Building safety data efficient management method and system based on safety large model

Abstract

This application provides a method and system for efficient management of building safety data based on a large safety model, relating to the field of building safety data management technology. It establishes a baseline for monitoring the expected behavior of sensors over time, identifies persistent deviations between sensor data and the expected behavior baseline based on this baseline, and performs behavioral logic consistency checks on sensor data and associated sensor data based on preset physical association rules. For sensor data identified as having persistent deviations or inconsistent behavioral logic, preprocessing is performed, including reducing data credibility or dynamic numerical correction. Finally, based on the large safety model, weighted learning and risk assessment are performed on the preprocessed target sensor data, enabling the model to intelligently adjust according to data quality and reliability, significantly improving the large safety model's ability to perceive real risks and the accuracy of risk assessment.

Description

Technical Field

[0001] This application relates to the field of building safety data management technology, and more specifically, to an efficient management method and system for building safety data based on a large safety model. Background Technology

[0002] In the field of modern construction engineering, advanced intelligent safety management systems are widely deployed to improve the safety management level of construction sites. These systems aim to achieve intelligent identification, prediction, and early warning of potential risks by integrating and analyzing massive amounts of safety-related data. At their core is typically a large-scale safety model capable of processing data from various heterogeneous sources, including but not limited to images and video data captured by various physical sensors and on-site monitoring cameras, as well as daily inspection reports and abnormal event records entered by safety inspectors. This approach aims to significantly improve the efficiency of data processing and the accuracy of risk assessment through automation and intelligence, thereby achieving "efficient management" of construction safety data.

[0003] However, in real-world long-term operating environments, seemingly insignificant physical factors can have a profound impact on system reliability. For example, anemometers deployed atop tower cranes, exposed to harsh outdoor conditions for extended periods, may experience slow and imperceptible physical aging of their internal mechanical rotating parts or electronic sensing elements. This aging doesn't manifest as sudden sensor failure or data output interruption, but rather as persistent, subtle measurement deviations within specific wind direction and speed ranges. This persistent, systematic underestimation constitutes a hidden flaw at the data source, setting the stage for subsequent data processing and risk assessment.

[0004] Furthermore, to ensure the safety model system can adapt to real-time changes in the construction site environment and maintain the accuracy of its judgments, the system typically incorporates a self-learning and optimization mechanism. This mechanism periodically adjusts model parameters and optimizes internal logic using information generated from recent site operations. During a specific learning cycle, if the construction site experiences prolonged periods of low wind speeds, or if the frequency of tower crane operations under high wind conditions is significantly reduced due to project scheduling adjustments, then aging anemometers will continuously report lower-than-normal wind speeds to the safety model system. In this context, the safety model, during its self-learning process, will be trained based on this systematically biased "historical" information. When processing this information, the model may incorrectly learn a correlation: that under current conditions, wind speed information has a relatively low weight in influencing the overall safety status of the tower crane. This "mislearning" is not a defect in the model itself, but rather an unexpected result produced by its predetermined optimization objective when faced with input information containing hidden biases, thus weakening the model's ability to perceive wind speed risks in the real world.

[0005] This hidden risk becomes apparent when a construction site enters a critical operational phase and weather conditions change abruptly. Suppose on a workday, the wind gradually increases, and the actual wind speed quietly exceeds the safety limits for tower crane operations. However, due to the aging of the anemometer, it still reports a wind speed far lower than the actual value to the safety model system. At this point, when the safety model performs a real-time risk assessment, because its previous learning has reduced its focus on wind speed information, it relies more heavily on other information that appears normal at the time. The model integrates this seemingly "normal" information with its weakened wind speed sensitivity, ultimately determining the current risk level as "overall risk is controllable," failing to identify the imminent high wind speed risk. This misjudgment directly results in the system failing to issue any high wind speed warnings, preventing on-site managers from obtaining timely information about the true wind conditions. Summary of the Invention

[0006] This application provides a method and system for efficient management of building safety data based on a large safety model. It aims to solve the problems that existing building safety management systems may experience continuous data deviation due to physical aging of sensors during long-term operation, and the large safety model may undergo "mislearning" during self-learning due to receiving input information with hidden biases, thereby weakening its ability to perceive real risks and resulting in the inability to accurately warn of high risks at critical moments, which in turn leads to safety accidents.

[0007] On the one hand, this application provides an efficient management method for building safety data based on a large safety model, including:

[0008] Establish a baseline for monitoring the expected behavior of sensors over time;

[0009] Based on the expected behavior baseline, identify persistent deviations between sensor data and the expected behavior baseline;

[0010] Based on preset physical association rules, the sensor data is checked for behavioral logic consistency with at least one associated sensor data;

[0011] The sensor data identified as having the persistent deviation or inconsistent behavior logic is preprocessed to obtain the target sensor data. The preprocessing includes at least one of reducing data credibility and dynamic numerical correction.

[0012] The target sensor data is weighted and risk assessed based on a large safety model.

[0013] Optionally, the step of identifying persistent deviations between sensor data and the expected behavior baseline, based on the expected behavior baseline, includes the following prior steps:

[0014] The raw sensor data is obtained by high-frequency acquisition and preprocessing of the raw signal through the sensor.

[0015] The original sensor data is decomposed intrinsically to obtain the basic signal components with different frequencies and amplitudes;

[0016] The features of the basic signal components are compared with the pre-established degradation mode feature set to identify the nonlinear noise components caused by the internal degradation of the sensor. The degradation mode feature set records the features of the basic signal components after intrinsic decomposition of the data collected by the sensor with different degrees of internal degradation.

[0017] The nonlinear noise component is removed from the raw sensor data to extract clean sensor data, and a health status index is assigned. The health status index is used to perform weighted learning and risk assessment on the sensor data based on a large safety model.

[0018] Optionally, the step of establishing a baseline for monitoring the expected behavior of the sensor over time includes:

[0019] Real-time sensing of multi-dimensional environmental parameters to obtain environmental parameter data, including salt spray concentration, humidity, temperature and corrosive gas concentration;

[0020] The microscopic physical state characteristics inside the sensor are periodically detected to obtain microscopic physical state characteristic values;

[0021] Establish nonlinear correlation rules between environmental parameter data, microscopic physical state characteristic values, and sensor measurement deviations to obtain a set of nonlinear correlation rules;

[0022] Based on the environmental parameter data, the microscopic physical state characteristic values, and the set of nonlinear association rules, a baseline for expected behavior is generated.

[0023] When the expected behavior baseline deviates continuously from the sensor data, the environmental parameter data and the microscopic physical state characteristic values ​​are analyzed, and the expected behavior baseline is corrected according to the set of nonlinear association rules.

[0024] Optionally, the preprocessing of the identified sensor data exhibiting persistent deviation or inconsistent behavioral logic to obtain target sensor data, wherein the preprocessing includes at least one step of reducing data reliability and dynamic numerical correction, includes:

[0025] Obtain historical data reliability records for sensor data;

[0026] When the credibility of a certain sensor data recorded in the sensor data credibility history is continuously lower than a preset threshold for a certain period of time, a credibility reduction risk assessment is triggered.

[0027] Based on the results of the confidence reduction risk assessment, it is determined whether the sensor is in a critical degradation state or has irreversible damage.

[0028] When the sensor is in a critical state of degradation or has irreversible damage, a recommendation is made to replace or repair the sensor, and the weight of the sensor data is temporarily reduced to the minimum, while redundant sensor data or inference based on associated sensor data is enabled.

[0029] Optionally, the step of analyzing the environmental parameter data and the microscopic physical state characteristic values, and correcting the expected behavior baseline according to the set of nonlinear association rules when the expected behavior baseline deviates continuously from the sensor data includes:

[0030] Time series analysis was performed on the environmental parameter data to obtain the evolution trend and abrupt change points of environmental factors;

[0031] Trend analysis of the microscopic physical state characteristics reveals the mode transitions of degradation within the sensor.

[0032] Based on the evolution trend and abrupt change points of the environmental factors, the mode change of the degradation inside the sensor, and the set of nonlinear association rules, the deviation patterns are classified to obtain the classification results.

[0033] Based on the classification results, the weights of the nonlinear association rules in the expected behavior baseline correction are adjusted to obtain the target expected behavior baseline;

[0034] The degree of fit between the obtained target expected behavior baseline and the sensor data is evaluated;

[0035] Based on the degree of fit, the baseline of the target expected behavior is triggered for correction and optimization.

[0036] Optionally, the step of performing time series analysis on the environmental parameter data to obtain the evolution trend and abrupt change points of environmental factors includes:

[0037] The environmental parameter data is decomposed into environmental components at different time scales.

[0038] Trend analysis was performed on the environmental components at different time scales to obtain the evolution trend of environmental factors;

[0039] Anomaly detection was performed on the environmental components at different time scales to obtain the abrupt change points of environmental factors;

[0040] The evolution trend and abrupt change points of the environmental factors are fused to obtain the evolution trend and abrupt change points of the environmental factors.

[0041] Optionally, the step of performing trend analysis on the microscopic physical state characteristic values ​​to obtain the mode transition of degradation within the sensor includes:

[0042] Multidimensional feature extraction is performed on the microscopic physical state feature values ​​to obtain a multidimensional feature vector;

[0043] Real-time correlation analysis is performed on the environmental parameter data to obtain environmental impact factors related to the multidimensional feature vector;

[0044] The multidimensional feature vector is coupled with the environmental impact factor to obtain the degradation pattern characteristics under environmental coupling.

[0045] Based on the degradation mode characteristics under environmental coupling, the degradation modes inside the sensor are dynamically clustered to obtain mode transition points;

[0046] At the mode transition point, the degradation modes inside the sensor are reclassified to obtain the target degradation mode.

[0047] Optionally, the step of performing real-time correlation analysis on the environmental parameter data to obtain environmental impact factors related to the multidimensional feature vector includes:

[0048] Multi-factor interactive analysis was performed on the environmental parameter data to obtain the nonlinear coupling relationship between environmental factors;

[0049] Based on the nonlinear coupling relationship between environmental factors and the multidimensional feature vector, a nonlinear contribution quantification model of environmental factors to the internal degradation mode of the sensor is constructed.

[0050] The nonlinear contribution quantification model is used to calculate and quantify the nonlinear coupling contribution of different environmental factors to the internal degradation mode of the sensor in real time, thereby obtaining the environmental impact factor.

[0051] Optionally, the step of performing multi-factor interactive analysis on the environmental parameter data to obtain the nonlinear coupling relationship between environmental factors includes:

[0052] The corrosion rate and hydrolysis rate of the material surface are calculated in real time based on the salt spray concentration, the humidity, the temperature, and the corrosive gas concentration.

[0053] Based on the material surface corrosion rate, the hydrolysis rate, and thermodynamic principles, the nonlinear coupling relationship between the environmental parameter data is derived in real time.

[0054] When the environmental parameter data changes significantly, the corrosion rate of the material surface and the hydrolysis rate are recalculated, and the nonlinear coupling relationship between the environmental factors is derived in real time.

[0055] On the other hand, this application provides an efficient building safety data management system based on a large safety model, the system comprising:

[0056] The baseline establishment module is used to establish a baseline for monitoring the expected behavior of the sensor over time.

[0057] A deviation identification module is used to identify a persistent deviation between sensor data and the expected behavior baseline based on the expected behavior baseline.

[0058] The consistency verification module is used to perform behavioral logic consistency verification between the sensor data and at least one associated sensor data based on preset physical association rules.

[0059] The data preprocessing module is used to preprocess the sensor data that is identified as having the persistent deviation or having inconsistent behavior logic to obtain the target sensor data. The preprocessing includes at least one of reducing data credibility and dynamic numerical correction.

[0060] The risk assessment module is used to perform weighted learning and risk assessment on the target sensor data based on a large safety model.

[0061] This application discloses a method and system for efficient management of building safety data based on a large-scale safety model. By establishing a baseline for monitoring the expected behavior of sensors over time and identifying persistent deviations between sensor data and the expected behavior baseline, it effectively detects hidden data deviations caused by physical aging of sensors. Simultaneously, by verifying the behavioral logic consistency between sensor data and associated sensor data based on preset physical association rules, it further eliminates logical inconsistencies between data, thereby comprehensively improving the reliability of the data source. For sensor data identified as having persistent deviations or inconsistent behavioral logic, this application performs preprocessing, including reducing data credibility or dynamically correcting numerical values, effectively mitigating the impact of unreliable data on subsequent analysis. Finally, based on the large-scale safety model, weighted learning and risk assessment are performed on the preprocessed target sensor data, enabling the model to intelligently adjust according to data quality and reliability. This avoids "mislearning" caused by receiving input information with hidden biases, thus significantly improving the large-scale safety model's ability to perceive real risks and the accuracy of risk assessment. This method overcomes the problems of difficulty in detecting data deviations caused by sensor aging and the susceptibility of models to being misled by biased data in existing technologies. It effectively solves the technical problem of being unable to accurately warn of high risks at critical moments, and provides a more reliable and intelligent solution for safety management at construction sites. Attached Figure Description

[0062] To illustrate this application more clearly, the accompanying drawings used in the embodiments will be briefly described below. Obviously, those skilled in the art can obtain other drawings based on these drawings without any creative effort.

[0063] Figure 1 The diagram above illustrates a flowchart of an efficient building safety data management method based on a large safety model, as exemplified in this embodiment.

[0064] Figure 2 The diagram above illustrates a structural schematic of an efficient building safety data management system based on a large safety model, as exemplified in the embodiments.

[0065] Figure reference numerals: 100, High-efficiency management system for building safety data based on a large safety model; 10, Baseline establishment module; 20, Deviation identification module; 30, Consistency verification module; 40, Data preprocessing module; 50, Risk assessment module. Detailed Implementation

[0066] The technical solutions of this application will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are merely some embodiments of this application, and not all embodiments. The components of this application described and shown in the accompanying drawings can generally be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of this application provided in the accompanying drawings is not intended to limit the scope of the claimed application, but merely to illustrate selected embodiments of this application. All other embodiments obtained by those skilled in the art based on the embodiments of this application without inventive effort are within the scope of protection of this application.

[0067] It should be noted that similar reference numerals and letters in the following figures indicate similar items; therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures. Furthermore, in the description of this application, terms such as "first," "second," etc., are used only to distinguish descriptions and should not be construed as indicating or implying relative importance.

[0068] In the field of modern construction engineering, advanced intelligent safety management systems are widely deployed to improve the safety management level of construction sites. These systems aim to achieve intelligent identification, prediction, and early warning of potential risks by integrating and analyzing massive amounts of safety-related data. At their core is typically a large-scale safety model capable of processing data from various heterogeneous sources, including but not limited to images and video data captured by various physical sensors and on-site monitoring cameras, as well as daily inspection reports and abnormal event records entered by safety inspectors. This approach aims to significantly improve the efficiency of data processing and the accuracy of risk assessment through automation and intelligence, thereby achieving "efficient management" of construction safety data. However, in actual long-term operating environments, some seemingly insignificant physical factors can have a profound impact on the reliability of the system, causing persistent deviations in sensor data or inconsistent behavioral logic, thus affecting the accuracy of the large-scale safety model's judgments and preventing the system from identifying potential risks in a timely manner.

[0069] like Figure 1 The illustration shows a flowchart of an efficient building safety data management method based on a large safety model, as shown in the example. This application proposes an efficient building safety data management method based on a large safety model, comprising:

[0070] S10, Establish a baseline for monitoring the expected behavior of the sensor over time;

[0071] Sensor data refers to raw or pre-processed data collected and transmitted in real time by various sensors (such as anemometers, thermometers, hygrometers, stress sensors, etc.) deployed on construction sites. This data forms the basis for assessing the health of building structures, environmental conditions, and operational safety. The expected behavior baseline, established through historical data analysis, physical model simulation, or expert experience, describes the expected output pattern or numerical range of sensors under normal operating conditions over time. This baseline is a crucial reference for determining whether sensor data is abnormal.

[0072] S20, Based on the expected behavior baseline, identify a persistent deviation between the sensor data and the expected behavior baseline;

[0073] S30, based on preset physical association rules, perform behavioral logic consistency verification on the sensor data and at least one associated sensor data;

[0074] Physical correlation rules, based on physical laws, engineering specifications, or equipment operating logic, describe the inherent logical relationships or mutual influence mechanisms between different sensor data. Examples include the relationship between tower crane wind speed and tower crane sway amplitude, or the spatial correlation between temperature sensor data from different locations.

[0075] S40, preprocess the identified sensor data with the persistent deviation or the sensor data with inconsistent behavior logic to obtain target sensor data. The preprocessing includes at least one of reducing data credibility and dynamic numerical correction.

[0076] S50, based on a large safety model, performs weighted learning and risk assessment on the target sensor data.

[0077] The safety big data model refers to an artificial intelligence model trained on a large amount of building safety-related data, possessing complex pattern recognition, risk prediction, and decision support capabilities. It can comprehensively analyze multi-source heterogeneous data to assess and issue early warnings about building safety status. The implementation environment of this application is typically an intelligent building safety management platform integrating IoT, cloud computing, and artificial intelligence technologies. This platform can receive, store, process, and analyze massive amounts of sensor data in real time and support the deployment and operation of the safety big data model.

[0078] The method proposed in this application first involves establishing a baseline for monitoring the expected behavior of sensors over time. This baseline is a crucial basis for determining whether sensor data is abnormal. As an optional implementation, historical data from specific types of sensors under different operating conditions is collected over a long period. Statistical methods (such as moving averages, exponential smoothing, etc.) or machine learning algorithms (such as time series prediction models ARIMA, LSTM, etc.) are then used to learn and predict the expected output value or behavior pattern of the sensor at a future point in time. For example, for a wind speed sensor, a predictive model can be built based on historical wind speed data, which can predict the expected wind speed range within a specific season and time period. Another approach is to combine the sensor's physical characteristics and environmental parameters to generate the expected behavior baseline by establishing a physical model. For example, for a temperature sensor, the expected changes in its internal temperature can be predicted using a heat conduction model based on parameters such as ambient temperature and solar radiation intensity.

[0079] Furthermore, based on the aforementioned expected behavior baseline, it is necessary to identify persistent deviations between the sensor data and the expected behavior baseline. As an optional implementation, the real-time acquired sensor data is compared point-by-point or segment-by-segment with the pre-established expected behavior baseline. A persistent deviation is identified when the sensor data consistently exceeds the normal fluctuation range defined by the expected behavior baseline for multiple consecutive sampling points or over a period of time. For example, if the wind speed value reported by the wind speed sensor is lower than the minimum wind speed indicated by the expected baseline for several consecutive hours, a persistent deviation is considered to exist. Another implementation method is to use statistical control charts (such as Shewhart control charts or CUSUM control charts) to monitor the changing trends of the sensor data. When data points consistently fall outside the control limits or exhibit specific non-random patterns, it indicates that the sensor data may have a persistent deviation.

[0080] Subsequently, based on preset physical association rules, the sensor data is checked for behavioral logic consistency with at least one associated sensor data. This check aims to identify any logical inconsistencies between the sensor data. As an optional implementation, a series of physical association rules are predefined, such as "tower crane wind speed sensor data and tower crane sway amplitude sensor data should be positively correlated" and "data from different temperature sensors within the same area should remain consistent within a reasonable range." When real-time sensor data is input, cross-validation is performed according to these rules. For example, if the wind speed sensor reports high wind speed, but the tower crane sway amplitude sensor reports low sway amplitude, there may be a behavioral logic inconsistency. Another implementation method utilizes an expert-knowledge-based rule engine or causal graph model to describe the complex relationships between sensor data. When sensor data does not conform to these preset causal relationships, it is determined to be behavioral logic inconsistent.

[0081] Next, sensor data exhibiting persistent deviations or inconsistent behavioral logic is preprocessed to obtain target sensor data. This preprocessing includes at least one of reducing data confidence and dynamic numerical correction. As an optional implementation, when sensor data is identified as anomalous, its confidence score is reduced. For example, the weight of anomalous data is reduced from 1.0 to 0.5 or lower to minimize its impact on subsequent large-scale safety model assessments. Simultaneously, redundant sensor data or inference based on associated sensor data can be used to compensate for anomalous data. For example, if a wind speed sensor's data is anomalous, data from other nearby wind speed sensors or reasonable wind speed values ​​can be inferred from surrounding environmental parameters. Another implementation involves dynamic numerical correction of anomalous data. For example, if a temperature sensor's data is consistently low, it can be corrected in real-time based on its historical deviation patterns and normal data from associated sensors, using methods such as regression analysis or Kalman filtering, to bring it closer to the true value.

[0082] Finally, a weighted learning and risk assessment of the target sensor data is performed based on a large-scale safety model. The large-scale safety model uses pre-processed target sensor data for learning and assessment. As an optional implementation, after receiving the target sensor data, the large-scale safety model weights the data according to its reliability. Data with high reliability has a larger weight in model learning and risk assessment, while data with low reliability has a correspondingly lower weight. For example, sensor data that has undergone dynamic numerical correction may have a slightly lower weight than completely normal data, but a higher weight than uncorrected abnormal data. The model integrates these weighted data with its internal risk assessment logic to output the current building safety risk level and potential risk points. As another implementation, the large-scale safety model performs pattern recognition and risk prediction based on the characteristics of the target sensor data and historical risk events. For example, the model may identify a correlation between a specific sensor data deviation pattern and historical tower crane overturning accidents, thereby issuing a high-risk warning in advance.

[0083] The overall technical solution of this application aims to address the problem in traditional building safety management where sensor data deviations due to physical aging or environmental influences affect the accuracy of safety model judgments. First, by establishing a baseline of expected behavior, a clear understanding of the normal behavior patterns of sensors can be achieved. When sensor data consistently deviates from this baseline, it indicates a potential sensor anomaly. For example, when a tower crane anemometer continuously reports low wind speeds due to aging, its data will significantly deviate from the preset expected wind speed baseline, thus being identified. Second, by verifying the consistency of behavioral logic, potential logical contradictions between different sensor data can be discovered. For example, if an anemometer reports low wind speeds, but the tower crane's sway amplitude sensor reports high sway amplitudes, this inconsistency will trigger an alarm. These deviations and inconsistencies are hidden defects that are difficult to detect using traditional methods.

[0084] After identifying these anomalous data, this application employs a preprocessing step to reduce the reliability of the anomalous data or dynamically correct its values. For example, for anemometer data that consistently underreports wind speed, its weight in the safety model is reduced, and it may be corrected by referencing other related sensors (such as data from nearby weather stations). Therefore, even if the sensor itself has problems, its output deviation data will not have an excessively negative impact on the safety model, avoiding a decrease in the model's sensitivity to real risks due to "mislearning." Finally, the safety model learns and performs risk assessments based on this preprocessed and weighted target sensor data. Because the data input to the model has been cleaned and corrected, the model can more accurately identify real risk patterns and provide reliable risk assessment results. For example, even if a particular anemometer data point is biased, after preprocessing, the safety model can still integrate other normal data and corrected wind speed data to accurately determine that the actual wind speed has exceeded the safety limit, thereby issuing a timely warning.

[0085] Compared with existing technologies, the core innovation of this application lies in its deep integration of sensor data anomaly identification, preprocessing, and security large-scale model application. Traditional methods often focus on single data anomaly detection or model training optimization, failing to effectively address the profound impact of sensor data source deviations on model accuracy. For example, existing technologies may only use threshold detection to determine whether sensor data is abnormal, but they are difficult to detect persistent, subtle deviations. This application, by establishing a baseline of expected behavior and verifying the consistency of behavioral logic, can more comprehensively and precisely identify hidden defects in sensor data.

[0086] In some embodiments, the step of identifying persistent deviations between sensor data and the expected behavior baseline, based on the expected behavior baseline, includes the following prior steps:

[0087] The raw sensor data is obtained by high-frequency acquisition and preprocessing of the raw signal through the sensor.

[0088] The original sensor data is decomposed intrinsically to obtain the basic signal components with different frequencies and amplitudes;

[0089] The features of the basic signal components are compared with the pre-established degradation mode feature set to identify the nonlinear noise components caused by the internal degradation of the sensor. The degradation mode feature set records the features of the basic signal components after intrinsic decomposition of the data collected by the sensor with different degrees of internal degradation.

[0090] The nonlinear noise component is removed from the raw sensor data to extract clean sensor data, and a health status index is assigned. The health status index is used to perform weighted learning and risk assessment on the sensor data based on a large safety model.

[0091] The process involves high-frequency acquisition of raw signals via sensors to obtain sufficiently dense raw signal data, capturing subtle changes and potential anomalies to ensure the data's temporal resolution meets the requirements of subsequent analysis. Pre-processing of the acquired raw signals, including filtering, noise reduction, sampling rate adjustment, and data normalization, aims to eliminate environmental interference, inherent sensor noise, and other non-critical factors, resulting in cleaner and more standardized raw sensor data.

[0092] Furthermore, signal processing techniques are employed to perform intrinsic mode decomposition on the raw sensor data, such as empirical mode decomposition (EMD), variational mode decomposition (VMD), or wavelet decomposition. This decomposes the complex, nonlinear raw sensor data into a series of fundamental signal components with different frequencies and amplitudes. These fundamental signal components are commonly referred to as intrinsic mode functions (IMFs) or modal components, each representing the oscillation mode of the original signal at different time scales or frequency ranges. The purpose is to decouple the complex signal into simpler, easier-to-analyze components, facilitating subsequent identification of noise components.

[0093] Specifically, specific features of each fundamental signal component are extracted, such as statistical characteristics like energy, frequency center, bandwidth, kurtosis, and kurtosis. These features are then matched against a pre-built database to compare the features of the fundamental signal components with a pre-established set of degradation mode features. This degradation mode feature set can be understood as a knowledge base or model library containing the features of fundamental signal components obtained after intrinsic decomposition of data collected by known sensors under different degradation states. For example, this set can record the energy attenuation, frequency drift, or waveform distortion of specific modal components after intrinsic decomposition of the sensor when it experiences different degrees of internal degradation, such as slight wear, corrosion, or circuit aging. By comparing these features, fundamental signal components that match the sensor's internal degradation mode can be identified, thus classifying them as nonlinear noise components caused by internal sensor degradation. The aim is to accurately distinguish between noise caused by problems within the sensor itself and signals caused by actual changes in the building structure.

[0094] Therefore, the identified nonlinear noise components are removed from the original sensor data, for example through subtraction, reconstruction, or filtering, to obtain clean sensor data that contains no or significantly reduces internal degradation noise, thus achieving the removal of the nonlinear noise components. Simultaneously, a health status index is assigned, which can be understood as assigning a quantified health score to the sensor data based on the degree of internal degradation. For example, the health status index can be a value between 0 and 1, where 1 indicates the sensor is in a fully healthy state, and 0 indicates the sensor is completely failed. The assignment of the health status index can be based on a comprehensive evaluation of factors such as the comparison results of degradation mode feature sets, the intensity of noise components, or historical degradation trends. The purpose is to provide an important reference for subsequent weighted learning and risk assessment of sensor data by a large-scale safety model, enabling the model to intelligently adjust according to the reliability of the data.

[0095] The technical solution of this application effectively solves the problem of data distortion caused by internal sensor degradation by introducing a deep evaluation and correction mechanism for sensor data quality before identifying persistent deviations between sensor data and the expected behavior baseline. Specifically, high-frequency acquisition and preprocessing ensure the integrity and usability of the original signal; intrinsic decomposition breaks down the complex original signal into easily analyzable basic signal components, providing a basis for noise identification; by comparing with the feature set of degradation patterns, nonlinear noise components generated by internal sensor degradation can be accurately located and identified, avoiding misjudging sensor problems as building structure anomalies. Subsequently, these noise components are removed, allowing subsequent deviation identification to be based on cleaner and more reliable sensor data, significantly improving the accuracy of identification. At the same time, assigning a health status index provides a quantitative basis for the reliability of sensor data for the safety model, enabling the large model to adjust data weights according to the actual health status of the sensor when performing weighted learning and risk assessment, thereby avoiding misjudgments caused by low-quality data and ensuring the objectivity and effectiveness of risk assessment.

[0096] Through the above technical solutions, this application can significantly improve the accuracy and reliability of building safety data management methods. Compared with directly using potentially damaged sensor data for deviation identification, this application ensures that the sensor data used for analysis is pure and accurate by pre-identifying and removing nonlinear noise caused by internal sensor degradation, thereby avoiding false alarms or missed alarms due to sensor malfunctions. Furthermore, the introduction of a health status index enables the safety model to intelligently adjust the weights of different sensor data during risk assessment, effectively reducing the negative impact of low-quality data on the overall assessment results. This makes the risk assessment results more accurate and objective, providing a more solid data foundation and decision-making basis for building safety management.

[0097] In some alternative embodiments, it is assumed that a vibration sensor in a building structure, due to long-term operation, experiences slight fatigue in its internal crystal element, resulting in weak nonlinear noise superimposed on the vibration signal it collects. Directly using this sensor data for deviation identification might lead to misjudgment as abnormal vibration in the structure. The technical solution of this application first obtains the original vibration signal through high-frequency acquisition and then performs intrinsic mode function decomposition, decomposing the signal into multiple intrinsic mode functions (IMFs). Subsequently, the characteristics of these IMFs (e.g., energy, frequency distribution) are compared with a pre-established set of degradation mode features containing different degrees of crystal fatigue characteristics. For example, a support vector machine (SVM) classifier identifies specific IMFs related to crystal fatigue as nonlinear noise components. Next, these noise IMFs are removed from the original vibration signal to obtain cleaner vibration data. Simultaneously, based on the identified degree of degradation, a low health status index, such as 0.7, is assigned to the sensor. In subsequent weighted learning and risk assessment of the large safety model, the clean data from this sensor will participate in the calculation with a weight of 0.7, while the weight of other health sensors may be 1. This automatically reduces the impact of the sensor data on the final result when using it for risk assessment, avoids misjudgment caused by sensor degradation, and ensures the accuracy of risk assessment.

[0098] In some embodiments, the step of establishing a baseline for monitoring the expected behavior of the sensor over time includes:

[0099] Real-time sensing of multi-dimensional environmental parameters to obtain environmental parameter data, including salt spray concentration, humidity, temperature and corrosive gas concentration;

[0100] The microscopic physical state characteristics inside the sensor are periodically detected to obtain microscopic physical state characteristic values;

[0101] Establish nonlinear correlation rules between environmental parameter data, microscopic physical state characteristic values, and sensor measurement deviations to obtain a set of nonlinear correlation rules;

[0102] Based on the environmental parameter data, the microscopic physical state characteristic values, and the set of nonlinear association rules, a baseline for expected behavior is generated.

[0103] When the expected behavior baseline deviates continuously from the sensor data, the environmental parameter data and the microscopic physical state characteristic values ​​are analyzed, and the expected behavior baseline is corrected according to the set of nonlinear association rules.

[0104] Specifically, various environmental sensors deployed in the built environment continuously collect environmental data related to the safety of building structures or equipment. These multi-dimensional environmental parameters, such as salt spray concentration, humidity, temperature, and corrosive gas concentration, can directly or indirectly reflect the severity of the sensor's operating environment and its impact on sensor performance. Environmental parameter data can be understood as these real-time collected environmental measurements, the purpose of which is to provide external environmental condition inputs for subsequent baseline generation and calibration.

[0105] This involves periodically acquiring physical property data of the materials, structure, or electronic components inside the sensor using non-invasive or minimally invasive testing techniques, thus enabling periodic detection of the sensor's internal microscopic physical state characteristics. These microscopic physical state characteristics may include, but are not limited to, changes in material resistivity, dielectric constant, surface morphology, internal stress, or fatigue damage, directly reflecting the sensor's health status and potential degradation trends. The aim is to more accurately assess the sensor's true performance under specific environmental conditions and distinguish between reading changes caused by environmental factors and those caused by sensor degradation itself.

[0106] In practical applications, mathematical models or rule sets are constructed using methods such as historical data analysis, physical modeling, or machine learning. These models describe the complex relationships between environmental factors, the internal state of the sensor, and the deviation of the sensor's actual measurements from its ideal behavior. This allows for the establishment of nonlinear correlation rules between environmental parameter data, microscopic physical state characteristics, and sensor measurement deviations. These sets of nonlinear correlation rules can capture the nonlinear coupling effects between multiple factors. For example, in high-temperature and high-humidity environments, the impact of corrosive gas concentrations on sensor performance may be amplified. The aim is to accurately predict the sensor's expected behavior and potential measurement deviations under different environmental and internal conditions.

[0107] Therefore, by using real-time acquired environmental parameter data and periodically detected microscopic physical state characteristic values ​​as inputs, and combining them with an established set of nonlinear correlation rules, the theoretical expected output range or value of the sensor under current conditions is dynamically calculated and generated, thus generating an expected behavior baseline. This expected behavior baseline is dynamically changing and can adapt to changes in the environment and the sensor's own state, thereby more accurately reflecting the sensor's "normal" behavior.

[0108] Furthermore, when the expected behavior baseline deviates persistently from the sensor data, the environmental parameter data and the microscopic physical state characteristic values ​​are analyzed, and the expected behavior baseline is corrected according to the set of nonlinear association rules. This means that when there is a significant and persistent difference between the actual sensor data and the dynamically generated expected behavior baseline, a feedback mechanism is activated. By re-examining the current environmental parameter data and microscopic physical state characteristic values, and combining them with the set of nonlinear association rules, it is possible to determine whether this deviation is caused by environmental changes, internal sensor degradation, or a genuine abnormal event, and accordingly adjust or optimize the expected behavior baseline to better fit the actual operating state of the sensor, ensuring the accuracy and adaptability of the baseline.

[0109] The technical solution of this application constructs a highly adaptive and accurate baseline for expected behavior by introducing multi-dimensional environmental parameters and the microscopic physical state characteristics of the sensor, and establishing nonlinear correlation rules between them and sensor measurement deviations. Specifically, real-time sensing of environmental parameter data enables the baseline to dynamically respond to changes in the external environment, avoiding misjudgments caused by environmental factors; periodic detection of microscopic physical state characteristic values ​​reveals the sensor's own degradation trend, taking into account the influence of internal sensor factors on the readings. Through a set of nonlinear correlation rules, the complex interactions between these multiple factors can be understood, thereby generating an expected baseline that more closely reflects the sensor's actual behavior. When the actual sensor data deviates continuously from this baseline, these environmental and internal state information, combined with the nonlinear correlation rules, can be used to intelligently correct the baseline, ensuring that the baseline always accurately reflects the normal operating state of the sensor and effectively distinguishes between real anomalies and normal fluctuations caused by environmental or self-degradation.

[0110] Through the above technical solution, this application can significantly improve the accuracy and adaptability of the expected behavior baseline. Compared with baselines that rely solely on historical data or simple models, the technical solution of this application can dynamically consider the impact of environmental changes and internal sensor degradation on sensor readings, thereby effectively reducing false alarm and false negative rates. Consequently, the identification of deviations between sensor data and the expected behavior baseline will be more accurate, providing higher-quality input data for subsequent weighted learning and risk assessment of large-scale safety models, thus improving the reliability and accuracy of the entire building safety data management system. Furthermore, through continuous baseline correction, it can better adapt to the long-term evolution of the building environment and the aging process of sensors, extending the effective service life of sensors and providing a more scientific basis for maintenance and replacement decisions.

[0111] In some embodiments, the preprocessing of the identified sensor data exhibiting the persistent deviation or inconsistent behavioral logic to obtain target sensor data, wherein the preprocessing includes at least one step of reducing data reliability and dynamic numerical correction, includes:

[0112] Obtain historical data reliability records for sensor data;

[0113] When the credibility of a certain sensor data recorded in the sensor data credibility history is continuously lower than a preset threshold for a certain period of time, a credibility reduction risk assessment is triggered.

[0114] Based on the results of the confidence reduction risk assessment, it is determined whether the sensor is in a critical degradation state or has irreversible damage.

[0115] When the sensor is in a critical state of degradation or has irreversible damage, a recommendation is made to replace or repair the sensor, and the weight of the sensor data is temporarily reduced to the minimum, while redundant sensor data or inference based on associated sensor data is enabled.

[0116] Specifically, the sensor data reliability history refers to a continuous record of the reliability of each sensor's data, collected and stored over a long period. This record may include timestamps, sensor identifiers, reliability scores, and information on any events that cause changes in reliability. This history can be obtained through real-time calculation and recording of indicators such as the stability of sensor output, deviation from the expected baseline, and consistency with other related sensor data. The preset threshold can be set according to the sensor type, importance, and security requirements of the application scenario. For example, it can be set to 0.7 or 0.8, indicating that when the reliability is below this value, the data reliability is questionable. The specific time length refers to the period during which the reliability remains below the preset threshold. For example, it can be set to 24 or 48 consecutive hours to avoid misjudgments triggered by instantaneous fluctuations.

[0117] When the above conditions are met, a credibility reduction risk assessment will be triggered. This assessment can employ machine learning models, expert systems, or statistical analysis methods, comprehensively considering various factors such as sensor type, installation location, environmental parameter data, microscopic physical state characteristics, and historical fault data to conduct an in-depth analysis of the sensor's current state. Based on the risk assessment results, it can be determined whether the sensor is in a critical degradation state or has suffered irreversible damage. A critical degradation state means that the sensor's performance has significantly decreased, but it has not yet completely failed, and there is still a possibility of repair or adjustment; irreversible damage indicates that the sensor has suffered structural damage and cannot be restored to normal function through simple repair.

[0118] When a sensor is determined to be in a critical degradation state or suffer irreversible damage, an automatic recommendation to replace or repair the sensor will be issued. This recommendation can be sent to the maintenance personnel's terminal device for timely intervention. Simultaneously, to ensure the integrity of the data chain and the accuracy of risk assessment, the weight of the sensor's data will be temporarily reduced to a minimum. This means that the sensor's data will have a significantly weakened influence in subsequent weighted learning and risk assessment of the large-scale security model. Furthermore, to compensate for the information loss caused by the decreased reliability of the sensor data, redundant sensor data will be enabled or inference based on associated sensor data will be performed. Enabling redundant sensor data means that if a backup sensor measuring the same or similar physical quantities exists, its data will be used preferentially. Inference based on associated sensor data means that data from other sensors that have a physical or logical connection with the sensor will be used, through data fusion, pattern recognition, or predictive models, to estimate and supplement the missing or unreliable data of the sensor.

[0119] The technical solution of this application establishes a historical record of sensor data reliability and triggers risk assessments based on preset thresholds and specific time lengths, enabling dynamic and intelligent identification of potential sensor degradation or damage issues. This shifts the approach from passively reducing data reliability to proactively assessing sensor health and implementing tiered response measures based on the assessment results. By determining whether a sensor is in a critical degradation state or suffers irreversible damage and issuing corresponding repair or replacement recommendations, preventative maintenance of the building safety monitoring system can be achieved. Simultaneously, by reducing the data weight of problematic sensors and enabling redundant or inferred data, the data quality input to the large-scale safety model is ensured, avoiding bias in the overall risk assessment due to a single sensor failure.

[0120] The aforementioned technical solution significantly improves the intelligence and reliability of building safety data management. This solution not only promptly detects and addresses sensor health issues, effectively preventing data distortion caused by sensor malfunctions, but also provides cleaner and more reliable input data for the safety model. This ensures that the safety model can perform accurate weighted learning and risk assessment based on high-quality data, greatly enhancing the overall robustness and scientific rigor of the building safety monitoring system and reducing potential safety risks.

[0121] In some optional embodiments, assume a vibration sensor is deployed in a building structural health monitoring system, whose main task is to monitor the frequency and amplitude of structural vibrations. The historical reliability data of this vibration sensor is continuously recorded. During a certain period, due to slight aging of the sensor's internal components, its output data begins to drift slightly, causing the calculated data reliability to remain below a preset threshold of 0.7 for 48 hours. At this point, a reliability reduction risk assessment is automatically triggered. This assessment module, combining information such as the sensor's model, years of use, ambient temperature, and humidity, analyzes the data using a pre-trained machine learning model and determines that the sensor is in a critical degradation state. Based on this determination, a notification is sent to maintenance personnel suggesting "checking and calibrating or replacing the vibration sensor." Simultaneously, to avoid affecting real-time monitoring and risk assessment, the data weight of the vibration sensor is temporarily reduced to a minimum, for example, from 1.0 to 0.1. If another redundant vibration sensor is deployed in the area, the data from the redundant sensor will be automatically switched or fused for analysis. If there is no redundant sensor, the data value that the vibration sensor should have will be inferred based on the data from other nearby related sensors (such as strain sensors and acceleration sensors) through a data fusion algorithm. This ensures that the safety model can continuously obtain relatively accurate structural vibration information, thereby maintaining effective monitoring of the building structure safety.

[0122] In some embodiments, the step of analyzing the environmental parameter data and the microscopic physical state characteristic values, and correcting the expected behavior baseline according to the set of nonlinear association rules when the expected behavior baseline deviates persistently from the sensor data includes:

[0123] Time series analysis was performed on the environmental parameter data to obtain the evolution trend and abrupt change points of environmental factors;

[0124] Trend analysis of the microscopic physical state characteristics reveals the mode transitions of degradation within the sensor.

[0125] Based on the evolution trend and abrupt change points of the environmental factors, the mode change of the degradation inside the sensor, and the set of nonlinear association rules, the deviation patterns are classified to obtain the classification results.

[0126] Based on the classification results, the weights of the nonlinear association rules in the expected behavior baseline correction are adjusted to obtain the target expected behavior baseline;

[0127] The degree of fit between the obtained target expected behavior baseline and the sensor data is evaluated;

[0128] Based on the degree of fit, the baseline of the target expected behavior is triggered for correction and optimization.

[0129] This study employs statistical or machine learning methods, such as Autoregressive Integral Moving Average (ARIMA), Long Short-Term Memory (LSTM) networks, or wavelet transform, to process multi-dimensional environmental parameter data. This process identifies long-term trends in environmental factors (e.g., seasonal temperature variations, slow increases in humidity) and abrupt changes (e.g., sudden increases in corrosive gas concentrations), enabling time-series analysis of the environmental parameter data. The aim is to comprehensively understand the dynamic impact of the external environment on sensor performance.

[0130] By continuously monitoring and analyzing the microscopic physical characteristics of the sensor (e.g., impedance, capacitance, material fatigue, etc.), trend analysis of these characteristics can be performed to reveal specific modes of degradation within the sensor. This may include identifying different stages from normal operation to initial degradation, accelerated degradation, and even the period before critical failure, with the aim of gaining a deeper understanding of the sensor's health status and potential failure mechanisms.

[0131] By combining the above analysis results with pre-established nonlinear correlation rules between environmental parameter data, microscopic physical state characteristics, and sensor measurement deviations, and using pattern recognition or clustering algorithms, persistent deviations between sensor data and the expected behavioral baseline are categorized into specific deviation patterns, thus achieving pattern classification. For example, deviations may be classified as "corrosive degradation caused by excessively high environmental humidity," "drift caused by aging of internal sensor components," or "transient failure caused by external impact," etc. The aim is to accurately diagnose the root cause of the deviation.

[0132] Once the deviation pattern is accurately classified, the contribution or weight of each rule in the set of nonlinear association rules in the expected behavior baseline correction model is dynamically adjusted based on the classification result to obtain the target expected behavior baseline. For example, if the deviation is diagnosed as being mainly caused by humidity, the nonlinear association rules related to humidity will be given higher weights. Thus, the generated target expected behavior baseline can more accurately reflect the expected behavior under the current environmental and sensor conditions.

[0133] After generating the baseline of the expected target behavior, the accuracy of the fit is quantified by calculating the difference between this baseline and the actual collected sensor data, such as the mean square error (MSE), root mean square error (RMSE), or correlation coefficient, to evaluate the degree of fit between the obtained baseline of the expected target behavior and the sensor data. The purpose is to objectively measure the correction effect.

[0134] If the assessed fit does not meet the preset satisfaction standard (e.g., the error exceeds a certain threshold), a further correction and optimization process will be automatically triggered, thus initiating correction and optimization of the target expected behavior baseline. This may include readjusting the weights of nonlinear association rules, refining deviation pattern classifications, or exploring other correction strategies until the target expected behavior baseline can adequately fit the actual sensor data. The aim is to ensure the continued effectiveness and accuracy of baseline correction.

[0135] The technical solution of this application, through time-series analysis of environmental parameter data, can dynamically capture the evolution trend and abrupt change points of environmental factors, thereby gaining a more comprehensive understanding of the impact of external conditions on sensor performance. Simultaneously, trend analysis of microscopic physical state characteristic values ​​allows for the identification of internal degradation mode changes within the sensor, revealing deep-seated changes in the sensor's own health status. Therefore, by combining the dynamic changes of environmental factors and the internal degradation modes of the sensor, and utilizing a pre-defined set of nonlinear association rules, the deviation patterns between sensor data and the expected behavior baseline can be finely classified. This classification result provides clear guidance for subsequent baseline correction, enabling the weights of the nonlinear association rules in the correction process to be adaptively adjusted according to the specific causes of deviation, thus generating a target expected behavior baseline that better reflects the current situation. Finally, by evaluating the degree of fit between the target expected behavior baseline and the sensor data, the effectiveness of the correction is ensured, and further optimization is triggered when the fit is poor, forming a closed-loop, adaptive baseline correction mechanism.

[0136] Through the above technical solution, this application overcomes the problem of insufficient adaptability of traditional baseline correction methods when facing complex and variable environments and sensor internal degradation. Specifically, by introducing time series analysis and trend analysis, environmental impacts and sensor degradation can be more accurately identified and quantified, thereby improving the accuracy of deviation pattern classification. This refined classification further supports the dynamic adjustment of nonlinear association rule weights, enabling the expected behavior baseline to respond more flexibly and accurately to changes in actual working conditions, avoiding lag or misjudgment caused by static correction models. Therefore, this application significantly improves the adaptability and correction accuracy of the expected behavior baseline, providing a more reliable and efficient foundation for building safety data management, effectively reducing false alarm and false negative rates, and ensuring the safe operation of building structures and equipment.

[0137] In some embodiments, the step of performing time series analysis on the environmental parameter data to obtain the evolution trend and abrupt change points of environmental factors includes:

[0138] The environmental parameter data is decomposed into environmental components at different time scales.

[0139] Trend analysis was performed on the environmental components at different time scales to obtain the evolution trend of environmental factors;

[0140] Anomaly detection was performed on the environmental components at different time scales to obtain the abrupt change points of environmental factors;

[0141] The evolution trend and abrupt change points of the environmental factors are fused to obtain the evolution trend and abrupt change points of the environmental factors.

[0142] This process involves decomposing the raw environmental parameter data into multiple components with specific frequencies and amplitudes at different time scales, achieving multi-scale decomposition of the environmental parameter data. For example, methods such as wavelet transform, empirical mode decomposition (EMD), or ensemble empirical mode decomposition (EEMD) can be used to decompose the environmental parameter data into high-frequency, mid-frequency, and low-frequency components, representing short-term fluctuations, periodic changes, and long-term trends, respectively. The aim is to reveal the intrinsic structure and changing patterns of environmental factors at different time scales.

[0143] Furthermore, for the decomposed environmental components at various scales, statistical or machine learning methods, such as moving average, linear regression, support vector regression (SVR), or Gaussian process regression, are used to extract and quantify their long-term or short-term evolution directions, enabling trend analysis of environmental components at different time scales. The aim is to accurately capture the gradual changes in environmental factors, such as seasonal increases in temperature or long-term accumulation of corrosive gas concentrations.

[0144] Furthermore, by setting statistical thresholds, using machine learning-based anomaly detection models (such as Isolation Forest, LOF, or One-Class SVM), or rule-based methods, data points or time periods that significantly deviate from normal patterns can be identified, enabling anomaly detection of environmental components at different time scales. These anomalies may represent sudden events in the environment, such as a sudden increase in salt spray concentration or an accidental leak of corrosive gas concentration. The aim is to promptly detect abnormal environmental conditions that may have an immediate impact on sensor performance.

[0145] Ultimately, the gradual change information obtained through trend analysis is integrated with the abrupt change information obtained through anomaly detection to fuse the evolutionary trends and abrupt change points of environmental factors. This fusion can be achieved through weighted averaging, decision trees, neural networks, or other fusion algorithms to form a comprehensive and accurate description of the evolutionary trends and abrupt change points of environmental factors. The aim is to provide a complete view that reflects both long-term environmental changes and captures short-term anomalies, providing more accurate input for subsequent baseline correction of expected behaviors.

[0146] The technical solution of this application, through refined time-series analysis of environmental parameter data, enables a comprehensive and in-depth understanding of the dynamic changes of environmental factors. Specifically, multi-scale decomposition allows for the examination of environmental data from both macroscopic and microscopic perspectives, avoiding the one-sidedness that may result from single-time-scale analysis. Based on this, trend analysis captures the long-term evolution patterns of environmental factors, while anomaly detection focuses on identifying sudden events in the environment. By effectively integrating information from these different levels, this technical solution can construct a more robust and comprehensive dynamic model of environmental factors. Therefore, when a persistent deviation occurs between the expected behavioral baseline and sensor data, the environmental parameter data can be analyzed more accurately to distinguish whether it is due to gradual environmental changes or sudden events, thus providing a more precise basis for subsequent correction of the expected behavioral baseline based on a set of nonlinear association rules.

[0147] The aforementioned technical solutions significantly enhance the depth and breadth of environmental parameter data analysis. Multi-scale decomposition enables the identification and separation of environmental impacts at different time scales, leading to a more accurate understanding of the mechanisms by which the environment affects sensor behavior. The combination of trend analysis and anomaly detection ensures that both slow environmental degradation and sudden environmental shocks can be identified promptly and effectively. This comprehensive analytical capability allows for more precise localization of environmental factors when the expected behavior baseline deviates from sensor data, thereby improving the accuracy and response speed of expected behavior baseline correction. This not only helps maintain the high reliability of sensor data but also provides more reliable foundational data for risk assessment based on large-scale safety models, ultimately enhancing the intelligence level and early warning capabilities of the entire building safety data management approach.

[0148] In some embodiments, the step of performing trend analysis on the microscopic physical state characteristic values ​​to obtain the mode transition of degradation within the sensor includes:

[0149] Multidimensional feature extraction is performed on the microscopic physical state feature values ​​to obtain a multidimensional feature vector;

[0150] Real-time correlation analysis is performed on the environmental parameter data to obtain environmental impact factors related to the multidimensional feature vector;

[0151] The multidimensional feature vector is coupled with the environmental impact factor to obtain the degradation pattern characteristics under environmental coupling.

[0152] Based on the degradation mode characteristics under environmental coupling, the degradation modes inside the sensor are dynamically clustered to obtain mode transition points;

[0153] At the mode transition point, the degradation modes inside the sensor are reclassified to obtain the target degradation mode.

[0154] Specifically, multidimensional feature extraction is performed on the microscopic physical state characteristic values ​​to obtain multidimensional feature vectors. These microscopic physical state characteristic values ​​may include, but are not limited to, the sensor's resistance, capacitance, mechanical vibration frequency, and changes in the material's microstructure (such as crack propagation and corrosion degree). By employing various signal processing and feature engineering techniques such as Principal Component Analysis (PCA), Independent Component Analysis (ICA), wavelet transform, and Fourier transform, multidimensional feature vectors that effectively characterize the sensor's internal state are extracted from the original microscopic physical state characteristic values. These feature vectors aim to capture key information about sensor degradation, such as degradation rate, degree of degradation, and type of degradation.

[0155] Real-time correlation analysis is performed on the environmental parameter data to obtain environmental impact factors related to the multidimensional feature vector. These environmental parameter data include salt spray concentration, humidity, temperature, and corrosive gas concentration. By establishing a real-time correlation model between the environmental parameter data and the multidimensional feature vector, using methods such as multiple regression analysis, mutual information analysis, and Granger causality tests, environmental factors that significantly influence the sensor's internal degradation mode are identified. These environmental factors are quantified as environmental impact factors to characterize their contribution and mechanism of action in the sensor degradation process.

[0156] By coupling the multidimensional feature vectors with the environmental influencing factors, degradation mode characteristics under environmental coupling are obtained. In practical applications, sensor degradation is not independent of the environment, but rather has a complex nonlinear coupling relationship with environmental factors. Therefore, deep coupling analysis is performed on the extracted multidimensional feature vectors and the environmental influencing factors obtained from real-time correlation analysis. For example, by constructing deep learning models (such as recurrent neural networks (RNNs) or long short-term memory networks (LSTMs)) or coupling algorithms based on physicochemical models, the true characteristics of the sensor's internal degradation modes under specific environmental conditions can be revealed. This allows for a more comprehensive and accurate analysis of degradation mode characteristics under environmental coupling.

[0157] Based on the degradation mode characteristics under environmental coupling, dynamic clustering is performed on the degradation modes inside the sensor to obtain mode transition points. Furthermore, using the obtained degradation mode characteristics under environmental coupling, dynamic clustering algorithms (such as K-means, DBSCAN, Gaussian Mixture Model, GMM, etc.) are employed to classify the degradation modes inside the sensor. Dynamic clustering can monitor changes in degradation modes in real time and identify mode transition points. Mode transition points represent the critical moments in the sensor's internal degradation process, transitioning from one state to another, such as moving from initial degradation to accelerated degradation, or from reversible damage to irreversible damage.

[0158] At the mode transition point, the internal degradation modes of the sensor are reclassified to obtain the target degradation mode. The detection of a mode transition point indicates a potential change in the degradation mechanism or dominant factor within the sensor. Therefore, at this point, the internal degradation modes of the sensor need to be reassessed and reclassified. By updating the clustering model or adjusting the classification rules, a more accurate and real-time target degradation mode can be obtained, providing a more precise basis for subsequent risk assessment and maintenance decisions.

[0159] The technical solution of this application can comprehensively capture subtle changes in sensor internal degradation by extracting multidimensional features from microscopic physical state characteristics. Subsequently, by real-time correlation analysis of environmental parameter data, environmental influencing factors that significantly affect sensor degradation are identified, thus avoiding the problem of confusing environmental impacts with sensor degradation itself. Furthermore, by coupling the multidimensional feature vectors with environmental influencing factors, a degradation mode feature under environmental coupling is constructed. This makes the understanding of sensor degradation modes deeper and more accurate, reflecting the complex degradation mechanisms under real-world operating conditions. Based on these degradation mode features under environmental coupling, dynamic clustering can identify the transition points of sensor internal degradation modes in real time and accurately, such as key stages from normal operation to initial degradation, and then to accelerated degradation. Reclassification at the mode transition points ensures accurate judgment of the sensor's current degradation state, thereby enabling timely adjustment of subsequent risk assessment strategies.

[0160] Through the above technical solution, this application can analyze the microscopic physical state characteristics inside the sensor more precisely and fully consider the complex influence of environmental factors on sensor degradation, thereby improving the accuracy and real-time performance of identifying changes in sensor internal degradation modes. This method can not only detect potential degradation risks of sensors earlier, but also distinguish between apparent anomalies caused by environmental changes and real anomalies caused by the sensor's own internal degradation, avoiding misjudgments. Therefore, it can provide more reliable sensor health status assessments for building safety management, support more accurate predictive maintenance and risk warnings, effectively extend sensor lifespan, and reduce building safety risks caused by sensor failures.

[0161] In some embodiments, the step of performing real-time correlation analysis on the environmental parameter data to obtain environmental impact factors related to the multidimensional feature vector includes:

[0162] Multi-factor interactive analysis was performed on the environmental parameter data to obtain the nonlinear coupling relationship between environmental factors;

[0163] Based on the nonlinear coupling relationship between environmental factors and the multidimensional feature vector, a nonlinear contribution quantification model of environmental factors to the internal degradation mode of the sensor is constructed.

[0164] The nonlinear contribution quantification model is used to calculate and quantify the nonlinear coupling contribution of different environmental factors to the internal degradation mode of the sensor in real time, thereby obtaining the environmental impact factor.

[0165] Specifically, advanced data analysis techniques, such as multivariate statistical analysis, machine learning algorithms (e.g., decision trees, random forests, neural networks), or physical model-based simulations, are used to delve into the complex interactions between environmental parameters such as salt spray concentration, humidity, temperature, and corrosive gas concentration, enabling multi-factor interactive analysis of these environmental parameter data. The aim is to reveal that these environmental factors do not act independently but rather influence each other in a nonlinear manner, jointly driving the sensor degradation process. This reveals the nonlinear coupling relationships between environmental factors; for example, high humidity and high corrosive gas concentrations may produce a more severe corrosion effect than either factor acting alone.

[0166] Based on the nonlinear coupling relationships between environmental factors and the multidimensional eigenvectors, a quantification model is constructed to measure the nonlinear contribution of environmental factors to the sensor's internal degradation modes. This can be understood as establishing a mathematical or computational model that quantifies the influence of different environmental factors and their nonlinear coupling relationships on the sensor's internal degradation modes. The multidimensional eigenvectors characterize multiple dimensions of the sensor's internal microscopic physical state, such as changes in material structure and degradation of electrochemical performance. This model aims to accurately assess the nonlinear contribution of each environmental factor (including its interaction with other factors) to the sensor's degradation modes. For example, through regression analysis, deep learning models, and other methods, the input of environmental parameters is mapped to the output of the sensor's degradation modes, taking into account their nonlinear characteristics.

[0167] In practical applications, by utilizing the established model and combining it with real-time monitored environmental parameter data, the specific influence weight or intensity of each environmental factor (and its interactions) on the sensor degradation mode is dynamically calculated and evaluated. This enables the real-time calculation and quantification of the nonlinear coupling contribution of different environmental factors to the sensor's internal degradation mode through the aforementioned nonlinear contribution quantification model. The aim is to obtain accurate environmental impact factors that can accurately reflect which environmental factors and their combinations play a dominant role in sensor degradation under the current environment, thereby providing a more reliable basis for subsequent degradation mode analysis and baseline correction of expected behavior.

[0168] The technical solution of this application, by introducing multi-factor interaction analysis, can comprehensively capture the complex nonlinear coupling relationships between environmental parameter data, overcoming the limitation of traditional single correlation analysis that may ignore the synergistic effects of multiple factors. It is precisely this deep understanding of these nonlinear coupling relationships that enables the construction of a more accurate nonlinear contribution quantification model. This model combines the nonlinear coupling relationships of environmental factors with a multidimensional eigenvector formed by the microscopic physical state characteristics of the sensor, thereby enabling a more accurate assessment of the nonlinear contributions of different environmental factors to the sensor's internal degradation modes. By calculating and quantifying these contributions in real time, this application obtains more refined and accurate environmental impact factors. These factors not only reflect the effects of individual environmental factors but, more importantly, reveal how their complex interactions jointly drive sensor degradation, thus providing a more solid data foundation for subsequent sensor degradation mode analysis.

[0169] Through the above technical solution, this application can more comprehensively and accurately identify the impact of environmental factors on the internal degradation modes of sensors. Compared with simply performing real-time correlation analysis, this application significantly improves the accuracy and reliability of environmental impact factor identification by considering the nonlinear coupling relationship between environmental factors and constructing a nonlinear contribution quantification model. This allows for a more in-depth analysis of the internal degradation modes of sensors, enabling the earlier and more accurate detection of potential degradation risks, thereby providing a more reliable basis for correcting the expected behavioral baseline, and ultimately improving the overall efficiency and accuracy of building safety data management.

[0170] In some embodiments, the step of performing multi-factor interactive analysis on the environmental parameter data to obtain the nonlinear coupling relationship between environmental factors includes:

[0171] The corrosion rate and hydrolysis rate of the material surface are calculated in real time based on the salt spray concentration, the humidity, the temperature, and the corrosive gas concentration.

[0172] Based on the material surface corrosion rate, the hydrolysis rate, and thermodynamic principles, the nonlinear coupling relationship between the environmental parameter data is derived in real time.

[0173] When the environmental parameter data changes significantly, the corrosion rate of the material surface and the hydrolysis rate are recalculated, and the nonlinear coupling relationship between the environmental factors is derived in real time.

[0174] Among these, the salt spray concentration, humidity, temperature, and corrosive gas concentration are key elements constituting the multi-dimensional environmental parameter data, directly affecting the corrosion and hydrolysis processes of materials. Specifically, by using a preset physicochemical model or empirical formula, combined with real-time monitored environmental parameter data, the corrosion rate and hydrolysis rate of materials under the current environment are quantitatively assessed, achieving real-time calculation of the material surface corrosion rate and hydrolysis rate. For example, the corrosion rate can be estimated using principles such as electrochemical impedance spectroscopy and weight loss methods, while the hydrolysis rate is closely related to factors such as the chemical composition of the material, temperature, and humidity. The thermodynamic principles can be understood as the laws of chemical thermodynamics, which describe the direction, extent, and energy conversion of chemical reactions, such as Gibbs free energy change and reaction enthalpy change. These principles provide a solid theoretical foundation for deriving the complex nonlinear coupling relationships between environmental parameter data. The nonlinear coupling relationship means that the influence of environmental factors on material degradation is not a simple linear superposition, but rather involves complex interactions and synergistic effects. For example, under specific humidity and temperature conditions, the presence of corrosive gases may significantly accelerate or inhibit the corrosion process, forming a nonlinear response. Furthermore, when the environmental parameter data changes significantly, such as exceeding a preset threshold or rate of change, it will trigger a recalculation of the material surface corrosion rate and the hydrolysis rate, and deduce the nonlinear coupling relationship between the environmental factors in real time, so as to ensure that the obtained coupling relationship can dynamically adapt to the complexity and variability of the actual environment.

[0175] The above technical solution enables precise quantification of the nonlinear contribution of environmental factors to the internal degradation modes of sensors. Specifically, by calculating the corrosion rate and hydrolysis rate of material surfaces in real time and deriving the nonlinear coupling relationship between environmental parameter data using thermodynamic principles, this application provides a deeper understanding of how environmental factors synergistically affect sensor performance. This dynamic and physicochemically based analytical method overcomes the problems of simplistic assumptions or insufficient empirical models that may exist in traditional methods, significantly improving the accuracy and real-time performance of deriving the nonlinear coupling relationship between environmental factors. This provides more reliable basic data for the subsequent construction of a nonlinear contribution quantification model, thereby improving the accuracy of predicting and evaluating the internal degradation modes of sensors, ultimately contributing to more efficient and reliable building safety data management.

[0176] This application also proposes an efficient building safety data management system based on a large safety model, such as... Figure 2 As shown, a high-efficiency building safety data management system 100 based on a large safety model is provided. The system includes:

[0177] The baseline establishment module 10 is used to establish a baseline for monitoring the expected behavior of the sensor over time.

[0178] Deviation identification module 20 is used to identify a persistent deviation between sensor data and the expected behavior baseline based on the expected behavior baseline;

[0179] The consistency verification module 30 is used to perform behavioral logic consistency verification between the sensor data and at least one associated sensor data based on preset physical association rules.

[0180] The data preprocessing module 40 is used to preprocess the sensor data that is identified as having the continuous deviation or having inconsistent behavior logic to obtain target sensor data. The preprocessing includes at least one of reducing data credibility and dynamic numerical correction.

[0181] The risk assessment module 50 is used to perform weighted learning and risk assessment on the target sensor data based on a large safety model.

[0182] The overall technical solution of this application aims to address the problem in traditional building safety management where sensor data deviations due to physical aging or environmental influences affect the accuracy of safety model judgments. First, through a baseline establishment module, the system gains a clear understanding of the normal behavior patterns of sensors. When sensor data consistently deviates from this baseline, the deviation identification module indicates a potential sensor anomaly. For example, when a tower crane anemometer continuously reports low wind speeds due to aging, its data will significantly deviate from the preset expected wind speed baseline, thus being identified by the system. Second, through a consistency verification module, the system can detect potential logical contradictions between different sensor data. For example, if a wind speed sensor reports low wind speed, but a tower crane sway amplitude sensor reports high sway amplitude, this inconsistency will trigger a system alarm. These deviations and inconsistencies are hidden defects that are difficult to detect using traditional methods.

[0183] After identifying these abnormal data, the data preprocessing module reduces the data's reliability or dynamically corrects its values. For example, for anemometer data that consistently underreports wind speed, the preprocessing module reduces its weight in the safety model and may refer to other related sensors (such as data from nearby weather stations) for correction. This way, even if the sensor itself has problems, its biased output data will not have an excessively negative impact on the safety model, preventing the model from becoming less sensitive to real risks due to "mislearning." Finally, the risk assessment module learns and performs risk assessment based on this preprocessed and weighted target sensor data. Because the data input to the model has been cleaned and corrected, the model can more accurately identify real risk patterns and provide reliable risk assessment results. For example, even if a particular anemometer's data is biased, after processing by the data preprocessing module, the risk assessment module can still combine other normal data and corrected wind speed data to accurately determine that the actual wind speed has exceeded the safety limit, thus issuing a timely warning.

[0184] Compared with existing technologies, the core innovation of this application lies in its deep integration of sensor data anomaly identification, preprocessing, and security large-scale model application, achieved through a modular system design. Traditional methods often focus on single data anomaly detection or model training optimization, failing to effectively address the profound impact of sensor data source deviations on model accuracy. For example, existing technologies may only use threshold detection to determine whether sensor data is abnormal, but they are difficult to detect persistent, systematic, and minute deviations. This application, through a baseline establishment module, a deviation identification module, and a consistency verification module, can more comprehensively and precisely identify hidden defects in sensor data.

[0185] The above description is merely an embodiment of this application and is not intended to limit the scope of protection of this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of protection of this application.

Claims

1. A method for efficient management of building safety data based on a large-scale safety model, characterized in that, include: Establish a baseline for monitoring the expected behavior of sensors over time; Based on the expected behavior baseline, identify persistent deviations between sensor data and the expected behavior baseline; Based on preset physical association rules, the sensor data is checked for behavioral logic consistency with at least one associated sensor data; The sensor data identified as having the persistent deviation or inconsistent behavior logic is preprocessed to obtain the target sensor data. The preprocessing includes at least one of reducing data credibility and dynamic numerical correction. The target sensor data is weighted and risk-assessed based on a large-scale safety model. The steps for establishing a baseline for monitoring the expected behavior of the sensor over time include: Real-time sensing of multi-dimensional environmental parameters to obtain environmental parameter data, including salt spray concentration, humidity, temperature and corrosive gas concentration; The microscopic physical state characteristics inside the sensor are periodically detected to obtain microscopic physical state characteristic values; Establish nonlinear correlation rules between environmental parameter data, microscopic physical state characteristic values, and sensor measurement deviations to obtain a set of nonlinear correlation rules; Based on the environmental parameter data, the microscopic physical state characteristic values, and the set of nonlinear association rules, a baseline for expected behavior is generated. When the expected behavior baseline deviates continuously from the sensor data, the environmental parameter data and the microscopic physical state characteristic values ​​are analyzed, and the expected behavior baseline is corrected according to the set of nonlinear association rules. The step of analyzing the environmental parameter data and the microscopic physical state characteristic values, and correcting the expected behavior baseline according to the set of nonlinear association rules when the expected behavior baseline deviates continuously from the sensor data includes: Time series analysis was performed on the environmental parameter data to obtain the evolution trend and abrupt change points of environmental factors; Trend analysis of the microscopic physical state characteristics reveals the mode transitions of degradation within the sensor. Based on the evolution trend and abrupt change points of the environmental factors, the mode change of the degradation inside the sensor, and the set of nonlinear association rules, the deviation patterns are classified to obtain the classification results. Based on the classification results, the weights of the nonlinear association rules in the expected behavior baseline correction are adjusted to obtain the target expected behavior baseline; The degree of fit between the obtained target expected behavior baseline and the sensor data is evaluated; Based on the degree of fit, the target expected behavior baseline is triggered for correction and optimization; The preset physical association rules refer to those established based on physical laws, engineering specifications, or equipment operating logic, used to describe the inherent logical relationships or mutual influence mechanisms between different sensor data. The microscopic physical state characteristics inside the sensor refer to the physical property data of the materials, structure, or electronic components inside the sensor.

2. The efficient management method for building safety data based on a large safety model according to claim 1, characterized in that, Prior to the step of identifying persistent deviations between sensor data and the expected behavior baseline, the following steps are included: The raw sensor data is obtained by high-frequency acquisition and preprocessing of the raw signal through the sensor. The original sensor data is decomposed intrinsically to obtain the basic signal components with different frequencies and amplitudes; The features of the basic signal components are compared with the pre-established degradation mode feature set to identify the nonlinear noise components caused by the internal degradation of the sensor. The degradation mode feature set records the features of the basic signal components after intrinsic decomposition of the data collected by the sensor with different degrees of internal degradation. The nonlinear noise component is removed from the raw sensor data to extract clean sensor data, and a health status index is assigned. The health status index is used to perform weighted learning and risk assessment on the sensor data based on a large safety model.

3. The efficient management method for building safety data based on a large safety model according to claim 1, characterized in that, The step of preprocessing the identified sensor data exhibiting persistent deviations or inconsistent behavioral logic to obtain target sensor data includes at least one of the following steps: reducing data reliability and dynamic numerical correction. Obtain historical data reliability records for sensor data; When the credibility of a certain sensor data recorded in the sensor data credibility history is continuously lower than a preset threshold for a certain period of time, a credibility reduction risk assessment is triggered. Based on the results of the confidence reduction risk assessment, it is determined whether the sensor is in a critical degradation state or has irreversible damage. When the sensor is in a critical degradation state or has irreversible damage, a recommendation to replace or repair the sensor is issued, and the weight of the sensor data is temporarily reduced to the minimum, while redundant sensor data or inference based on associated sensor data is enabled. The specific time length refers to the time period during which the credibility remains below the preset threshold.

4. The efficient management method for building safety data based on a large safety model according to claim 1, characterized in that, The step of performing time series analysis on the environmental parameter data to obtain the evolution trend and abrupt change points of environmental factors includes: The environmental parameter data is decomposed into environmental components at different time scales. Trend analysis was performed on the environmental components at different time scales to obtain the evolution trend of environmental factors; Anomaly detection was performed on the environmental components at different time scales to obtain the abrupt change points of environmental factors; The evolution trend and abrupt change points of the environmental factors are fused to obtain the evolution trend and abrupt change points of the environmental factors.

5. The efficient management method for building safety data based on a large safety model according to claim 1, characterized in that, The step of performing trend analysis on the microscopic physical state characteristic values ​​to obtain the mode transition of internal degradation of the sensor includes: Multidimensional feature extraction is performed on the microscopic physical state feature values ​​to obtain a multidimensional feature vector; Real-time correlation analysis is performed on the environmental parameter data to obtain environmental impact factors related to the multidimensional feature vector; The multidimensional feature vector is coupled with the environmental impact factor to obtain the degradation pattern characteristics under environmental coupling. Based on the degradation mode characteristics under environmental coupling, the degradation modes inside the sensor are dynamically clustered to obtain mode transition points; At the mode transition point, the degradation modes inside the sensor are reclassified to obtain the target degradation mode.

6. The efficient management method for building safety data based on a large safety model according to claim 5, characterized in that, The step of performing real-time correlation analysis on the environmental parameter data to obtain environmental impact factors related to the multidimensional feature vector includes: Multi-factor interactive analysis was performed on the environmental parameter data to obtain the nonlinear coupling relationship between environmental factors; Based on the nonlinear coupling relationship between environmental factors and the multidimensional feature vector, a nonlinear contribution quantification model of environmental factors to the internal degradation mode of the sensor is constructed. The nonlinear contribution quantification model is used to calculate and quantify the nonlinear coupling contribution of different environmental factors to the internal degradation mode of the sensor in real time, thereby obtaining the environmental impact factor.

7. The efficient management method for building safety data based on a large safety model according to claim 6, characterized in that, The step of performing multi-factor interactive analysis on the environmental parameter data to obtain the nonlinear coupling relationship between environmental factors includes: The corrosion rate and hydrolysis rate of the material surface are calculated in real time based on the salt spray concentration, the humidity, the temperature, and the corrosive gas concentration. Based on the material surface corrosion rate, the hydrolysis rate, and thermodynamic principles, the nonlinear coupling relationship between the environmental parameter data is derived in real time. When the environmental parameter data changes significantly, the corrosion rate of the material surface and the hydrolysis rate are recalculated, and the nonlinear coupling relationship between the environmental factors is derived in real time.

8. A high-efficiency management system for building safety data based on a large safety model, used to execute the method as described in any one of claims 1-7, characterized in that, The system includes: The baseline establishment module is used to establish a baseline for monitoring the expected behavior of the sensor over time. A deviation identification module is used to identify a persistent deviation between sensor data and the expected behavior baseline based on the expected behavior baseline. The consistency verification module is used to perform behavioral logic consistency verification between the sensor data and at least one associated sensor data based on preset physical association rules. The data preprocessing module is used to preprocess the sensor data that is identified as having the persistent deviation or having inconsistent behavior logic to obtain the target sensor data. The preprocessing includes at least one of reducing data credibility and dynamic numerical correction. The risk assessment module is used to perform weighted learning and risk assessment on the target sensor data based on a large safety model.

Images