An intelligent pressure detection method suitable for basement side wall

Abstract

The application relates to the technical field of underground engineering structure safety monitoring, and discloses an intelligent pressure detection method suitable for basement side walls. The method starts multiple sensing units, acquires multi-dimensional sensing signals with space coding, generates calibrated stress data streams through signal purification and compensation. The data streams are introduced into a dynamically updated knowledge graph network for situational analysis. The network integrates structural component relationships, historical load and geological archive information. In this context, a pressure evolution deduction process is started to simulate the stress diffusion process. Through multiple rounds of iteration comparison with the knowledge base, stress abnormal evolution trajectories deviating from the normal mode are identified. The time-space coordinates and diffusion rate of the trajectories are extracted as core features and encapsulated as abnormal event descriptors. The application realizes situational deep interpretation of pressure data and early identification of abnormal evolution trajectories.

Description

Technical Field

[0001] This invention relates to the field of underground engineering structure safety monitoring technology, specifically to an intelligent pressure detection method applicable to basement sidewalls. Background Technology

[0002] In the field of underground engineering structural safety monitoring, existing technologies mainly rely on discrete sensors deployed at key points of the structure to collect pressure data and employ static alarm mechanisms with fixed thresholds. These methods can only provide point-based instantaneous values ​​and cannot characterize the spatial distribution and dynamic transmission patterns of pressure throughout the entire sidewall structure system. Furthermore, conventional signal processing techniques often focus on filtering out electrical or environmental noise; the processed data remains essentially an isolated numerical sequence detached from the specific engineering physics context, lacking the ability to interpret it within a specific context.

[0003] Because the data is disconnected from structural mechanics models, historical operating conditions, and external environmental information, existing systems struggle to understand the true engineering meaning of pressure readings. This leads to a heavy reliance on human experience and simple thresholds for judging abnormal states, making it impossible to effectively distinguish between normal time-varying loads and abnormal precursors that characterize potential structural risks. Existing methods lack the ability to analyze and extrapolate the evolution of the pressure field in the spatiotemporal dimensions; they can only trigger alarms after values ​​exceed preset limits, representing a reactive, ex-post response that fails to identify and warn of abnormal development trends early.

[0004] There is a need for a method that can dynamically analyze multi-point pressure data collected in real time within a contextualized model that integrates structural physical relationships, historical load patterns, and environmental knowledge, and simulate the pressure propagation process based on this model, thereby proactively identifying abnormal patterns that deviate from normal mechanical behavior in the evolution trajectory. Summary of the Invention

[0005] The purpose of this invention is to provide an intelligent pressure detection method suitable for basement sidewalls to solve the problems mentioned in the background art.

[0006] To achieve the above objectives, the present invention provides an intelligent pressure detection method suitable for basement sidewalls, the method comprising:

[0007] Multiple sets of sensing units deployed at a specified depth on the side wall of the basement are activated to acquire multi-dimensional physical quantities, including pressure amplitude and direction of action, in a periodic sampling mode, forming a set of sensing signals containing spatial coding identifiers.

[0008] The sensing signal group is subjected to multi-level signal purification processing to remove the environmental noise substrate, separate the pure stress component representing the stress of the structure, and perform signal attenuation compensation to generate a calibrated stress data stream.

[0009] The calibrated stress data stream is imported into a dynamically updated knowledge graph network. The knowledge graph network uses the side wall structural components as nodes and the mechanical transmission relationship between components as edges. It continuously obtains historical load records, nearby geological activity reports and construction archives related to the side wall from external databases as auxiliary knowledge to perform contextualized analysis of the calibrated stress data stream.

[0010] In the context of the knowledge graph network, the stress evolution simulation process is initiated. Starting from the calibrated stress data stream, the diffusion, superposition and attenuation process of stress in the side wall structure is simulated. Through multiple rounds of iterative comparison with the auxiliary knowledge in the knowledge graph network, the abnormal stress evolution trajectory that contradicts the known normal mechanical behavior pattern is identified.

[0011] The initial spatiotemporal coordinates, diffusion rate, and morphological changes of the stress anomaly evolution trajectory are extracted as core features, and these core features are encapsulated into an anomaly stress event descriptor that can be processed by subsequent decision logic.

[0012] Preferably, the step of performing multi-level signal purification processing on the sensed signal group, stripping away the environmental noise substrate, separating the pure stress component representing the structural stress, and performing signal attenuation compensation to generate a calibrated stress data stream, specifically includes:

[0013] The sensing signal group is initially decomposed to distinguish the regular environmental interference signals caused by the diurnal temperature cycle and the periodic fluctuation of the groundwater level, as well as the irregular background noise caused by the slight drift of the sensing unit itself, and to establish the dynamic noise spectrum of the current monitoring period.

[0014] Using the dynamic noise spectrum, the sensing signal group is subjected to composite filtering in the frequency and time domains to filter out the known environmental noise base and obtain a preliminary stress signal containing residual noise.

[0015] Based on the material aging coefficient and installation time of the sensing unit, a signal attenuation compensation curve is calculated, and the amplitude recovery correction is performed on the initial stress signal.

[0016] From the corrected preliminary stress signal, based on the characteristic shape and energy distribution of the stress waveform, the response waveform directly related to the load on the side wall structure is extracted, spurious stress signals caused by non-structural reasons are eliminated, and finally aggregated to form the calibrated stress data stream.

[0017] Preferably, the step of importing the calibrated stress data stream into a dynamically updated knowledge graph network, wherein the knowledge graph network uses sidewall structural members as nodes and the mechanical transfer relationships between members as edges, and continuously obtains historical load records, nearby geological activity reports, and construction archives related to the sidewall from external databases as auxiliary knowledge, in order to perform contextualized analysis of the calibrated stress data stream, specifically including:

[0018] A topological mechanical model of the basement sidewalls is constructed. Each wall, corner, and connection point with the floor / top slab is defined as a node in the knowledge graph network. The load transmission path and attenuation relationship between nodes are defined as directed weighted edges, thus completing the structural initialization of the knowledge graph network.

[0019] Establish continuous connections with multiple heterogeneous external databases, extract typical load patterns from the historical load records according to the preset synchronization strategy, analyze potential geological disturbance factors from the adjacent geological activity reports, obtain structural details and material parameters from the construction archives, and attach the information as attribute labels or constraint rules to the corresponding nodes and edges in the knowledge graph network.

[0020] Each data point in the calibrated stress data stream is mapped to a corresponding node in the knowledge graph network based on its spatial encoding identifier.

[0021] By combining the attribute labels and constraint rules currently attached to the node, as well as the mechanical transfer properties of its associated edges, the rationality score of the stress data in the local context of the knowledge graph network is calculated, and all contextual factors that are significantly related to the current stress data are recorded to complete the contextualized analysis.

[0022] Preferably, in the context of the knowledge graph network, the stress evolution simulation process is initiated, starting with the calibrated stress data stream to simulate the diffusion, superposition, and attenuation of stress in the sidewall structure. Through multiple rounds of iterative comparison with auxiliary knowledge in the knowledge graph network, anomaly stress evolution trajectories that contradict known normal mechanical behavior patterns are identified. Specifically, this includes:

[0023] Using the nodes mapped in the knowledge graph network by the calibrated stress data stream as stress injection points, and based on the mechanical transmission relationships and weights defined in the knowledge graph network, numerical simulations of stress diffusion are performed along the directed edges.

[0024] In each diffusion simulation step, the attenuation and deformation of stress during transmission are calculated based on the material damping and connection stiffness parameters defined by the node's own attribute labels.

[0025] The stress distribution state obtained from each step of the simulation is compared in real time with the normal stress threshold range and historical stress mode defined by auxiliary knowledge at the node.

[0026] When the simulated stress value continues to exceed the normal threshold range, or its distribution pattern deviates from the historical pattern, record the node position, stress state and deviation degree of the current simulation step, and connect the abnormal state points in chronological order and transmission direction.

[0027] The simulation continues until the stress decays to a negligible level, or until all relevant nodes in the knowledge graph network have been traversed. Finally, the sequence of anomalous state points formed by the connection is marked as a complete stress anomaly evolution trajectory.

[0028] Preferably, the extraction of the initial spatiotemporal coordinates, diffusion rate, and morphological changes of the stress anomaly evolution trajectory as core features, and the encapsulation of these core features into an anomaly stress event descriptor that can be processed by subsequent decision-making logic, specifically includes:

[0029] The first point marked as an abnormal state is parsed from the stress anomaly evolution trajectory, and its corresponding knowledge graph network node identifier and timestamp are extracted as the starting spatiotemporal coordinates of the abnormal pressure event.

[0030] Calculate the ratio of stress propagation distance to time interval between adjacent abnormal state points on the stress anomaly evolution trajectory to obtain the stress diffusion rate on the path, and calculate the average diffusion rate of the entire trajectory.

[0031] Analyze the changes in stress distribution morphology along the stress anomaly evolution trajectory, including the changing trend of stress peak values ​​and the movement path of stress concentration areas;

[0032] The initial spatiotemporal coordinates, the average diffusion rate, the changes in the diffusion rate, and the key parameters of the stress distribution morphology changes are organized according to a predefined data structure to generate the abnormal pressure event descriptor containing a unique event identifier.

[0033] Preferably, after generating the abnormal pressure event descriptor, the method further includes:

[0034] The abnormal stress event descriptor is input into a causal reasoning engine, and the causal model of the causal reasoning engine is pre-constructed based on the structure and auxiliary knowledge of the knowledge graph network;

[0035] Within the framework of the causal model, the causal reasoning engine enumerates potential causal hypotheses that lead to the phenomena represented by the abnormal stress event descriptor. These potential causal hypotheses include internal structural damage types, external improper loads, and sudden changes in environmental factors.

[0036] The causal reasoning engine retrieves evidence from the calibrated stress data stream, auxiliary knowledge in the knowledge graph network, and a library of historical anomalous stress event descriptors to support or refute each potential causal hypothesis.

[0037] Based on the retrieved evidence, the causal reasoning engine calculates a confidence score for each potential causal hypothesis and generates a potential causal inference report sorted by probability in descending order based on the scores.

[0038] Preferably, within the framework of the causal model, the causal inference engine enumerates potential causal hypotheses leading to the phenomenon represented by the abnormal stress event descriptor, specifically including:

[0039] The causal reasoning engine loads a causal model describing the causal relationships between various elements within the basement sidewall system. The causal model considers pressure anomalies as "results" and structural states, external actions, environmental variables, etc., as possible "cause" nodes, and defines the direction and intensity of causal influence between nodes.

[0040] The causal reasoning engine takes the feature parameters in the abnormal stress event descriptor as input and traces back in the causal model to the set of all "cause" nodes that can lead to this "result".

[0041] For each traced "cause" node, a specific fault or event scenario described in natural language is generated by combining its path weight with the "result" node and the node's prior probability, thus forming a complete potential cause hypothesis.

[0042] Preferably, based on the retrieved evidence, the causal inference engine calculates a confidence score for each potential causal hypothesis and generates a potential causal inference report sorted by probability in descending order based on the scores, specifically including:

[0043] For each of the potential causal hypotheses, retrieve direct observation data, historical similar cases, and domain rules related to the potential causal hypotheses from the knowledge graph network and external data;

[0044] Based on the strength of the causal relationship between each piece of evidence retrieved and the potential causal hypothesis, and the reliability level of the evidence itself, the degree of support or rejection of the evidence for the potential causal hypothesis is calculated.

[0045] By combining the support and rejection of all evidence, and using the Bayesian update rule, the posterior probability of the potential cause hypothesis under all available evidence is calculated as its confidence score.

[0046] All potential causal hypotheses are sorted from highest to lowest confidence score, and a summary of key evidence supporting each hypothesis is attached to generate a structured report of potential causal inferences.

[0047] Preferably, after generating the potential cause inference report, the method further includes:

[0048] The current complete processing procedure, including the calibrated stress data stream, the abnormal pressure event descriptor, the potential cause inference report, and the key intermediate data in the causal reasoning process, is packaged into a knowledge case.

[0049] The knowledge cases are stored in a continuously evolving case library, which is indexed in multiple dimensions according to stress anomaly patterns, causes, and structural parts.

[0050] Periodically cluster and pattern mining are performed on all knowledge cases in the case library to discover recurring stress anomalies and causal combinations. These stress anomaly and causal combination patterns are then used as new empirical rules and fed back into the auxiliary knowledge of the knowledge graph network to enhance the accuracy of subsequent contextualized analysis. They also serve as new prior knowledge to update the causal model of the causal reasoning engine.

[0051] Preferably, the step of periodically clustering and pattern mining all knowledge cases in the case library to discover recurring stress anomalies and causal combinations, and then feeding these stress anomaly and causal combination patterns as new empirical rules into the auxiliary knowledge of the knowledge graph network, specifically includes:

[0052] Set a fixed time period to trigger a full scan and analysis task of the case library;

[0053] The abnormal stress event descriptor feature vectors of all knowledge cases in the case library are extracted and the finally confirmed cause labels are combined to form a high-dimensional dataset.

[0054] An unsupervised learning algorithm is used to perform cluster analysis on the high-dimensional dataset to identify multiple case groups that cluster in the feature space;

[0055] For each case group, analyze the consistency of abnormal characteristics and causes among the cases within it, and extract the common and stable abnormal patterns and cause association rules shared by the case group.

[0056] Each extracted association rule is formatted as “IF abnormal pattern THEN possible cause” and assigned an initial confidence weight based on the number of cases and consistency in the case group.

[0057] The formatted association rules are used as incremental knowledge and synchronously updated to the rule base of the knowledge graph network, and the causal strength parameters between corresponding nodes in the causal reasoning engine are adjusted.

[0058] Compared with the prior art, the beneficial effects of the present invention are:

[0059] By constructing a knowledge graph network that integrates the mechanical relationships of structural components with multi-source historical information, real-time acquired stress data is analyzed within a specific structural topology and historical context. This method associates discrete sensor readings with specific structural nodes and force transmission paths, and cross-validates them using historical load patterns, geological reports, and construction archives. This allows the system to understand the physical meaning of pressure data based on complete contextual information. This technology transforms data interpretation from processing isolated numerical sequences to analyzing contextualized information with clear structural semantics, improving the accuracy and reliability of condition assessment.

[0060] Based on the knowledge graph network, the system initiates a stress evolution simulation process, dynamically simulating the diffusion path and intensity change of stress in the structure according to the mechanical transmission relationship. This process uses real-time calibration data as initial conditions and continuously compares it with normal mechanical behavior patterns stored in the knowledge base through multiple iterations. This mechanism can capture subtle evolutionary characteristics of stress distribution in the spatiotemporal dimensions, identifying abnormal trajectories that may not exceed thresholds in absolute values ​​but have systematically deviated from normal patterns in their development. This technology shifts the monitoring focus from post-event threshold alarms to pre-event evolution trend warnings.

[0061] Based on the abnormal evolution trajectories identified through the aforementioned deduction and comparison process, the system automatically extracts core physical characteristics such as their starting position, occurrence time, spatial diffusion rate, and morphological changes, and encapsulates them into structured abnormal event descriptors. These descriptors fully preserve the spatiotemporal dynamic characteristics of the abnormal patterns, providing directly processable and physically meaningful information input for subsequent risk assessment and decision analysis. This processing flow forms a complete technical closed loop from data acquisition, context analysis, dynamic deduction to feature generation. Attached Figure Description

[0062] Figure 1 This is a schematic diagram illustrating the working principle of the intelligent pressure detection method for basement sidewalls described in this invention.

[0063] Figure 2 A flowchart for multi-stage signal purification processing;

[0064] Figure 3 Flowchart generated for abnormal stress event descriptors;

[0065] Figure 4A comparison chart of key indicators at each stage of the entire process of basement sidewall pressure testing;

[0066] Figure 5 This is a heatmap showing the correlation between the dimensions of abnormal features. Detailed Implementation

[0067] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0068] Please see Figure 1 This invention provides an intelligent pressure detection method for basement sidewalls. The method includes: activating multiple sets of sensing units deployed at a specified depth on the basement sidewall. These sensing units operate in a periodic sampling mode, collecting multi-dimensional physical quantities including pressure amplitude and direction. The signals collected by each set of sensing units contain their unique spatial coding identifier, forming a sensing signal group. Subsequently, multi-level signal purification processing is performed on the sensing signal group to remove the environmental noise substrate and separate the pure stress components representing the structural stress. This process also includes signal attenuation compensation, ultimately outputting a calibrated stress data stream. The calibrated stress data stream is imported into a dynamically updated knowledge graph network. This knowledge graph network uses the structural components of the sidewall as nodes and the mechanical transmission relationships between components as edges. It continuously acquires auxiliary knowledge such as historical load records, nearby geological activity reports, and construction archives from external databases, using this knowledge to perform contextualized analysis of the input stress data stream. Within the context provided by the knowledge graph network, a stress evolution simulation process is initiated. Using a calibrated stress data stream as initial input, the diffusion, superposition, and attenuation of stress in the node network of the sidewall structure are simulated. The simulation process is iteratively compared with auxiliary knowledge in the knowledge graph network in multiple rounds to identify abnormal stress evolution trajectories that deviate from known normal mechanical behavior patterns. Core features such as the initial spatiotemporal coordinates, diffusion rate, and morphological changes of the abnormal stress evolution trajectory are extracted and encapsulated into structured abnormal stress event descriptors for use by subsequent decision-making systems.

[0069] In one embodiment of the present invention, see [reference] Figure 2The sensed signal set is initially decomposed to distinguish between regular environmental interference signals caused by diurnal temperature cycles and periodic fluctuations in groundwater levels, and irregular background noise caused by minor drifts of the sensing unit itself. This establishes a dynamic noise spectrum reflecting the characteristics of the current monitoring cycle. Using the dynamic noise spectrum, a composite filtering operation in the frequency and time domains is performed on the sensed signal set to filter out the identified environmental noise base, obtaining a preliminary stress signal containing residual noise. Based on the material aging coefficient and cumulative installation time of the sensing unit, a signal attenuation compensation curve is calculated, and this curve is used to correct the amplitude of the preliminary stress signal. From the corrected preliminary stress signal, based on the characteristic shape and energy distribution of the stress waveform, the response waveform directly related to the load on the sidewall structure is extracted, while spurious stress signals caused by non-structural reasons are eliminated. Finally, the data is aggregated to form a calibrated stress data stream.

[0070] A topological mechanical model of the basement sidewalls is constructed. Each independent wall, structural corner, and connection point with the floor or roof slab are explicitly defined as nodes in a knowledge graph network. The load transfer paths and attenuation relationships between these nodes are defined as directed weighted edges, completing the structural initialization of the knowledge graph network. Continuous connections are established with multiple heterogeneous external databases. Following a pre-defined synchronization strategy, typical load patterns are extracted from historical load records, potential geological disturbance factors are analyzed from nearby geological activity reports, and structural details and material parameters are obtained from construction archives. This information is then used as attribute labels or constraint rules and attached to the corresponding nodes and edges in the knowledge graph network. During the data import phase, each data point in the calibrated stress data stream is mapped to a corresponding node in the knowledge graph network based on its spatial coding identifier. Combining the currently attached attribute labels and constraint rules of the node, as well as the mechanical transfer properties of its associated edges, the rationality score of the stress data point in the local context of the knowledge graph network is calculated. All contextual factors significantly related to the current stress data are recorded, completing contextualized analysis.

[0071] In practice, the sensing signal group undergoes multi-level signal purification processing. The sensing signal group contains periodic sampling data from multiple sets of sensing units deployed at a specified depth on the basement sidewall. The data includes multi-dimensional physical quantities such as pressure amplitude and direction of action. In practice, the sensing signal group is initially decomposed to distinguish regular environmental interference signals caused by diurnal temperature cycles and periodic fluctuations in groundwater levels, as well as irregular background noise caused by the slight drift of the sensing units themselves. For example, the pressure amplitude exhibits sinusoidal fluctuations with a 24-hour period during the monitoring period, which is identified as an environmental interference signal. At the same time, the signal baseline exhibits slow random shifts, which are identified as background noise. A dynamic noise spectrum for the current monitoring period is established, which quantifies the frequency amplitude characteristics of these interference signals. In some embodiments, a dynamic noise spectrum is used to perform composite filtering on the sensing signal group in both the frequency and time domains. In the frequency domain, a band-stop filter is used to remove characteristic frequency components corresponding to environmental interference signals in the dynamic noise spectrum. In the time domain, a moving average filter is used to suppress irregular background noise. After filtering out the known environmental noise floor, a preliminary stress signal containing residual noise is obtained. It can be understood that the preliminary stress signal still contains random noise components not modeled in the dynamic noise spectrum. In a specific implementation, a signal attenuation compensation curve is calculated based on the material aging coefficient of the sensing unit and the installation time. The material aging coefficient of the sensing unit is provided by the manufacturer, and the installation time is obtained from the deployment log. The signal attenuation compensation curve is used to correct the decrease in the sensitivity of the sensing unit over time, performing amplitude recovery correction on the preliminary stress signal. The correction process uses the following formula:

[0072]

[0073] in: This represents the compensated signal amplitude. This represents the initial amplitude of the stress preliminary signal. Indicates the material aging coefficient. This indicates the installation time. Optionally, response waveforms directly related to the sidewall structure under load are extracted from the corrected preliminary stress signal based on the characteristic shape and energy distribution of the stress waveform. Waveform segments with fast rise edges and high energy concentration are identified as structural response waveforms. False stress signals with dispersed energy and irregular shapes are eliminated. Finally, the signals are aggregated to form a calibrated stress data stream, which includes timestamps, spatial coding identifiers, and pure stress components.

[0074] In practical implementation, a dynamically updated knowledge graph network is constructed, and a topological mechanical model of the basement sidewalls is built. Each wall, corner, and connection point with the floor or roof slab is defined as a node in the knowledge graph network. For example, a wall numbered W1 corresponds to one node. The load transmission path and attenuation relationship between nodes are defined as directed weighted edges, and the edge weights are determined based on structural mechanics calculations, thus completing the structural initialization of the knowledge graph network. In some embodiments, a continuous connection with multiple heterogeneous external databases is established. According to a preset synchronization strategy, typical load patterns such as daily peak load distribution are extracted from historical load records, potential geological disturbance factors such as soil pressure variation rate are analyzed from nearby geological activity reports, and structural details and material parameters such as concrete elastic modulus are obtained from construction archives. This information is then attached as attribute labels or constraint rules to the corresponding nodes and edges in the knowledge graph network. It can be understood that attribute labels include the node's allowable stress threshold, and constraint rules include the attenuation coefficient of the load transmitted by the edge. In practice, each data point in the calibrated stress data stream is mapped to a corresponding node in the knowledge graph network based on its spatial encoding identifier. For example, the spatial encoding identifier "W1-D5" is mapped to the node in the knowledge graph network identified as "wall W1, depth 5 meters". Optionally, the rationality score of the stress data in the local context of the knowledge graph network is calculated by combining the attribute labels and constraint rules currently attached to the node and the mechanical transfer attributes of its associated edges. The rationality score is calculated by comparing the deviation between the current stress value and the average of the node's historical stress patterns, and all contextual factors significantly related to the current stress data are recorded, such as groundwater pressure data within the same time period, thus completing the contextualized analysis.

[0075] In one embodiment of the present invention, see [reference] Figure 3 In the context of a knowledge graph network, a stress evolution simulation process is initiated. This process uses nodes mapped in the knowledge graph network from calibrated stress data streams as stress injection points. Based on the defined mechanical transmission relationships and edge weights in the knowledge graph network, numerical simulations of stress diffusion are performed along directed edges. In each diffusion simulation step, the attenuation and deformation of stress during transmission to the next node are calculated according to parameters such as material damping and connection stiffness defined by the attribute labels of the current node. The stress distribution state obtained from each simulation step is compared in real time with the normal stress threshold range and historical stress patterns defined by auxiliary knowledge at that node. When the simulated stress value continuously exceeds the normal threshold range, or its distribution pattern significantly deviates from the historical pattern, the node position, stress state, and degree of deviation of the current simulation step are recorded, and these abnormal state points are connected in chronological order and transmission direction. The simulation continues until the stress attenuates to a negligible level, or all relevant nodes in the knowledge graph network have been traversed. Finally, the sequence of connected abnormal state points is marked as a complete stress anomaly evolution trajectory.

[0076] The core features of the stress anomaly evolution trajectory are extracted and encapsulated into an abnormal pressure event descriptor. The first point marked as an anomalous state is parsed from the stress anomaly evolution trajectory, and its corresponding knowledge graph network node identifier and timestamp are extracted as the starting spatiotemporal coordinates of the abnormal pressure event. The ratio of stress propagation distance to time interval between adjacent anomalous state points on the stress anomaly evolution trajectory is calculated to obtain the instantaneous diffusion rate of stress on each path segment, and then the average diffusion rate of the entire trajectory is calculated. The changes in the distribution pattern of stress values ​​on the stress anomaly evolution trajectory are analyzed, including the changing trend of stress peak values ​​and the movement path of stress concentration areas. The starting spatiotemporal coordinates, average diffusion rate, changes in diffusion rate, and key parameters of stress distribution pattern changes are organized according to a predefined data structure to generate an abnormal pressure event descriptor containing a unique event identifier.

[0077] In practical implementation, the stress evolution simulation process is initiated within the context of a knowledge graph network. Nodes mapped in the knowledge graph network from the calibrated stress data stream are used as stress injection points. Numerical simulations of stress diffusion are performed along directed edges based on the mechanical transmission relationships and weights defined in the knowledge graph network. In some embodiments, at each diffusion simulation step, the attenuation and deformation of stress during its transmission to the next node are calculated based on the material damping and connection stiffness parameters defined by the current node's own attribute labels. The calculation process uses the following formula:

[0078]

[0079] in: This represents the stress value transmitted from node i into the connecting edge. This represents the stress value transmitted from the connecting edge to the adjacent node j. This represents the attenuation coefficient calculated based on the connection attribute between node i and node j. The stress distribution obtained from each simulation step is calculated by combining the material damping parameters of node i, node j, and the stiffness parameters defined by the connecting edges. It can be understood that the stress distribution obtained from each simulation step is compared in real time with the normal stress threshold range and historical stress patterns defined by auxiliary knowledge on the nodes. The normal stress threshold range is a dynamic interval calculated based on the statistical distribution of the node's historical stress data. When the simulated stress value continuously exceeds the normal threshold range or its distribution pattern deviates from the historical pattern, the node position, stress state, and degree of deviation of the current simulation step are recorded, and these abnormal state points are connected in chronological order and direction of transmission. Optionally, the simulation step size of the pressure evolution deduction process is dynamically determined based on the material wave velocity properties of the edges in the knowledge graph network. The pressure evolution deduction process continues to simulate until the stress decays to a negligible level or all relevant nodes in the knowledge graph network have been traversed. Finally, the sequence of abnormal state points formed by the connection is marked as a complete stress anomaly evolution trajectory.

[0080] In specific implementation, the core features of the stress anomaly evolution trajectory are extracted and encapsulated as an abnormal pressure event descriptor. The first point marked as an abnormal state is parsed from the stress anomaly evolution trajectory, and its corresponding knowledge graph network node identifier and timestamp are extracted as the starting spatiotemporal coordinates of the abnormal pressure event. The ratio of stress propagation distance to time interval between adjacent abnormal state points on the stress anomaly evolution trajectory is calculated to obtain the instantaneous diffusion rate of stress on each path segment. Then, the arithmetic mean of all instantaneous diffusion rates on the entire trajectory is calculated as the average diffusion rate. In some embodiments, the distribution pattern of stress values ​​on the stress anomaly evolution trajectory, including the trend of stress peak value changes, is analyzed. The movement path of the stress concentration region can be understood as follows: the trend of stress peak change is quantified by fitting the first derivative of the peak sequence, and the movement path of the stress concentration region is described by connecting the centroid coordinate sequence of the concentration region. The key parameters of the initial spatiotemporal coordinates, average diffusion rate, changes in diffusion rate, and changes in stress distribution morphology are organized according to a predefined data structure to generate an abnormal pressure event descriptor containing a unique event identifier. Optionally, the predefined data structure includes fields for storing spatial coordinates, time series, rate arrays, and morphological vectors. The abnormal pressure event descriptor exists in the form of a structured data object for transmission between system modules.

[0081] In one embodiment of the invention, an abnormal stress event descriptor is input into a causal inference engine. The causal model of this engine is pre-constructed based on the structure of a knowledge graph network and integrated auxiliary knowledge. Within the framework of the causal model, the causal inference engine enumerates potential causal hypotheses leading to the phenomenon represented by the abnormal stress event descriptor. These potential causal hypotheses include categories such as internal structural damage types, inappropriate external loading, and abrupt changes in environmental factors. The causal inference engine retrieves evidence supporting or refuting each potential causal hypothesis from a calibrated stress data stream, auxiliary knowledge in the knowledge graph network, and a historical abnormal stress event descriptor library. Based on the retrieved evidence, the causal inference engine calculates a confidence score for each potential causal hypothesis and generates a potential causal inference report sorted by probability in descending order based on the scores.

[0082] In specific implementations, the abnormal stress event descriptor is input into a causal reasoning engine. The causal model of the causal reasoning engine is pre-constructed based on the structure of a knowledge graph network and integrated auxiliary knowledge. The causal model is expressed in the form of a directed graph, where nodes represent different state or event variables, and directed edges represent causal relationships between variables. In some embodiments, the causal reasoning engine enumerates potential causal hypotheses leading to the phenomenon represented by the abnormal stress event descriptor within the framework of the causal model. The enumeration process starts with the result state defined by the abnormal stress event descriptor and performs a reverse graph traversal in the causal model. Each upstream cause node encountered during the traversal corresponds to a specific fault or event scenario. The generated potential causal hypotheses include internal structural damage types such as concrete cracking, external improper load effects such as sudden overloading in adjacent areas, and sudden changes in environmental factors such as a sharp drop in groundwater levels. It can be understood that the causal reasoning engine generates a structured object for each potential causal hypothesis, containing a hypothesis identifier, a description of the hypothesis content, and path weight information in the causal model. In its implementation, the causal inference engine retrieves evidence supporting or refuting each potential causal hypothesis from calibrated stress data streams, auxiliary knowledge in the knowledge graph network, and a historical abnormal stress event descriptor library. The retrieval process is based on the relationships defined in the causal model. For example, for the hypothesis of "concrete cracking," it retrieves the most recent stress mutation records and historical cracking case records for the corresponding structural parts; for the hypothesis of "rapid drop in groundwater level," it retrieves data records from hydrological monitoring units within the same time period. Optionally, based on the retrieved evidence, the causal inference engine calculates a confidence score for each potential causal hypothesis using the following formula:

[0083]

[0084] in: This represents the posterior probability, or confidence score, of hypothesis H given evidence E. This represents the likelihood of observing evidence E under the assumption that H is true. This represents the prior probability of hypothesis H. The marginal probability of evidence E is represented by the conditional probability and prior probability parameters required for calculation, which are obtained from statistical information and expert experience rules in the historical case library. In some embodiments, the causal inference engine generates a potential cause inference report sorted by probability in descending order based on the confidence score. The potential cause inference report is presented in list form, and each item in the list contains a description of a potential cause hypothesis, its calculated confidence score, and a list of key evidence summaries supporting or refuting the hypothesis. It can be understood that the evidence summary list is generated directly from the core information fields extracted from the retrieved evidence data.

[0085] In one embodiment of the present invention, the causal inference engine loads a causal model describing the causal relationships between various elements within the basement sidewall system. This model considers pressure anomalies as "result" nodes and structural states, external effects, environmental variables, etc., as possible "cause" nodes, and explicitly defines the direction and intensity of causal influence between nodes. The causal inference engine takes the feature parameters in the abnormal pressure event descriptor as input and traces back through the causal model to the set of all "cause" nodes that can lead to this "result." For each traced "cause" node, it combines the path weight between it and the "result" node with the node's prior probability to generate a specific fault or event scenario described in natural language, constituting a complete potential causal hypothesis.

[0086] The process by which the causal inference engine calculates confidence scores and generates inference reports includes the following steps: For each potential causal hypothesis, relevant direct observation data, historical similar cases, and domain rules are retrieved from the knowledge graph network and external related data. Based on the strength of the causal association between each piece of evidence and the potential causal hypothesis, and the reliability level of the evidence itself, the support or rejection of the hypothesis is calculated. Combining the support and rejection of all evidence, a Bayesian update rule is applied to calculate the posterior probability of the potential causal hypothesis under all currently available evidence, which serves as its confidence score. All potential causal hypotheses are sorted from highest to lowest confidence score, and a summary of key evidence supporting each hypothesis is attached, generating a structured potential causal inference report.

[0087] In specific implementation, the causal reasoning engine enumerates potential causal hypotheses through the following operations: The causal reasoning engine loads a causal model describing the causal relationships between various elements within the basement sidewall system. The causal model is a directed acyclic graph (DAG), where node variables represent states such as pressure anomalies, internal structural damage, improper external loads, and sudden changes in environmental factors. The directed edges in the graph explicitly define the direction and intensity of the causal influence between node variables. For example, there exists a directed edge from the node "sharp drop in groundwater level" to the node "pressure anomaly on the sidewall," and the edge weight represents the intensity of the causal influence. In some embodiments, the causal reasoning engine takes the feature parameters in the abnormal pressure event descriptor as input, such as the initial spatiotemporal coordinates and stress distribution morphology changes. It then traces back in the causal model to the set of all "cause" nodes that can lead to this "result." The tracing process uses a depth-first search algorithm, starting from the "result" node representing the current pressure anomaly state and recursively visiting the upstream "cause" nodes along all incoming edges. For each traced "cause" node, a specific fault or event scenario described in natural language is generated by combining its path weight with the "result" node and the node's prior probability, thus forming a complete potential cause hypothesis. For example, when the node "concrete carbonization leads to steel corrosion" is accessed, the hypothesis "the load-bearing capacity of the steel reinforcement inside wall W1 is reduced due to carbonization and corrosion, causing local stress anomalies" is generated by combining its path weight of 0.75 with the "pressure anomaly" node and the prior probability of the node of 0.05.

[0088] In its implementation, the causal inference engine calculates confidence scores and generates inference reports based on retrieved evidence. This process includes systematic evidence evaluation and probability calculation. For each potential causal hypothesis, it retrieves relevant direct observation data, historical similar cases, and domain rules from knowledge graph networks and external data. The retrieval is based on predefined semantic mapping relationships. For example, for the hypothesis of "adjacent foundation pit excavation," the system retrieves records with the "rapid lateral displacement" feature from construction log databases, vibration sensing data from adjacent areas, and historical abnormal pressure event descriptor libraries. The causal inference engine calculates the support or rejection of each piece of evidence for the potential causal hypothesis based on the strength of the causal association between each piece of evidence and the potential causal hypothesis, as well as the reliability level of the evidence itself. The strength of the causal association is quantified based on the weights of the edges in the causal model. The reliability level of the evidence is predefined by the authority and timeliness of the data source. The formulas for calculating support and rejection are as follows:

[0089]

[0090] in: This indicates the relevance value of the evidence. Indicates the causal relationship strength coefficient. Indicates the reliability level coefficient of the evidence. This symbol represents the consistency between the evidence and the hypothesis. By combining the support and rejection strengths of all evidence, and applying the Bayesian update rule, the posterior probability of the potential causal hypothesis under all available evidence is calculated as its confidence score. Optionally, the Bayesian update uses the aforementioned calculated correlation impact values ​​of each piece of evidence as a likelihood adjustment factor. After calculating the confidence scores of all potential causal hypotheses, the causal inference engine sorts all potential causal hypotheses from highest to lowest confidence score and attaches a summary of key evidence supporting each hypothesis, generating a structured potential causal inference report. The report presents the key summaries in tabular form. See Table 1.

[0091] Table 1: Assessment Table of Evidence Relevance and Reliability

[0092]

[0093] See Figure 4 This study demonstrates the variations in data processing time, data accuracy, and anomaly detection rate across different processing stages in an intelligent stress detection method. Specifically: Signal Acquisition Stage: Data processing time is approximately 2 seconds, data accuracy is approximately 95%, and the anomaly detection rate is at a basic level (close to 0). This stage focuses on quickly acquiring the raw signal and does not involve in-depth analysis. Data Purification Stage: Data processing time increases to 8 seconds, data accuracy rises to approximately 90% (with slight fluctuations), and the anomaly detection rate remains low. This stage focuses on noise removal and signal calibration, without anomaly analysis. Context Analysis Stage: Data processing time jumps to 16 seconds, the anomaly detection rate rapidly increases to approximately 85%, and data accuracy remains stable. In the first stage, with over 90% accuracy, the knowledge graph network is used to contextualize stress data, significantly improving anomaly detection capabilities. In the anomaly deduction stage, data processing time further increases to 22 seconds, with anomaly detection rate approaching 95% and data accuracy stabilizing at around 98%. Through stress diffusion simulation and comparison with historical patterns, the accuracy of anomaly detection is continuously enhanced. In the causal reasoning stage, data processing time reaches 30 seconds, with both data accuracy and anomaly detection rate approaching 100%. This stage completes anomaly feature extraction and causal reasoning, achieving optimal convergence of metrics across the entire process. The diagram clearly illustrates the synergistic change pattern of "increased processing time - stable accuracy - increasing detection rate" in the detection process, reflecting the matching relationship between algorithm complexity and performance metrics at each stage.

[0094] In one embodiment of the present invention, after generating the potential cause inference report, the method further includes steps of knowledge accumulation and system evolution. The current complete processing, including the calibrated stress data stream, abnormal pressure event descriptors, potential cause inference report, and key intermediate data in the causal reasoning process, is packaged into a knowledge case. This knowledge case is stored in a continuously evolving case library, which is indexed in multiple dimensions according to pressure anomaly patterns, causes, and structural locations. All knowledge cases in the case library are periodically clustered and pattern mined to discover recurring combinations of pressure anomalies and causes. These mined patterns of pressure anomalies and causes are used as new empirical rules and fed back into the auxiliary knowledge of the knowledge graph network to enhance the accuracy of subsequent contextualized analysis. They also serve as new prior knowledge to update the causal model of the causal reasoning engine.

[0095] A fixed time period is set to trigger a full scan and analysis of the case library. Anomaly stress event descriptor feature vectors and finally confirmed causal labels for all knowledge cases are extracted from the case library to form a high-dimensional dataset. An unsupervised learning algorithm is used to perform cluster analysis on this high-dimensional dataset, identifying multiple case groups clustered in the feature space. For each case group, the consistency of anomaly features and causes among its internal cases is analyzed, extracting common and stable anomaly patterns and causal association rules. Each extracted association rule is formatted as "IF anomaly pattern THEN possible cause" and assigned an initial confidence weight based on the number and consistency of cases within the case group. The formatted association rules are then used as incremental knowledge to synchronously update the rule base of the knowledge graph network, and the causal strength parameters between corresponding nodes in the causal inference engine are adjusted.

[0096] In practice, after generating the potential cause inference report, the steps of knowledge accumulation and system evolution are performed. The current complete processing process, including the calibrated stress data stream, abnormal pressure event descriptors, potential cause inference reports, and key intermediate data in the causal reasoning process, is packaged into a knowledge case. The knowledge case is stored in the form of a structured file. Its data structure includes a header for recording the case identifier and generation timestamp, and multiple data areas to store the above-mentioned information. The knowledge case is stored in a continuously evolving case library. The case library is indexed in multiple dimensions according to the pressure anomaly pattern, cause, and structural part. For example, the pressure anomaly pattern index is built based on the morphological change parameters in the abnormal pressure event descriptor to create an inverted index, and the cause index is built based on the hypothesis type ranked first in the potential cause inference report. In some embodiments, clustering and pattern mining are performed on all knowledge cases in the case library periodically. A fixed time period is set to trigger a full scan and analysis task of the case library. The abnormal stress event descriptor feature vectors of all knowledge cases in the case library and the finally confirmed causal labels constitute a high-dimensional dataset. An unsupervised learning algorithm is used to perform clustering analysis on the high-dimensional dataset. It can be understood that the unsupervised learning algorithm uses the density-based DBSCAN algorithm to identify multiple case groups clustered in the feature space. For each case group, the consistency of abnormal features and causes among its internal cases is analyzed to extract common and stable abnormal patterns and causal association rules. Each extracted association rule is formatted as "IF abnormal pattern THEN possible cause" and assigned an initial confidence weight based on the number and consistency of cases in the case group. The initial confidence weight is calculated using the formula:

[0097]

[0098] in: This represents the initial confidence weight assigned to the association rule. This indicates the number of cases within a case group. This indicates the total number of cases in the case library. This indicates the consistency rate of causal labels within a case group.

[0099] In practice, the formatted association rules are synchronously updated to the rule base of the knowledge graph network as incremental knowledge, and the causal strength parameters between corresponding nodes in the causal inference engine are adjusted. The rule base of the knowledge graph network is an independent storage module used to store empirical rules in the form of "IF-THEN". The update operation includes inserting new rules into the rule base and attaching their initial confidence weight fields to the rules. The adjustment of the causal model in the causal inference engine is based on the mapping relationship between "abnormal patterns" and "possible causes" defined in the new rules, modifying or adding directed edges between corresponding node variables and updating the weight parameters of the edges. Optionally, the rule base is updated using a version management mechanism. A new rule version is generated after each full analysis. The system uses the latest version of the rule set by default in subsequent contextualized analysis and causal inference. In some embodiments, the knowledge graph network loads empirical rules from the rule base in parallel when loading auxiliary knowledge. When calculating the rationality score of node stress data, in addition to attribute labels and constraint rules, it also matches the abnormal pattern conditions in empirical rules. It can be understood that when the matching degree between real-time data features and the "IF" part of an empirical rule exceeds a threshold, the "THEN" part of the rule will be used as a high-priority clue to influence the conclusion of contextualized analysis. Optionally, when enumerating potential causal hypotheses, the causal inference engine will prioritize searching for empirical rules in the rule base that match the current abnormal stress event descriptor features, and enumerate the "possible causes" pointed to by the rules as high-priority potential causal hypotheses for evidence retrieval. This utilizes historical experience to enhance the accuracy of subsequent contextualized analysis and serves as new prior knowledge to update the causal model of the causal inference engine.

[0100] See Figure 5 In the high-dimensional feature analysis and pattern mining stage of the intelligent pressure detection method for basement sidewalls, heatmaps are used to quantify the linear correlation between the core feature dimensions of abnormal pressure events. The heatmap shows the Pearson correlation coefficient distribution for six core feature dimensions: initial time coordinates (quantized), initial spatial coordinates (quantized), average diffusion rate, diffusion rate change rate, peak stress change, and morphological change parameters. The correlation coefficient between the diagonal positions (feature and itself) is 1.00 (perfectly positive correlation); the coefficients at off-diagonal positions range from -0.09 to 0.21, with the diffusion rate change rate and morphological change parameters showing a correlation coefficient of 0.21 (relatively strong positive correlation). The absolute values ​​of the correlation coefficients between the remaining feature dimensions are all below 0.1, indicating that the linear correlation between most abnormal feature dimensions is weak. The analysis results of this heatmap can support subsequent clustering and pattern mining of high-dimensional datasets: the existence of weakly correlated feature dimensions indicates that the feature space of abnormal pressure events has high independence, requiring unsupervised clustering algorithms (such as DBSCAN) for group division to identify stable abnormal patterns.

[0101] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0102] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A smart pressure detection method suitable for basement sidewalls, characterized in that, include: Multiple sets of sensing units deployed at a specified depth on the side wall of the basement are activated to acquire multi-dimensional physical quantities, including pressure amplitude and direction of action, in a periodic sampling mode, forming a set of sensing signals containing spatial coding identifiers. The sensing signal group is subjected to multi-level signal purification processing to remove the environmental noise substrate, separate the pure stress component representing the stress of the structure, and perform signal attenuation compensation to generate a calibrated stress data stream. The calibrated stress data stream is imported into a dynamically updated knowledge graph network. The knowledge graph network uses the side wall structural components as nodes and the mechanical transmission relationship between components as edges. It continuously obtains historical load records, nearby geological activity reports and construction archives related to the side wall from external databases as auxiliary knowledge to perform contextualized analysis of the calibrated stress data stream. In the context of the knowledge graph network, the stress evolution simulation process is initiated. Starting from the calibrated stress data stream, the diffusion, superposition and attenuation process of stress in the side wall structure is simulated. Through multiple rounds of iterative comparison with the auxiliary knowledge in the knowledge graph network, the abnormal stress evolution trajectory that contradicts the known normal mechanical behavior pattern is identified. The initial spatiotemporal coordinates, diffusion rate, and morphological changes of the stress anomaly evolution trajectory are extracted as core features, and the core features are encapsulated into an abnormal stress event descriptor that can be processed by subsequent decision logic. The process involves importing the calibrated stress data stream into a dynamically updated knowledge graph network. This knowledge graph network uses sidewall structural members as nodes and the mechanical transfer relationships between members as edges. It continuously retrieves historical load records, nearby geological activity reports, and construction archives related to the sidewall from external databases as supplementary knowledge to contextualize the calibrated stress data stream. Specifically, this includes: A topological mechanical model of the basement sidewalls is constructed. Each wall, corner, and connection point with the floor / top slab is defined as a node in the knowledge graph network. The load transmission path and attenuation relationship between nodes are defined as directed weighted edges, thus completing the structural initialization of the knowledge graph network. Establish continuous connections with multiple heterogeneous external databases, extract typical load patterns from the historical load records according to the preset synchronization strategy, analyze potential geological disturbance factors from the adjacent geological activity reports, obtain structural details and material parameters from the construction archives, and attach the information as attribute labels or constraint rules to the corresponding nodes and edges in the knowledge graph network. Each data point in the calibrated stress data stream is mapped to a corresponding node in the knowledge graph network based on its spatial encoding identifier. By combining the attribute labels and constraint rules currently attached to the node, as well as the mechanical transfer properties of its associated edges, the rationality score of the stress data in the local context of the knowledge graph network is calculated, and all contextual factors that are significantly related to the current stress data are recorded to complete the contextualized analysis.

2. The intelligent pressure detection method for basement sidewalls according to claim 1, characterized in that, The process of performing multi-level signal purification processing on the sensed signal group, removing the environmental noise substrate, separating the pure stress component representing the structural stress, and performing signal attenuation compensation to generate a calibrated stress data stream, specifically includes: The sensing signal group is initially decomposed to distinguish the regular environmental interference signals caused by the diurnal temperature cycle and the periodic fluctuation of the groundwater level, as well as the irregular background noise caused by the slight drift of the sensing unit itself, and to establish the dynamic noise spectrum of the current monitoring period. Using the dynamic noise spectrum, the sensing signal group is subjected to composite filtering in the frequency and time domains to filter out the known environmental noise base and obtain a preliminary stress signal containing residual noise. Based on the material aging coefficient and installation time of the sensing unit, a signal attenuation compensation curve is calculated, and the amplitude recovery correction is performed on the initial stress signal. From the corrected preliminary stress signal, based on the characteristic shape and energy distribution of the stress waveform, the response waveform directly related to the load on the side wall structure is extracted, spurious stress signals caused by non-structural reasons are eliminated, and finally aggregated to form the calibrated stress data stream.

3. The intelligent pressure detection method for basement sidewalls according to claim 2, characterized in that, In the context of the knowledge graph network, a stress evolution simulation process is initiated. Starting with the calibrated stress data stream, the diffusion, superposition, and attenuation of stress in the sidewall structure are simulated. Through multiple rounds of iterative comparison with auxiliary knowledge in the knowledge graph network, anomaly evolution trajectories that contradict known normal mechanical behavior patterns are identified. Specifically, this includes: Using the nodes mapped in the knowledge graph network by the calibrated stress data stream as stress injection points, and based on the mechanical transmission relationships and weights defined in the knowledge graph network, numerical simulations of stress diffusion are performed along the directed edges. In each diffusion simulation step, the attenuation and deformation of stress during transmission are calculated based on the material damping and connection stiffness parameters defined by the node's own attribute labels. The stress distribution state obtained from each step of the simulation is compared in real time with the normal stress threshold range and historical stress mode defined by auxiliary knowledge at the node. When the simulated stress value continues to exceed the normal threshold range, or its distribution pattern deviates from the historical pattern, record the node position, stress state and deviation degree of the current simulation step, and connect the abnormal state points in chronological order and transmission direction. The simulation continues until the stress decays to a negligible level, or until all relevant nodes in the knowledge graph network have been traversed. Finally, the sequence of anomalous state points formed by the connection is marked as a complete stress anomaly evolution trajectory.

4. The intelligent pressure detection method for basement sidewalls according to claim 3, characterized in that, The process of extracting the initial spatiotemporal coordinates, diffusion rate, and morphological changes of the stress anomaly evolution trajectory as core features, and encapsulating these core features into an abnormal stress event descriptor that can be processed by subsequent decision-making logic, specifically includes: The first point marked as an abnormal state is parsed from the stress anomaly evolution trajectory, and its corresponding knowledge graph network node identifier and timestamp are extracted as the starting spatiotemporal coordinates of the abnormal pressure event. Calculate the ratio of stress propagation distance to time interval between adjacent abnormal state points on the stress anomaly evolution trajectory to obtain the stress diffusion rate on the path, and calculate the average diffusion rate of the entire trajectory. Analyze the changes in stress distribution morphology along the stress anomaly evolution trajectory, including the changing trend of stress peak values ​​and the movement path of stress concentration areas; The initial spatiotemporal coordinates, the average diffusion rate, the changes in the diffusion rate, and the key parameters of the stress distribution morphology changes are organized according to a predefined data structure to generate the abnormal pressure event descriptor containing a unique event identifier.

5. The intelligent pressure detection method for basement sidewalls according to claim 4, characterized in that, After generating the abnormal stress event descriptor, the method further includes: The abnormal stress event descriptor is input into a causal reasoning engine, and the causal model of the causal reasoning engine is pre-constructed based on the structure and auxiliary knowledge of the knowledge graph network; Within the framework of the causal model, the causal reasoning engine enumerates potential causal hypotheses that lead to the phenomena represented by the abnormal stress event descriptor. These potential causal hypotheses include internal structural damage types, external improper loads, and sudden changes in environmental factors. The causal reasoning engine retrieves evidence from the calibrated stress data stream, auxiliary knowledge in the knowledge graph network, and a library of historical anomalous stress event descriptors to support or refute each potential causal hypothesis. Based on the retrieved evidence, the causal reasoning engine calculates a confidence score for each potential causal hypothesis and generates a potential causal inference report sorted by probability in descending order based on the scores.

6. The intelligent pressure detection method for basement sidewalls according to claim 5, characterized in that, Within the framework of the causal model, the causal inference engine enumerates potential causal hypotheses leading to the phenomena represented by the abnormal stress event descriptor, specifically including: The causal reasoning engine loads a causal model describing the causal relationships between various elements within the basement sidewall system. The causal model considers pressure anomalies as "results," structural states, external actions, and environmental variables as possible "cause" nodes, and defines the direction and intensity of causal influence between nodes. The causal reasoning engine takes the feature parameters in the abnormal stress event descriptor as input and traces back in the causal model to the set of all "cause" nodes that can lead to this "result". For each traced "cause" node, a specific fault or event scenario described in natural language is generated by combining its path weight with the "result" node and the node's prior probability, thus forming a complete potential cause hypothesis.

7. The intelligent pressure detection method for basement sidewalls according to claim 6, characterized in that, Based on the retrieved evidence, the causal inference engine calculates a confidence score for each potential causal hypothesis and generates a potential causal inference report sorted by probability in descending order based on the scores, specifically including: For each of the potential causal hypotheses, retrieve direct observation data, historical similar cases, and domain rules related to the potential causal hypotheses from the knowledge graph network and external data; Based on the strength of the causal relationship between each piece of evidence retrieved and the potential causal hypothesis, and the reliability level of the evidence itself, the degree of support and rejection of the evidence for the potential causal hypothesis is calculated. By combining the support and rejection of all evidence, and using the Bayesian update rule, the posterior probability of the potential cause hypothesis under all available evidence is calculated as its confidence score. All potential causal hypotheses are sorted from highest to lowest confidence score, and a summary of key evidence supporting each hypothesis is attached to generate a structured report of potential causal inferences.

8. The intelligent pressure detection method for basement sidewalls according to claim 7, characterized in that, After generating the potential cause inference report, the method further includes: The current complete processing procedure, including the calibrated stress data stream, the abnormal pressure event descriptor, the potential cause inference report, and the key intermediate data in the causal reasoning process, is packaged into a knowledge case. The knowledge cases are stored in a continuously evolving case library, which is indexed in multiple dimensions according to stress anomaly patterns, causes, and structural parts. Periodically cluster and pattern mining are performed on all knowledge cases in the case library to discover recurring stress anomalies and causal combinations. These stress anomaly and causal combination patterns are then used as new empirical rules and fed back into the auxiliary knowledge of the knowledge graph network to enhance the accuracy of subsequent contextualized analysis. They also serve as new prior knowledge to update the causal model of the causal reasoning engine.

9. The intelligent pressure detection method for basement sidewalls according to claim 8, characterized in that, The process of periodically clustering and pattern mining all knowledge cases in the case library to discover recurring stress anomalies and causal combinations, and then feeding these stress anomaly and causal combination patterns as new empirical rules into the auxiliary knowledge of the knowledge graph network, specifically includes: Set a fixed time period to trigger a full scan and analysis task of the case library; The abnormal stress event descriptor feature vectors of all knowledge cases and the finally confirmed cause labels are extracted from the case library to form a high-dimensional dataset. An unsupervised learning algorithm is used to perform cluster analysis on the high-dimensional dataset to identify multiple case groups that cluster in the feature space; For each case group, analyze the consistency of abnormal characteristics and causes among the cases within it, and extract the common and stable abnormal patterns and cause association rules shared by the case group. Each extracted association rule is formatted as "IF abnormal pattern THEN possible cause" and assigned an initial confidence weight based on the number of cases and consistency in the case group. The formatted association rules are used as incremental knowledge and synchronously updated to the rule base of the knowledge graph network, and the causal strength parameters between corresponding nodes in the causal reasoning engine are adjusted.

Images