress characterization sequence required for fatigue assessment, so that subsequent cycle counting and life calculation have clear physical meaning. To ensure the feasibility of the conversion process, this embodiment uses the linear elastic stress recovery relationship to represent the stress time history of the key component:

[0038] in, As a key component At any moment The stress time history vector; As a key component The transformation matrix related to the material and cross section; As a key component At any moment The strain time history vector; Identify key components; It is a time variable.

[0039] Then, a cycle counting algorithm is used to process the stress time history, identify and statistically analyze the independent stress cycles contained therein, and record the amplitude and mean of each stress cycle. In this embodiment, cycle counting uses the impact time window corresponding to the train passing event as the basic statistical unit, so that the stress cycle formed by each train passing can be consistently identified under the same window. Afterwards, the fatigue performance parameters of the materials corresponding to the key components, which are pre-stored and associated with the key component identifier, are queried, and each identified stress cycle is converted into a damage increment according to the stress-life relationship and the mean stress correction rule. To keep the damage increment calculation simple and feasible, this embodiment uses the following form to represent the single-cycle damage increment:

[0040] in, As a key component In the Damage increment corresponding to each stress cycle; The life mapping function is determined by the fatigue performance parameters; For the first The amplitude of each stress cycle; For the first The average of one stress cycle; Identify key components; This is the cyclic index.

[0041] Finally, all the damage increments are accumulated in chronological order to output the cumulative fatigue damage of the key component from the start of construction to the current calculation time. In this embodiment, the accumulation process uses the sequence of train passing events and the sequence of construction stages as time indices: train passing events are used to characterize the cyclic load sequence, and construction stages are used to characterize the impact of boundary condition changes caused by foundation pit excavation, support, and dewatering on the fatigue evolution path. The cumulative fatigue damage can be expressed as:

[0042] in, As a key component At the calculation time The cumulative fatigue damage; From the start of construction to the calculation time The total number of stress cycles identified cumulatively; For the first Damage increment per stress cycle; Identify key components; For calculating the time; This is the cyclic index.

[0043] In summary, step S2, through the process of "defining the set of key components - analyzing the time window of train passage impact - transient dynamic analysis - organizing the dynamic response dataset - stress time history conversion - cycle counting - life mapping - damage increment accumulation," transforms the reliable digital twin model and train passage event organization mechanism obtained in step S1 into an executable component-level fatigue damage quantification output. This output is strictly bound to the key component identifier and retains the index relationship between train passage events and construction stages, enabling a traceable engineering interpretation path for the accumulated fatigue damage. It can serve as a direct input for subsequent steps to integrate construction disturbance parameters and calculate the construction disturbance entropy index, and further provides a structured data foundation for the observation node evidence of the dynamic Bayesian network.

[0044] S3. The calculation of the construction disturbance entropy index, which comprehensively reflects the system risk, by fusing the cumulative fatigue damage degree and the construction disturbance parameters includes: normalizing and weighting the construction disturbance parameters and the cumulative fatigue damage degree to form a multidimensional disturbance state feature vector. A probability distribution vector characterizing the perturbation intensity distribution is constructed based on each component in the perturbation state feature vector. The basic perturbation entropy value is obtained by calculating the probability distribution vector using the information entropy function; The basic disturbance entropy value is smoothed in the time dimension and weighted and aggregated in the spatial dimension to generate a comprehensive construction disturbance entropy index. The dynamic risk level is determined based on the historical statistical distribution of the construction disturbance entropy index, and the construction disturbance entropy index and the dynamic risk level are output.

[0045] Step S3 specifically includes: In this embodiment, step S3 is used to fuse the construction disturbance parameters with the cumulative fatigue damage output in step S2 under a unified statistical caliber, forming a construction disturbance entropy index that can comprehensively characterize the evolution of system risk, and providing a dynamic risk level based on historical statistical distribution. The key technical problem addressed by step S3 is that during the construction of subway foundation pits, construction disturbance parameters are characterized by multi-source heterogeneity, inconsistent dimensions, and amplitudes spanning orders of magnitude. The cumulative fatigue damage reflects the internal stress degradation process of key components. If the two are directly linearly superimposed or compared using a fixed threshold, situations may arise where "a certain dimension dominates the result due to dimensional advantage" or "a staged disturbance peak masks the continuous accumulation of risk," making it difficult to form a stable measure of system risk. This invention adopts a workflow of "normalization processing and weighted fusion—probability distribution construction—information entropy calculation—spatiotemporal aggregation—historical distribution grading," so that macroscopic construction disturbances and microscopic damage states are characterized as changes in disturbance intensity distribution within a unified statistical structure, and the degree of distribution dispersion is characterized by entropy values, thereby forming a comparable index for risk states.

[0046] (i) Normalization and weighted fusion to form perturbation state feature vector At each calculation moment, the system obtains construction disturbance parameters from step S1 and cumulative fatigue damage degree bound to the key component identifier from step S2. To avoid fusion bias caused by dimensional differences, this embodiment first performs normalization processing on each component participating in the fusion. Assume the set of components to be fused consists of several construction disturbance parameter components and several cumulative fatigue damage degree components, and uses a unified symbol to represent the original value of each component at the current moment. The normalization process can be expressed as:

[0047] in, For the first The normalized values ​​of each component; For the first The original values ​​of each component; For the first The lower bound of each component within the preset reference window; For the first The upper bound of each component within the preset reference window; To prevent stable terms with a positive denominator of zero; For component indexes.

[0048] After normalization, this embodiment weights and fuses the construction disturbance parameters and cumulative fatigue damage to form a multidimensional disturbance state feature vector. The weighting is used to reflect the differences in risk contribution and data reliability between different components in engineering semantics. The weights can be mapped to the quality / confidence labels generated in step S1, suppressing low-confidence components during fusion and thus improving the robustness of the entropy index to noise and anomalies. The disturbance state feature vector can be expressed as:

[0049] in, The feature vector of the perturbation state; For the first The fusion weights of each component; The total number of components participating in the fusion This is a transpose operation; For the first The normalized values ​​of each component.

[0050] The purpose of the above-mentioned normalization and weighted fusion steps is to establish a state representation structure that is "dimensionless, weighted, and comparable", so that the cumulative fatigue damage degree is no longer used as an isolated value in subsequent calculations, but enters the feature vector as a component of the same scale as the construction disturbance parameters. This allows the feature vector to simultaneously contain two types of information: external disturbance intensity and internal damage evolution, providing a more complete state description basis for subsequent entropy calculations.

[0051] (ii) Construct the probability distribution vector and calculate the basic perturbation entropy value To transform multidimensional feature vectors into distributed objects that can be entropy-measured, this embodiment is based on Construct a probability distribution vector characterizing the intensity distribution of the disturbance. This step shifts the focus from absolute amplitude to changes in the relative proportions of each component, making it suitable for identifying situations where the risk source shifts from a single disturbance to a multi-factor coupling. In this embodiment, an exponential mapping is used to achieve nonnegativity and normalization, and the probability distribution vector can be expressed as:

[0052] in, The probability distribution vector is the first... One component; The perturbation state feature vector The One component; It is an exponential function; For summation index; The total number of components; For component indexes.

[0053] The information entropy function is then used to calculate the basic perturbation entropy value on the probability distribution vector. The basic perturbation entropy value reflects the dispersion of the perturbation contribution among the components: when a few components dominate the probability distribution, the entropy value is low; when multiple components contribute significantly simultaneously, the entropy value increases. The basic perturbation entropy value can be expressed as:

[0054] in, Based on the basic perturbation entropy value; These are the components of the probability distribution vector; It is the natural logarithm function; The total number of components; For component indexes.

[0055] (III) Temporal smoothing and spatial weighted aggregation to generate a comprehensive construction disturbance entropy index Because construction sites experience short-term impacts, localized noise, and occasional anomalies, using only instantaneous foundation disturbance entropy values ​​as risk characterization can easily lead to misjudgments due to momentary spikes or abrupt changes in indicators due to localized anomalies. Therefore, this embodiment performs time-dimensional smoothing on the foundation disturbance entropy values, making the indicators more focused on continuous evolution trends. Simultaneously, it weights and aggregates the entropy values ​​of different spatial units, making the indicators more consistent with the spatial characteristics of the influence range of the subway foundation pit distributed along the subway structure axis. Specifically, this embodiment divides the engineering impact zone into several spatial units. Within each spatial unit, the foundation disturbance entropy value is calculated based on the corresponding construction disturbance parameters and cumulative fatigue damage, followed by time smoothing.

[0056] in, For the first The smoothed perturbation entropy value of each time window; This represents the smoothed perturbation entropy value of the previous time window; For the first The basic perturbation entropy value for each time window; The smoothing coefficient is and satisfies ; Index for time windows.

[0057] After obtaining the smoothed disturbance entropy value of each spatial unit, this embodiment performs weighted aggregation based on the spatial influence weight to generate a comprehensive construction disturbance entropy index:

[0058] in, For the first The comprehensive construction disturbance entropy index for each time window; spatial unit The aggregate weights and satisfy ; For time windows Lower spatial unit The smoothed perturbation entropy value; This represents the total number of spatial units. For spatial unit indexing.

[0059] The combined use of time smoothing and spatial aggregation is as follows: time smoothing suppresses short-term spikes caused by a single train passing event or instantaneous construction operation, so that the indicator reflects the continuous accumulation of risks; spatial aggregation integrates information from multiple points, components and disturbance sources in the affected section of the subway into a single indicator under a unified weight structure, so as to avoid the mis-amplification of single-point anomalies at the global scale and maintain the risk sensitivity to the adjacent subway structure area.

[0060] This processing method matches the cumulative characteristics of accumulated fatigue damage in step S2, ensuring that short-term fluctuations in construction disturbance parameters do not mask the continuous contribution of damage accumulation to risk, thereby providing more stable observational evidence input for subsequent dynamic Bayesian networks.

[0061] (iv) Determine and output the dynamic risk level based on historical statistical distribution. In obtaining the comprehensive construction disturbance entropy index Subsequently, this embodiment determines the dynamic risk level based on the historical statistical distribution of this indicator. This approach specifically addresses the problem of the fixed threshold method being difficult to transfer across different projects, construction organizations, and subway operating densities: the same absolute threshold may be too conservative under high-frequency train conditions and too lenient under low-frequency conditions, leading to an imbalance in the early warning strategy. Dynamic grading based on historical statistical distribution uses the project's own data as a reference, ensuring that the risk level is adapted to the project baseline. This embodiment uses historical sample quantiles to determine the level threshold and applies the current... The data is compared with a threshold range to output the construction disturbance entropy index and the corresponding dynamic risk level. The dynamic risk level is then compared with... Maintain a one-to-one correspondence and continue to use it as observational evidence or a basis for judgment in subsequent steps.

[0062] In summary, step S3 normalizes and weights the construction disturbance parameters and cumulative fatigue damage, mapping the external disturbance intensity and internal damage evolution into a unified disturbance state feature vector. It further constructs a probability distribution vector, calculates information entropy, and performs spatiotemporal aggregation, ultimately outputting a comprehensive construction disturbance entropy index and dynamic risk level. This output builds upon the reliable data foundation and component-level damage quantification results from steps S1 and S2, providing structured and comparable observation node inputs for subsequent probabilistic inference based on dynamic Bayesian networks.

[0063] S4. The step of using the construction disturbance entropy index and the cumulative fatigue damage degree as observation nodes to input the dynamic Bayesian network, updating the network parameters online, and outputting the real-time occurrence probability of various engineering failure modes includes: Construct a dynamic Bayesian network comprising observation layer nodes, intermediate mechanism layer nodes, and top-level failure layer nodes, wherein the observation layer nodes include at least the construction disturbance entropy index and the cumulative fatigue damage degree; The construction disturbance entropy index and the cumulative fatigue damage value output in real time from the previous steps are used as evidence of the observation layer node at the current moment and input into the dynamic Bayesian network. Based on the input evidence, the posterior probability distribution of all nodes in the dynamic Bayesian network is updated online using a probabilistic inference algorithm. The update process includes both the correction of the state probability of the intermediate mechanism layer nodes and the recalculation of the occurrence probability of the top-level failure layer nodes. The top-level failure layer nodes correspond to various predefined engineering failure modes. After updating and calculation, the real-time occurrence probability of each type of engineering failure mode under the current evidence is output.

[0064] Step S4 specifically includes: In this embodiment, step S4 is used to take the construction disturbance entropy index output in real time in step S3 and the cumulative fatigue damage degree output in real time in step S2 as the observation layer evidence input of the dynamic Bayesian network, and update the posterior probability distribution of each node in the network online through the probabilistic inference algorithm, thereby outputting the real-time occurrence probability of various predefined engineering failure modes under the current evidence conditions.

[0065] The core technical problem to be addressed in step S4 is that the construction risks of subway foundation pits exhibit significant multi-factor coupling and uncertainty superposition characteristics. The sources of risk simultaneously include external construction disturbances, cumulative responses caused by cyclical train loads, geological and parameter uncertainties, and changes in boundary conditions during the construction phase. Relying solely on a single indicator threshold or deterministic discrimination method makes it difficult to transfer risks between different working conditions and to explain the reasons for differences in risk probabilities under the same external response level.

[0066] This invention uses the construction disturbance entropy index of macroscopic disturbance structure and the cumulative fatigue damage degree of microscopic component degradation process as observation node evidence, enabling probabilistic reasoning to simultaneously receive observation information of system-level state and component-level state, and realize traceable inference of intermediate mechanism state and top-level failure event in dynamic network structure, thereby transforming state description into failure probability quantification.

[0067] (I) Constructing the hierarchical structure and node semantic mapping of dynamic Bayesian networks This embodiment constructs a dynamic Bayesian network comprising observation layer nodes, intermediate mechanism layer nodes, and top-level failure layer nodes. The observation layer nodes include at least a construction disturbance entropy index and cumulative fatigue damage degree, serving as the receivers of the outputs from steps S3 and S2. The intermediate mechanism layer nodes characterize key potential states influencing the occurrence of failure events. The top-level failure layer nodes correspond to various predefined engineering failure modes. To adapt the network structure to changes in the construction phase, this embodiment organizes the dynamic Bayesian network using discrete time slices. These discrete time slices are consistent with the time window used in calculating the comprehensive construction disturbance entropy index in step S3, ensuring a consistent time dimension for the injection of observational evidence, posterior updates, and probability outputs. Through this hierarchical structure, the intermediate mechanism layer nodes act as a probability transmission bridge between observational evidence and failure events, enabling the network to express the statistical correlation between observation and failure while retaining a certain degree of engineering mechanism interpretability.

[0068] In terms of invention concept, the purpose of this hierarchical dynamic structure is to solve the problem of scale inconsistency between observed variables and failure events: construction disturbance entropy index Reflects changes in the system's disturbance distribution structure and cumulative fatigue damage. Reflecting the cumulative degradation of key components over time, these two aspects correspond to the macro-risk situation and micro-degradation evidence, respectively. By listing them side by side as observation layer nodes and linking them through intermediate mechanism layer nodes, the inference results can be made sensitive to both sudden changes in the perturbation structure and continuous accumulation of damage, thus adapting to the risk evolution characteristics of short-term strong perturbations and long-term cumulative degradation coexisting in subway scenarios.

[0069] (ii) Alignment of observational evidence input with time window diameter The construction disturbance entropy index and cumulative fatigue damage value output in real time from the previous steps are used as evidence input to the dynamic Bayesian network for the observation layer nodes at the current moment. In this embodiment, the construction disturbance entropy index is taken from the output of step S3. The time window index is To eliminate the ambiguity between the calculation time scope in step S2 and the time window scope in step S3, this embodiment uses a time window... The end time is recorded as and at time S2 Output cumulative fatigue damage As a time window The corresponding observational evidence input, the This identifies key components. Therefore, the time window... Construction disturbance entropy index Cumulative fatigue damage at the end of the time window Establish corresponding relationships on the same time scale to ensure the synchronicity and traceability of evidence input.

[0070] In terms of evidence input implementation, this embodiment can discretize or perform interval mapping on the observed evidence to adapt it to the state-space representation of a dynamic Bayesian network. For example, it can... Mapped to a finite number of risk situation levels, The data is mapped to several degenerate interval states, so that the observational evidence enters the reasoning module in the form of states, while maintaining consistency with the dynamic risk level output in step S3.

[0071] The targeted value of this evidence input mechanism lies in the fact that construction disturbance parameters and train operating loads exhibit significant time non-stationarity; instantaneous anomalies do not equate to a risk accumulation trend. After step S3 spatiotemporal aggregation, it possesses trend stability; In step S2, a correspondence is established between train passing events and construction stages, providing a traceable cumulative caliber. Using both as evidence input can simultaneously suppress instantaneous spike misjudgments and capture cumulative degradation contributions, improving the engineering consistency of the failure probability output.

[0072] (III) Online updating of probabilistic inference algorithms and calculation of posterior distribution Based on the input evidence, the posterior probability distribution of all nodes in the dynamic Bayesian network is updated online using a probabilistic inference algorithm. This update process simultaneously includes correcting the state probabilities of intermediate mechanism layer nodes and recalculating the occurrence probabilities of top-level failure layer nodes. This embodiment employs a recursive Bayesian update framework: In the time window Upon arrival, the observation evidence is injected into the observation layer nodes. Using the conditional probability structure and time transition structure of the network, the posterior distribution of the intermediate mechanism layer nodes is first updated, and then the probability is propagated from the intermediate mechanism layer nodes to the top failure layer nodes to obtain the posterior probability of each failure mode.

[0073] For ease of explanation, this embodiment can be summarized by the following formula for calculating the posterior probability of a certain failure mode within the current time window:

[0074] in, For the first Failure modes similar to engineering failures within time windows The real-time probability of occurrence; The first given by the prior structure and time transition model of dynamic Bayesian networks Failure modes in time windows The predicted weights; In the time window Observational evidence Under the conditions Evidence likelihood weights for different failure modes; The total number of predefined engineering failure modes; For summation index; For failure mode index; Index for time windows; For time windows The comprehensive construction disturbance entropy index; For time windows End time Key components The cumulative fatigue damage; For time windows The end time; Identify key components.

[0075] It should be noted that the online update not only updates the probability of the top-level failure layer nodes, but also updates the probability of the intermediate mechanism layer nodes. This enables the network to continuously correct its judgment of the mechanism state as construction progresses, disturbance intensity changes, and cumulative damage increases, thereby avoiding the lack of intermediate state explanations when only a black box is output for failure events.

[0076] (iv) Real-time occurrence probability of output engineering failure modes The top-level failure layer nodes correspond to various predefined engineering failure modes. After update calculation, the real-time occurrence probability of each type of engineering failure mode under the current evidence is output. In this embodiment, the output is organized with failure mode identifiers, time window indexes, and real-time occurrence probabilities, and can be compared with the construction disturbance entropy index output in step S3. The dynamic risk level corresponds to the same time window scale and can be correlated with the cumulative fatigue damage level formed by identifying key components in step S2. Establishing connections facilitates multi-dimensional collaborative judgment and matching of handling suggestions in subsequent steps. This output organization method ensures a data structure that maintains consistency between failure probability and evidence source, time window index and failure mode semantics, which is beneficial for threshold comparison and identification of dominant risk factors during subsequent early warning rule execution.

[0077] In summary, step S4 constructs a dynamic Bayesian network structure consisting of an observation layer, a mechanism layer, and a failure layer to incorporate the construction disturbance entropy index. Cumulative fatigue damage at the end of the time window As input for observational evidence, a probabilistic inference algorithm is used to update the posterior distribution of the network online and output the real-time probability of occurrence of each engineering failure mode. This step builds upon the reliable data foundation provided in step S1 and the structured evidence output formed in steps S2 and S3, transforming risk assessment from deterministic threshold discrimination to probabilistic inference output, and providing quantitative basis and traceable evidence for the generation of linkage early warning and disposal suggestions in the subsequent step S5.

[0078] S5. The step of performing multi-dimensional collaborative judgment based on the real-time occurrence probability, the cumulative fatigue damage degree, and the construction disturbance entropy index to generate graded early warning information and corresponding handling suggestions includes: The system continuously receives and integrates the real-time occurrence probability, the cumulative fatigue damage degree, and the construction disturbance entropy index. Based on the preset multi-dimensional linkage early warning rules, it performs collaborative comparison and logical judgment on the values ​​or levels of the above indicators and the corresponding thresholds to determine the current comprehensive early warning level and identify the dominant risk factor. According to the comprehensive early warning level and the dominant risk factor, it matches the corresponding disposal suggestions from the preset disposal measures library. Finally, a graded early warning information containing the comprehensive early warning level and the proposed action measures is generated and output.

[0079] Step S5 specifically includes: In this embodiment, step S5 is used to continuously receive and integrate the real-time occurrence probability of engineering failure modes output by step S4, the cumulative fatigue damage degree output by step S2 at the end of the time window, and the construction disturbance entropy index and dynamic risk level output by step S3 within the same time window scale. Based on the preset multi-dimensional linkage early warning rules, collaborative comparison and logical judgment are completed to determine the comprehensive early warning level and identify the dominant risk factor. Then, disposal suggestions are matched from the preset disposal measures library, and finally, graded early warning information is generated and output. The key problem that step S5 aims to solve is that the construction risk of subway foundation pits is simultaneously manifested as short-term fluctuations of external disturbances and long-term accumulation of component degradation. Moreover, it is affected by the subway operating load, construction stage boundary conditions, monitoring noise, and model uncertainty. The single threshold method is prone to biases such as oversensitivity to short-term peaks leading to false alarms or insensitivity to long-term degradation leading to missed alarms.

[0080] In step S5, this invention employs a multi-dimensional linkage criterion of "probabilistic evidence, damage evidence, and entropy situation evidence." By utilizing the complementary characteristics of these three types of evidence in a statistical sense, a collaborative judgment mechanism is constructed. This ensures that the comprehensive early warning level reflects both the probabilistic inference results of top-level failure events and the cumulative constraints of component degradation, while remaining sensitive to changes in the system's disturbance distribution structure. This results in an executable hierarchical early warning and response suggestion matching process.

[0081] To ensure consistency in the evidence, this embodiment records the time window index of step S3 as... And record the end time of this time window as Step S4 within the time window The output of the first The real-time occurrence probability of a failure mode in a project is denoted as: The comprehensive construction disturbance entropy index output in step S3 is marked as follows: Step S2 at time Key components of the output The cumulative fatigue damage is denoted as Step S5 first involves organizing the three types of indicators into a unified event-based system: [The process involves] organizing all... , , Index by Time Window Inputs are aggregated into the same batch, and missing and delayed values ​​are processed for consistency to ensure that the evidence entering the linkage warning rule belongs to the synchronous snapshot under the same time window scale, thus avoiding cross-window mixing and causing caliber deviation.

[0082] Subsequently, based on preset multi-dimensional linkage early warning rules, probabilistic evidence, damage evidence, and entropy situation evidence are mapped in a hierarchical manner. In this embodiment, the level of entropy situation evidence is not redefined, but the dynamic risk level output in step S3 is directly adopted, denoted as... This ensures that the entropy status level is consistent with the definition used in step S3. When only [the following information is obtained] in the engineering implementation... When the numerical value does not directly carry the dynamic risk level output from step S3, the following segmented threshold mapping method can be used to... Equivalent mapping is This mapping serves only as an equivalent implementation of the caliber and does not constitute a new hierarchical system.

[0083] To facilitate the calculation of linkage rules, this embodiment further defines a time window. The aggregate statistics of maximum failure probability and maximum component damage degree as rule inputs

[0084] in, For time windows The maximum real-time occurrence probability of various engineering failure modes; For the first Failure modes similar to engineering failures within time windows The real-time probability of occurrence; For engineering failure mode index; Index for time windows.

[0085]

[0086] in, For a moment The maximum cumulative fatigue damage of each key component; As a key component At any moment The cumulative fatigue damage; Identify key components; For time windows The end time; Index for time windows.

[0087] In obtaining , and Subsequently, this embodiment classifies probabilistic evidence and damage evidence into different levels to obtain probability levels. Damage level The probability level uses a threshold. Damage levels are classified using thresholds. The threshold for classification is pre-set and assimilated into the rule configuration file by the project management team based on historical statistical distribution, design control indicators, and operational management requirements.

[0088] in, For time windows The probability level; For time windows The maximum probability of failure; The probability level threshold is 1; The probability level threshold is 2; The probability level threshold is 3; Index for time windows.

[0089]

[0090] in, For time windows The degree of damage; For a moment The maximum cumulative fatigue damage; The damage level threshold is 1; The damage level threshold is 2; The damage level threshold is 3; Index for time windows.

[0091] When using the equivalent mapping method to obtain the entropy situation level, it can be... By threshold Mapped to :

[0092] in, For time windows The entropy status level is consistent with the dynamic risk level output in step S3; For time windows The comprehensive construction disturbance entropy index; The threshold for the status quo is 1; The entropy status level threshold is 2; The entropy status level threshold is 3; Index for time windows.

[0093] After completing the level mapping, this embodiment performs collaborative comparison and logical judgment of multi-dimensional linkage early warning rules to determine the comprehensive early warning level. To ensure that the early warning results remain sensitive to high-risk evidence, while avoiding the occasional anomaly of a single piece of evidence directly triggering the highest level, this embodiment adopts a combined strategy at the rule layer: "the primary level is determined by the supremacy of the three levels, plus consistency constraints." The comprehensive early warning level can be calculated as follows:

[0094] in, For time windows The overall early warning level; Probability level; Damage level; Entropy status level; Index for time windows.

[0095] In engineering implementation, the rule engine further... The triggering of consistency constraints is used to reduce the probability of false alarms.

[0096] For example, when Depend on Individual lifting and When all are at a low level, the rule engine marks the result as a situational alert and requires that the same upward trend be met within a certain number of consecutive time windows before the comprehensive warning level is raised. when Continue to rise During short-term declines, the rules engine maintains the overall warning level without downgrading, reflecting the long-term significance of the monotonic accumulation of cumulative fatigue damage for risk constraints.

[0097] The aforementioned linkage mechanism combines the immediate sensitivity of probabilistic evidence, the cumulative constraint of damage evidence, and the structural sensitivity of entropy state evidence into an executable criterion, thereby forming differentiated decision-making criteria for situations such as short-term disturbance spikes, construction phase switching drift, and slow rise of component degradation.

[0098] In determining Subsequently, this embodiment identifies the dominant risk factor to drive the matching of treatment recommendations. The dominant risk factor identification rule is based on the relative magnitude of the three levels: when achieve And when it leads the other levels, the dominant risk factor is determined to be the probability of failure; when achieve And when it leads the other levels, the dominant risk factor is determined to be cumulative fatigue damage; when achieve When it leads other levels, the dominant risk factor is determined to be dominated by the perturbation entropy state. In the event of a tie, the rule engine makes a decision according to the preset priority and records the tie information in the output to ensure that the determination process of the dominant risk factor is traceable.

[0099] Finally, this embodiment matches corresponding disposal suggestions from a pre-set disposal measure library based on the comprehensive early warning level and the dominant risk factor. The disposal measure library stores multiple disposal measure records in a structured manner. Each record includes the applicable comprehensive early warning level range, the applicable dominant risk factor category, the applicable construction stage label, and a disposal suggestion template.

[0100] The matching process first filters by comprehensive early warning level, then by dominant risk factor, and then by construction stage to obtain a set of disposal suggestions. The set is then sorted by priority to form an output suggestion sequence.

[0101] To formally describe this matching process, this embodiment represents the mapping relationship of the treatment measure library as follows:

[0102] in, For time windows A set of proposed solutions; For matching and mapping of the disposal measures library; For time windows The overall early warning level; Index for time windows.

[0103] When generating tiered early warning information, this embodiment will Collection of disposal recommendations Encapsulate it as an early warning output object, and carry accompanying fields related to the evidence in the output object, including the time window index. Comprehensive construction disturbance entropy index Maximum failure probability Maximum cumulative fatigue damage The results of the determination of the leading risk factors form structured information that can be used for subsequent recording, auditing and review.

[0104] Example 2 A system employing a safety assessment method for construction of foundation pits adjacent to subway lines, characterized in that it includes: The system comprises the following modules: a multi-source acquisition and assimilation module, a multi-source monitoring data module for the foundation pit and adjacent subway structures, and subway operation load data, which collects construction disturbance parameters, multi-source monitoring data, and subway operation load data. It constructs and updates a foundation pit-subway digital twin model consistent with the physical state using spatiotemporal alignment and dynamic data assimilation methods. A dynamic response extraction module extracts the dynamic response of key components based on the foundation pit-subway digital twin model and calculates the cumulative fatigue damage of key components under cyclic loading. A disturbance entropy index calculation module integrates cumulative fatigue damage and construction disturbance parameters to calculate a construction disturbance entropy index that comprehensively reflects system risk. A Bayesian probabilistic reasoning module uses the construction disturbance entropy index and cumulative fatigue damage as observation nodes input to a dynamic Bayesian network, updates network parameters online, and outputs the real-time occurrence probability of various engineering failure modes. A coordinated early warning decision-making module performs multi-dimensional collaborative judgment based on real-time occurrence probability, cumulative fatigue damage, and construction disturbance entropy index, generating tiered early warning information and corresponding handling suggestions.

[0105] The technical features of this invention not described can be implemented by or using existing technology, and will not be repeated here. Of course, the above description is not a limitation of this invention, and this invention is not limited to the examples above. Any changes, modifications, additions or substitutions made by those skilled in the art within the scope of this invention should also be within the protection scope of this invention.

Claims

1. A method for safety assessment of construction of foundation pits adjacent to subway lines, characterized in that, Includes the following steps: S1. Collect construction disturbance parameters, multi-source monitoring data of the foundation pit and adjacent subway structures, and subway operation load data. Construct and update a foundation pit-subway digital twin model consistent with the physical state using spatiotemporal alignment and dynamic data assimilation methods. S2. Extract the dynamic response of key components based on the foundation pit-subway digital twin model and calculate the cumulative fatigue damage degree of key components under cyclic loads. S3. Integrate the cumulative fatigue damage degree and the construction disturbance parameters to calculate the construction disturbance entropy index, which comprehensively reflects the system risk. S4. Use the construction disturbance entropy index and the cumulative fatigue damage degree as observation nodes to input into a dynamic Bayesian network, update network parameters online, and output the real-time occurrence probability of various engineering failure modes. S5. Based on the real-time occurrence probability, the cumulative fatigue damage degree, and the construction disturbance entropy index, perform multi-dimensional collaborative judgment to generate graded early warning information and corresponding handling suggestions.

2. The method for safety assessment of construction of foundation pits near subway lines according to claim 1, characterized in that, The collection of construction disturbance parameters, multi-source monitoring data of the foundation pit and adjacent subway structures, and subway operation load data specifically includes: simultaneous collection and preprocessing of the three types of data. The first category is construction disturbance parameters, which continuously acquire real-time engineering parameters reflecting the intensity of excavation, support, and dewatering activities by connecting the automated control system for foundation pit construction with monitoring instruments. The second category is multi-source monitoring data of the foundation pit and adjacent subway structure. It involves deploying sensor arrays at key parts of the engineering structure and collecting physical quantities that reflect the mechanical state of the structure and the response of the soil layer according to a preset sampling strategy. These physical quantities include, but are not limited to, deformation, stress, pore water pressure and vibration acceleration. The vibration data is then sliced ​​into time slices according to the train passing events. The third category is subway operation load data, which is obtained through data interfaces or dedicated monitoring equipment to acquire train operation status parameters in the affected sections; All collected raw data streams are subject to uniform time synchronization and spatial coordinate association, and undergo standardized preprocessing including outlier identification, noise filtering, and confidence assessment to finally generate a multi-source fusion dataset that is time-synchronized, coordinate-associated, and includes quality / confidence labels.

3. The method for safety assessment of construction of subway foundation pits according to claim 2, characterized in that, The construction and updating of the foundation pit-subway digital twin model, consistent with the physical state, through spatiotemporal alignment and dynamic data assimilation methods specifically includes: First, a spatiotemporal alignment operation is performed on the multi-source fusion dataset. The spatiotemporal alignment operation is to uniformly transform the spatial coordinates of all monitoring data to an engineering coordinate system or a preset reference coordinate system based on the subway structure axis, and to synchronously calibrate the timestamps of all data. Then, based on this aligned dataset, a parameterized foundation pit-subway coupled mechanical model is iteratively updated using a dynamic data assimilation method. The dynamic data assimilation method uses key soil parameters and structural parameters in the coupled mechanical model as state variables to be corrected, and spatiotemporally aligned monitoring data as observation constraints. A sequential assimilation algorithm is used to compare the observation data with the model prediction values ​​in a rolling manner, and the state variables and model states are adjusted in reverse according to the comparison results so that the prediction residuals meet the preset consistency criteria. By iteratively executing the above assimilation process, the output of the coupled mechanical model is made consistent with the real-time monitoring under the consistency criterion, thereby obtaining and continuously updating the foundation pit-subway digital twin model that is consistent with the physical state.

4. The method for safety assessment of construction of subway foundation pits according to claim 3, characterized in that, The extraction of the dynamic response of key components based on the foundation pit-subway digital twin model includes: under the condition that the foundation pit-subway digital twin model satisfies the preset consistency criterion, and based on the preset engineering safety concerns, defining a set of key components from the engineering structure represented by the foundation pit-subway digital twin model. Simultaneously, the impact time window of the train passing through the adjacent subway tunnel is analyzed and generated in real time from the subway operation load data, and a correspondence is established between the impact time window and the train passing event; Then, the foundation pit-subway digital twin model is driven to perform transient dynamic analysis for each of the impact time windows, and the complete dynamic time history data of each component in the key component set within the corresponding impact time window is located and extracted from the analysis results. Finally, a dynamic response dataset is generated and output to characterize the dynamic behavior of each key component. The dynamic response dataset is indexed by the key component identifier and the sequence of train passing events, and serves as the direct input for calculating the cumulative fatigue damage.

5. The method for safety assessment of construction of foundation pits near subway lines according to claim 4, characterized in that, The calculation of the cumulative fatigue damage of the key components under cyclic loading includes: taking the dynamic response dataset output in the previous step as input, for each key component in the dataset; First, the relevant physical quantities in its dynamic time history data are converted into the stress time history of the key component, and the stress time history is then correlated with the train passing event and the construction stage. Then, the stress time history is processed using a cycle counting algorithm to identify and count the independent stress cycles contained therein, and the amplitude and mean of each stress cycle are recorded. Next, the fatigue performance parameters of the materials corresponding to the key components, which are pre-stored and associated with the key component identifier, are queried, and each identified stress cycle is converted into a damage increment according to the stress-life relationship and the mean stress correction rule. Finally, all the damage increments are accumulated in chronological order to output the cumulative fatigue damage of the key component from the start of construction to the current calculation time.

6. The method for safety assessment of construction of subway foundation pits according to claim 5, characterized in that, The calculation of the construction disturbance entropy index, which comprehensively reflects the system risk, by integrating the cumulative fatigue damage degree and the construction disturbance parameters includes: normalizing and weighting the construction disturbance parameters and the cumulative fatigue damage degree to form a multidimensional disturbance state feature vector. A probability distribution vector characterizing the perturbation intensity distribution is constructed based on each component in the perturbation state feature vector. The basic perturbation entropy value is obtained by calculating the probability distribution vector using the information entropy function; The basic disturbance entropy value is smoothed in the time dimension and weighted and aggregated in the spatial dimension to generate a comprehensive construction disturbance entropy index. The dynamic risk level is determined based on the historical statistical distribution of the construction disturbance entropy index, and the construction disturbance entropy index and the dynamic risk level are output.

7. The method for safety assessment of construction of foundation pits near subway lines according to claim 6, characterized in that, The process of using the construction disturbance entropy index and the cumulative fatigue damage degree as observation nodes to input into a dynamic Bayesian network, updating network parameters online, and outputting the real-time occurrence probability of various engineering failure modes includes: Construct a dynamic Bayesian network comprising observation layer nodes, intermediate mechanism layer nodes, and top-level failure layer nodes, wherein the observation layer nodes include at least the construction disturbance entropy index and the cumulative fatigue damage degree; The construction disturbance entropy index and the cumulative fatigue damage value output in real time from the previous steps are used as evidence of the observation layer node at the current moment and input into the dynamic Bayesian network. Based on the input evidence, the posterior probability distribution of all nodes in the dynamic Bayesian network is updated online using a probabilistic inference algorithm. The update process includes both the correction of the state probability of the intermediate mechanism layer nodes and the recalculation of the occurrence probability of the top-level failure layer nodes. The top-level failure layer nodes correspond to various predefined engineering failure modes. After updating and calculation, the real-time occurrence probability of each type of engineering failure mode under the current evidence is output.

8. The method for safety assessment of construction of subway foundation pits according to claim 7, characterized in that, The step of generating graded early warning information and corresponding handling suggestions by performing multi-dimensional collaborative judgment based on the real-time occurrence probability, the cumulative fatigue damage degree, and the construction disturbance entropy index includes: The system continuously receives and integrates the real-time occurrence probability, the cumulative fatigue damage degree, and the construction disturbance entropy index. Based on the preset multi-dimensional linkage early warning rules, it performs collaborative comparison and logical judgment on the values ​​or levels of the above indicators and the corresponding thresholds to determine the current comprehensive early warning level and identify the dominant risk factor. According to the comprehensive early warning level and the dominant risk factor, it matches the corresponding disposal suggestions from the preset disposal measures library. Finally, a graded early warning information containing the comprehensive early warning level and the proposed action measures is generated and output.

9. A system employing the safety assessment method for construction of subway foundation pits as described in any one of claims 1-8, characterized in that, include: Multi-source acquisition and assimilation module: Collects construction disturbance parameters, multi-source monitoring data of the foundation pit and adjacent subway structures, and subway operation load data. Constructs and updates a foundation pit-subway digital twin model consistent with the physical state through spatiotemporal alignment and dynamic data assimilation methods. Dynamic response extraction module: Extracts the dynamic response of key components based on the foundation pit-subway digital twin model and calculates the cumulative fatigue damage of key components under cyclic loads. Disturbance entropy index calculation module: Merges the cumulative fatigue damage and construction disturbance parameters to calculate a construction disturbance entropy index that comprehensively reflects the system risk. Bayesian probabilistic inference module: Uses the construction disturbance entropy index and the cumulative fatigue damage as observation nodes to input a dynamic Bayesian network, updates network parameters online, and outputs the real-time occurrence probability of various engineering failure modes. Linked early warning decision module: Based on the real-time occurrence probability, the cumulative fatigue damage degree and the construction disturbance entropy index, it performs multi-dimensional collaborative judgment to generate graded early warning information and corresponding handling suggestions.

Images