Real-time settlement monitoring device for building ground and use method of real-time settlement monitoring device

Abstract

The invention discloses a building ground real-time settlement monitoring device and a use method thereof, and belongs to the field of building structure safety monitoring. The monitoring device comprises a hierarchical sensor network which is used for carrying out multi-time-scale real-time data acquisition and comprehensively obtaining deformation data and related environmental parameters of a building structure; the data processing and analyzing module is used for performing real-time processing and intelligent analysis on the acquired data, and identifying and classifying abnormal deformation characteristics of the building structure in time; the deep learning prediction module is used for quantitatively predicting the probability state and the evolution trend of building settlement by constructing a multi-scale time sequence prediction model; the multi-factor analysis module is used for carrying out coupling modeling and comprehensive analysis on the environmental factors, the structural characteristics and the abnormal evolution process so as to identify key influence factors and action mechanisms thereof; and the risk assessment and early warning module is used for performing grading assessment on the building settlement risk based on the prediction and analysis result and generating corresponding early warning information and decision support schemes.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of building structure safety monitoring, more particularly, to a building ground real-time settlement monitoring device and a method thereof. BACKGROUND

[0002] Building ground settlement is an important factor affecting structural safety and functional use, and its abnormal change may cause building cracks, structural damage, and even serious safety accidents. Therefore, accurate monitoring of building ground settlement is of great significance to ensure building safety. In the prior art, ground settlement monitoring mainly relies on manual measurement or periodic observation, such as leveling, total station measurement, and GPS positioning. Although these methods can obtain settlement data, they have problems such as long observation period, data update lag, difficulty in covering key parts, and large interference from environmental factors, which cannot meet the needs of modern buildings, especially large or high-rise buildings, for real-time and high-precision settlement monitoring.

[0003] In recent years, with the development of sensor technology and data acquisition technology, automated monitoring methods have been gradually applied to building settlement monitoring. For example, sensor arrays are arranged at key parts of building foundations and load-bearing structures to achieve continuous data acquisition, and data processing modules are used for foundation analysis and trend judgment. However, the existing methods still have some deficiencies in practical application: first, a single type of sensor or a simple sensor network cannot simultaneously capture the instantaneous abnormality, short-term fluctuation, and long-term evolution trend of building settlement; second, the data analysis method relies on empirical formulas or simple statistical models, and it is difficult to fully utilize multi-source, multi-dimensional, and high-frequency data to comprehensively analyze the coupling relationship between environmental factors, structural characteristics, and settlement evolution; third, there is a lack of reliable risk classification evaluation and dynamic early warning means, which cannot provide real-time and operable decision-making basis for management and emergency.

[0004] In summary, how to realize multi-time scale real-time monitoring, abnormality identification, evolution trend prediction, and risk classification early warning of building ground settlement has become a technical problem to be solved. SUMMARY

[0005] In order to overcome the series of defects existing in the prior art, the purpose of the present application is to provide a building ground real-time settlement monitoring device, which includes the following modules.

[0006] Hierarchical sensor network for carrying out multi-time scale real-time data acquisition and comprehensively obtaining deformation data of building structure and related environmental parameters.

[0007] Data processing and analysis module for real-time processing and intelligent analysis of collected data, and timely identification and classification of abnormal deformation characteristics of building structure.

[0008] A deep learning prediction module is configured to quantitatively predict the probability state and evolution trend of building settlement by constructing a multi-scale time series prediction model.

[0009] A multi-factor analysis module is configured to model and comprehensively analyze the environmental factors, structural characteristics and abnormal evolution process, so as to identify the key influencing factors and their action mechanisms.

[0010] A risk assessment and early warning module is configured to grade the building settlement risk based on the prediction and analysis results, and generate corresponding early warning information and decision support scheme.

[0011] Further, the hierarchical sensor network includes a basic layer sensor array, a structural layer sensor node and an environmental monitoring sensor group, wherein: the basic layer sensor array adopts a combination of high-precision laser displacement sensors and digital tilt sensors, is installed at key parts of the building foundation according to the grid arrangement principle, and the sensor spacing is set in the range of 5-15 meters according to the building span, so as to ensure that the 0.1 millimeter level of micro deformation can be captured; the structural layer sensor node adopts wireless strain sensors and three-axis acceleration sensors, is arranged along the main load-bearing structure of the building, and realizes real-time monitoring of the structural stress state change and dynamic response; the environmental monitoring sensor group includes soil moisture sensors, underground water level monitoring sensors, temperature and humidity sensors and ground vibration sensors, and is used to obtain external environmental parameters affecting building settlement.

[0012] Further, the data processing and analysis module includes a data preprocessing unit, a feature extraction unit and an anomaly detection unit, wherein: the data preprocessing unit is used for noise filtering and signal enhancement processing of the original sensor data, and simultaneously corrects sensor drift and suppresses environmental interference; the feature extraction unit extracts multi-frequency domain features based on wavelet transform, and reduces the data dimension by combining the principal component analysis method, so as to extract key feature parameters capable of representing the building deformation state; the anomaly detection unit establishes a dynamic control limit by using a statistical process control method, identifies the abnormal deformation mode by comparing with the historical baseline data, triggers an abnormal early warning when three consecutive sampling points exceed the control limit, and automatically adjusts the detection sensitivity according to different settlement development stages, so as to meet the monitoring requirements.

[0013] Further, the deep learning prediction module adopts a hybrid architecture combining long short-term memory neural network and attention mechanism, captures the time series dependence of settlement data by constructing a multi-layer LSTM network, and uses attention mechanism to adaptively allocate the weights of data at different time steps, thereby improving the prediction accuracy of key time nodes; the deep learning prediction module includes a data normalization layer, a feature encoding layer, a time series modeling layer, and an output decoding layer, wherein: the data normalization layer performs standardization processing on the input multi-dimensional time series data, the feature encoding layer converts sensor data into a high-dimensional feature vector, the time series modeling layer learns the dynamic pattern of settlement evolution through a bidirectional long short-term memory neural network, and the output decoding layer generates the settlement probability distribution prediction result in the future time window.

[0014] Further, the multi-factor analysis module includes an environmental factor influence evaluation unit, a structural property analysis unit, and a settlement mechanism identification unit, wherein: the environmental factor influence evaluation unit is used to quantitatively analyze the action relationship of various environmental factors on settlement, and establish the response function and sensitivity coefficient thereof; the structural property analysis unit evaluates the mechanical state and safety reserve level of the building structure through structural mechanics modeling and measured data correction; the settlement mechanism identification unit analyzes the spatio-temporal evolution characteristics of settlement through data mining and pattern recognition, identifies the dominant factors and triggering conditions, extracts typical settlement patterns, and establishes a discrimination criterion.

[0015] Further, the multi-factor analysis module further includes a dynamic weight distribution mechanism implementation unit for adaptively adjusting the weight coefficients of each influencing factor according to different stages of settlement evolution, which specifically includes the following steps.

[0016] By calculating the covariance matrix between each influencing factor and the settlement response in real time, the real-time influence strength of the factor is dynamically evaluated.

[0017] The sliding window mechanism is adopted to perform real-time statistics on the relationship between the factors and the settlement in the last 72 hours, and the contribution variance and stability index of each factor are calculated. When the contribution variance of a certain factor exceeds the set threshold, the weight coefficient of the factor is automatically increased to ensure that the key factor can be given priority at the critical moment.

[0018] By calculating the marginal contribution of each factor to the reduction of settlement prediction error, the importance ranking of the factors is updated in real time, and the weight of the top three factors is increased by 20-30%, while the weight of the last three factors is reduced accordingly, to realize the dynamic redistribution of weight resources.

[0019] Considering the prediction accuracy, calculation efficiency and model stability, the Pareto optimal solution set is used to select the best weight configuration scheme to ensure that the weight of the key factor is increased while the overall performance is not affected.

[0020] The application also aims to provide a building ground real-time settlement monitoring device use method, comprising the following steps.

[0021] According to the initial state of the settlement monitoring device, the key positions of the building, and the preset monitoring accuracy requirements, the sampling frequency configuration of each sensor is determined, and a multi-time scale data acquisition scheme of the hierarchical sensor network is generated in combination with the key positions of the building and the sampling frequency configuration.

[0022] The hierarchical sensor network is controlled to collect building structure deformation data and environmental parameter data in real time according to the multi-time scale data acquisition scheme.

[0023] Whenever the hierarchical sensor network completes a round of data collection, the currently collected structure deformation data is marked as target monitoring data, the abnormality detection parameters of the data processing and analysis module are configured based on the target monitoring data and a preset adaptive threshold, real-time abnormality recognition of the target monitoring data is performed to identify the instantaneous abnormal deformation of the building structure.

[0024] In the real-time abnormality recognition process, the duration of the instantaneous abnormal deformation is analyzed to obtain the duration characteristics of the instantaneous abnormal deformation in real time, the abnormal deformation is classified and labeled in combination with the instantaneous abnormal deformation and the duration characteristics, and the corresponding feature fingerprint data is extracted.

[0025] If the duration of the instantaneous abnormal deformation does not reach the preset classification standard, the real-time abnormality recognition is maintained until the classification of the abnormality is completed.

[0026] If the instantaneous abnormal deformation has been classified and labeled, the abnormality recognition process of the target monitoring data is ended, and the classification result is transmitted to the deep learning prediction module.

[0027] By fusing feature data of different time windows and introducing an attention mechanism, the probability state quantitative prediction of the evolution of building abnormal deformation to continuous settlement is realized.

[0028] Based on multivariate time series analysis, the contribution weight of each influence factor to abnormal evolution is calculated, and the analysis initial parameters of the multi-factor analysis module are obtained.

[0029] According to the analysis initial parameters, the environmental parameter data, and the probability state quantitative prediction result, spatial correlation analysis is performed to realize the multi-factor coupling relationship identification of environmental factors, structure characteristics, and abnormal evolution process.

[0030] The abnormal classification result and the probability state quantitative prediction result are mapped to a two-dimensional risk matrix, and the evaluation benchmark parameters of the risk assessment and early warning module are obtained.

[0031] Based on the evaluation benchmark parameters and the two-dimensional risk matrix, automatic evaluation and early warning of the building structure evolution risk are realized, and comprehensive decision support information including a settlement trend curve, a risk analysis report and emergency disposal suggestions are generated.

[0032] In the process of generating the comprehensive decision support information, early warning notifications are pushed in real time, and the risk analysis report and the emergency disposal suggestions are combined to guide on-site emergency response.

[0033] If the building structure evolution risk level does not reach the early warning standard, the normal monitoring state is maintained and data collection is continued.

[0034] If the building structure evolution risk level has reached the early warning standard, an emergency plan is immediately started, and the building ground real-time settlement monitoring device is controlled to enter a high-frequency monitoring mode.

[0035] Further, the duration analysis is used to statistically analyze the duration characteristics of the instantaneous abnormal deformation in real time, and the abnormal deformation is classified and labeled based on the instantaneous abnormal deformation and the duration characteristics, including the following steps.

[0036] A time window is established for each instantaneous abnormal deformation event, and its starting time and duration are tracked in real time, so as to quantify the duration characteristics of the abnormal deformation.

[0037] The key duration parameters of the abnormal deformation event are calculated, including the duration, cumulative deformation amplitude and trend change.

[0038] According to the deviation degree of the key duration parameters of the abnormal deformation relative to the normal range, the abnormal intensity is evaluated and a reference is provided for classification.

[0039] The abnormal deformation is classified based on the duration characteristics and the intensity evaluation results, and it is determined whether the preset threshold is reached.

[0040] The classification results are labeled in a structured manner, the abnormal type and characteristic parameters are recorded, and are synchronized to the deep learning prediction module and the multi-factor analysis module to update the abnormal deformation feature library.

[0041] Further, the probability state quantitative prediction is realized.

[0042] The Monte Carlo method is used to randomly sample the prediction model parameters, generate multiple potential parameter combinations, construct different prediction trajectories and quantify the prediction uncertainty.

[0043] The parameters obtained by random sampling are input into the prediction model to generate multiple settlement prediction trajectories covering different future states.

[0044] The predicted trajectories are statistically analyzed to calculate the occurrence probability of different settlement levels, form a probability distribution, and extract key indicators, including mean prediction, confidence interval, and extreme event probability.

[0045] In the prediction process, model uncertainty, parameter uncertainty, and observation uncertainty are comprehensively considered, and the posterior distribution of the model parameters is updated through Bayesian inference.

[0046] The prediction results in the form of probability distribution are taken as the output to provide quantitative reference for decision-making, including average state, confidence interval range, and extreme settlement event probability.

[0047] Further, according to the analysis of the initial parameters, the environmental parameter data, and the probabilistic state quantitative prediction results, spatial correlation analysis is performed to identify the multi-factor coupling relationship between environmental factors, structural characteristics, and abnormal evolution process, including the following steps.

[0048] A spatial weight matrix is constructed according to the spatial positions and structural connectivity of each part of the building, which is used to describe the spatial relationship between different monitoring points.

[0049] The analysis of the initial parameters, environmental parameter data, and probabilistic state prediction results is unified and coded to facilitate the unified processing of multiple factors in spatial correlation analysis.

[0050] The spatial autocorrelation index is used to quantify the aggregation characteristics of the settlement of each part of the building, identify abnormal settlement hot and cold spot areas, and provide reference for causal relationship analysis.

[0051] Based on the constructed spatial weight matrix, a regression model considering spatial effects is established to analyze the influence of environmental factors and structural characteristics on abnormal settlement evolution, while eliminating the interference of spatial dependence on parameter estimation.

[0052] Through main effect analysis and interaction effect analysis, the direct action and synergistic effect of each environmental factor and structural characteristic on the abnormal evolution process are quantitatively evaluated, and the multi-factor coupling mechanism is revealed.

[0053] The results of spatial correlation analysis and coupling relationship identification are structured and output to provide quantitative basis for the multi-factor analysis module and risk assessment module, and update the analysis parameters of the monitoring device.

[0054] Compared with the prior art, the present application has the following beneficial effects: the present application combines hierarchical sensor network, multi-time scale data acquisition, data processing and analysis, deep learning prediction, multi-factor coupling analysis, and risk assessment and early warning, realizes abnormal identification, probabilistic evolution prediction, and automatic risk grading early warning of building structure settlement, and thus provides quantitative, intelligent, and real-time decision support for structure safety management. BRIEF DESCRIPTION OF DRAWINGS

[0055] Figure 1 A module communication timing diagram of the building ground real-time settlement monitoring device of the application. In the figure, the hierarchical sensor network collects building structure deformation data and environmental parameters, and sends the data to the data processing and analysis module for real-time processing and abnormality recognition; the processed data is transmitted to the deep learning prediction module and the multi-factor analysis module at the same time, the deep learning prediction module performs probability state prediction of abnormal deformation to settlement evolution, and the multi-factor analysis module completes the coupling analysis of environmental factors, structure characteristics and abnormal evolution process; finally, the risk assessment and early warning module summarizes the prediction and analysis results, performs risk grading assessment, and generates settlement trend, early warning information and emergency disposal suggestions comprehensive decision support information.

[0056] Figure 2 A flowchart of the method for using the building ground real-time settlement monitoring device of the application. DETAILED DESCRIPTION

[0057] In order to make the purpose, technical scheme and advantages of the present application clearer, the technical scheme in the embodiments of the present application will be described in more detail below with reference to the drawings in the embodiments of the present application. In the drawings, the same or similar reference numerals represent the same or similar elements or elements having the same or similar functions throughout. The described embodiments are part of the embodiments of the present application, not all of the embodiments.

[0058] Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without making creative efforts belong to the scope of protection of the present application.

[0059] The embodiments described below with reference to the drawings and the directional words are exemplary and are intended to explain the present application, and cannot be understood as limiting the present application.

[0060] As shown in Figure 1 A building ground real-time settlement monitoring device, comprising the following modules.

[0061] The hierarchical sensor network is used to carry out real-time data collection of multiple time scales, and comprehensively obtain deformation data of the building structure and related environmental parameters.

[0062] The data processing and analysis module is used for real-time processing and intelligent analysis of the collected data, and timely identification and classification of abnormal deformation characteristics of the building structure.

[0063] The deep learning prediction module quantitatively predicts the probability state and evolution trend of building settlement by constructing a multi-scale time series prediction model.

[0064] A multi-factor analysis module is used to couple and comprehensively analyze environmental factors, structural characteristics and abnormal evolution processes to identify key influencing factors and their mechanisms.

[0065] A risk assessment and early warning module is used to grade the building subsidence risk based on the prediction and analysis results and generate corresponding early warning information and decision support schemes.

[0066] The building ground real-time subsidence monitoring device realizes dynamic perception, intelligent analysis, trend prediction and risk warning of the whole process of building subsidence through the synergistic effect of multiple modules in overall design. Firstly, the device can monitor the building ground and related environment in real time at different time scales through the setting of hierarchical sensor network. This multi-level and multi-dimensional data collection method not only improves the completeness and accuracy of data, but also ensures that the tiny deformation signal can be captured in the early stage of building subsidence, thereby avoiding the hysteresis and omission caused by too large sampling interval or insufficient collection dimension in traditional monitoring methods. Secondly, in the data processing and analysis link, the real-time data processing module equipped in the device can denoise, correct and intelligently classify the multi-source data collected by the sensor, so that the complex subsidence process can be decomposed into different feature types and identified. For example, when the building foundation appears uneven settlement, local abnormal subsidence or deformation related to environmental load changes, clear analysis results can be quickly generated to reduce the ambiguity and error of manual interpretation.

[0067] In the prediction aspect, the device introduces a deep learning method, considers the subsidence process as a typical time series problem, and simulates the dynamic evolution of subsidence by constructing a multi-scale time series prediction model. Compared with the traditional prediction method based on empirical formula or single regression model, the deep learning model can capture both short-term fluctuations and long-term trends, and can automatically learn the potential nonlinear relationship. This means that it can not only judge the current state, but also predict the subsidence trend and its uncertainty range in the future period of time, thereby providing proactive reference for building safety management. The prediction of this feature significantly improves the initiative and scientificity of risk management.

[0068] In addition, the introduction of the multi-factor analysis module further enhances the comprehensiveness and interpretability of the device. In the process of building settlement, the influencing factors are often superimposed in multiple aspects, including changes in geological conditions, fluctuations in groundwater level, differences in construction load distribution, and the stiffness and flexibility characteristics of the building structure itself. By coupling modeling of these environmental factors, structural characteristics, and abnormal settlement processes, not only can the identification be stopped at the phenomenon level, but also the key influencing factors and their mechanisms can be further explored. Such analysis helps to determine whether the settlement problem is caused by external environmental disturbances or by design and construction defects of the building itself, so that targeted and effective management and maintenance can be achieved in subsequent treatment and governance.

[0069] In terms of risk management and early warning, the device establishes a hierarchical risk assessment and early warning mechanism by combining prediction results with multi-factor analysis results. Unlike traditional monitoring systems that only issue an alarm when a certain fixed threshold is reached, the device can dynamically calculate the risk level based on the settlement trend, abnormal evolution speed, and comprehensive weight of influencing factors, and generate hierarchical warning information. For low-risk states, daily attention and inspection can be prompted; for medium-risk states, further detection and local reinforcement can be suggested; and for high-risk states, emergency warning and decision support schemes can be quickly triggered to remind relevant management personnel to take immediate preventive measures. Such a hierarchical early warning system not only avoids false positives and false negatives caused by overly single threshold settings, but also allows for differentiated response strategies at different risk stages, thereby achieving rational allocation of resources and efficient emergency response.

[0070] In summary, the advantages of the building ground real-time settlement monitoring device mainly lie in the following aspects: 1) it can accurately capture and continuously track the settlement in the early stages; 2) it can clearly analyze complex settlement processes; 3) it can give reasonable trend extrapolation before problems appear; 4) it can reveal the essential mechanism of settlement through multi-factor modeling; 5) it can ensure that different levels of risk are responded to and managed accordingly.

[0071] In summary, the building ground real-time settlement monitoring device proposed in this embodiment, through the collaborative work of sensor networks, data analysis, deep learning prediction, multi-factor coupling modeling, and hierarchical risk warning, realizes the complete chain from information collection to risk control. Its innovation lies not only in the significant improvement of monitoring capability, but also in the scientificity and practicality of prediction and early warning mechanisms, thereby significantly enhancing the building safety assurance capability in actual engineering applications.

[0072] Further, the hierarchical sensor network comprises a basic layer sensor array, a structural layer sensor node and an environmental monitoring sensor group, wherein: the basic layer sensor array adopts a combination of high-precision laser displacement sensors and digital tilt sensors, is installed at key parts of the building foundation according to the grid arrangement principle, the sensor spacing is set in the range of 5-15 meters according to the span of the building, so as to ensure that the micro deformation of 0.1 millimeter level can be captured; the structural layer sensor node adopts wireless strain sensors and three-axis acceleration sensors, is arranged along the main load-bearing structure of the building, and realizes real-time monitoring of the stress state change and dynamic response of the structure; the environmental monitoring sensor group comprises soil moisture sensors, underground water level monitoring sensors, temperature and humidity sensors and ground vibration sensors, and is used to obtain external environmental parameters affecting the building settlement.

[0073] As can be seen from the above, through the multi-dimensional cooperative monitoring of the basic layer, the structural layer and the environmental layer, high-precision and all-around perception of the building settlement is realized. The basic layer sensor array can capture micro deformation below millimeter level, ensuring early identification of abnormalities; the structural layer node obtains stress and vibration information in real time, reflecting the dynamic response of the building under load; the environmental monitoring sensor group comprehensively collects hydrological, climatic and ground disturbance external factors, providing sufficient data support for settlement cause analysis. Through this hierarchical arrangement and data fusion method, the building settlement state can be accurately reflected at multiple scales from micro to macro, significantly improving the reliability of monitoring and the scientificity of early warning.

[0074] Further, the data processing and analysis module comprises a data preprocessing unit, a feature extraction unit and an anomaly detection unit, wherein: the data preprocessing unit is used for noise filtering and signal enhancement processing of the original sensor data, and simultaneously corrects sensor drift and suppresses environmental interference; the feature extraction unit extracts multi-frequency domain features based on wavelet transform, and reduces the data dimension by combining a principal component analysis method, so as to extract key feature parameters capable of representing the deformation state of the building; the anomaly detection unit establishes a dynamic control limit by using a statistical process control method, identifies an abnormal deformation mode by comparing with historical reference data, triggers an abnormal early warning when three consecutive sampling points exceed the control limit, and automatically adjusts the detection sensitivity according to different settlement development stages, so as to meet the monitoring requirements.

[0075] The data preprocessing unit adopts a Kalman filter as a core noise filtering algorithm, sets a process noise covariance Q=1×10 -5 and a measurement noise covariance R=1×10 -2Recursive optimal estimation is performed to realize the complete filtering process of state prediction, error covariance calculation, gain coefficient update and state correction, to achieve a signal-to-noise ratio improvement of more than 25 dB, and the processing delay is less than 10 milliseconds; at the same time, the filtering parameters are dynamically adjusted according to the real-time data statistical characteristics, and updated once every 100 sampling points. The signal enhancement processing adopts a fourth-order Butterworth band-pass filter, and the passband range is set to 0.01-50Hz to cover the settlement characteristics of different frequency bands such as long-term trend settlement (0.01-1Hz), daily temperature effect (1-10Hz) and structure vibration response (10-50Hz). Through the adaptive gain control algorithm, different sensor signals are normalized to a unified power level, and the signal quality improvement rate is more than 95%. The sensor drift correction adopts the recursive least squares algorithm, sets a 24-hour sliding window to linearly fit the data in the window y(t)=a×t+b, takes the average value of the drift rate of multiple windows for correction, and establishes a three-level drift warning mechanism: drift rate > 0.05mm / day triggers a reminder, > 0.10mm / day triggers a warning, > 0.20mm / day triggers an alarm and suggests checking the sensor. After correction, the drift error can be controlled within ±0.02mm; for a multi-sensor monitoring network, a collaborative correction mechanism is realized, the adjacent sensor redundant information is used to cross-verify the drift estimation, the sensors with abnormal drift are automatically marked and the data quality is down-weighted, and a sensor health report is generated regularly. The environmental interference suppression adopts an adaptive notch filter, which is set to a quality factor of 30 and a notch depth of ≥40dB for power frequency interference (50Hz / 60Hz), high-order harmonics (100Hz, 150Hz, 120Hz, 180Hz) and mechanical vibration interference sources, and realizes zero-phase filtering; the main interference frequency components are automatically identified through spectrum analysis, and the corresponding notch filter is dynamically configured, so that the environmental interference power is reduced to 1 / 100 of the original, and the suppression effect is more than-40dB. After the complete data preprocessing process, the quality of the original sensor data is significantly improved: the signal-to-noise ratio is improved by more than 25dB, the drift error is within ±0.02mm, the environmental interference suppression is more than-40dB, the processing delay is less than 10ms, and the data integrity rate is more than 99.8%, providing a high-quality data basis for subsequent feature extraction and anomaly detection.

[0076] Among them, the feature extraction unit adopts Daubechies db8 wavelet as the transform base function to carry out 6-layer wavelet decomposition, and the preprocessed settlement signal is decomposed into 7 frequency bands: cD1 (25-50 Hz) corresponds to high-frequency noise and vibration, cD2-cD3 (6.25-25 Hz) corresponds to structural vibration response and short-term disturbance, cD4-cD6 (0.78-6.25 Hz) corresponds to daily fluctuation and temperature effect and underground water level change, and cA6 (0-0.78 Hz) corresponds to long-term settlement trend. The mean, standard deviation, maximum amplitude and energy of each layer wavelet coefficient are extracted to form 28 basic features; further, time-frequency joint features (adjacent layer coefficient correlation coefficient, skewness kurtosis, energy distribution, wavelet entropy, 32 dimensions in total), nonlinear features (approximate entropy, sample entropy, fractal dimension, 20 dimensions in total), frequency domain features (main frequency component, frequency band energy ratio, 16 dimensions in total) and time domain expansion features (zero-crossing rate, peak factor, margin factor, waveform factor, 24 dimensions in total) are constructed, and finally a complete feature vector of 128 dimensions is formed, which fully describes the time-frequency characteristics of the settlement signal. Principal component analysis reduces the 128-dimensional high-dimensional features to 16 principal components through eigenvalue decomposition of the covariance matrix, with an information retention rate of 98.5% and a dimension reduction ratio of 8:1; the 16 principal components have clear physical meaning: PC1-PC3 mainly come from low-frequency wavelet coefficients reflecting long-term settlement trend (contribution rate about 40%), PC4-PC7 mainly come from medium-frequency wavelet coefficients reflecting seasonal and daily periodic changes (contribution rate about 30%), PC8-PC12 mainly come from higher-frequency wavelet coefficients reflecting short-term fluctuations and structural responses (contribution rate about 20%), and PC13-PC16 mainly come from high-order statistical features reflecting the randomness and complexity of the data (contribution rate about 8.5%). At the same time, the incremental PCA algorithm is used to support the online update of the principal component model, and the covariance matrix is updated using the incremental eigenvalue decomposition algorithm every 1000 new samples to ensure that the PCA model always reflects the statistical characteristics of the current data. In addition, key parameters such as displacement amplitude features (peak-to-peak value, effective value, absolute average value, peak factor), frequency component features (main frequency identified by FFT analysis, frequency band energy distribution and frequency centroid), phase difference features (instantaneous phase extracted by Hilbert transform and phase synchronization index calculated to analyze the cooperativity of multiple points), and energy density features (time domain energy density, frequency domain energy density, wavelet energy density and normalized energy entropy) are extracted, and the importance of the features is evaluated through analysis of variance, correlation analysis and information gain to realize adaptive feature selection, remove redundant features with importance lower than the threshold, retain the top 16 core features in importance, and use Z-score standardization to ensure the comparability of different features.The performance indicators of the feature extraction unit are: the feature dimension is reduced from 120-dimensional sensor data to 16-dimensional principal components after extracting 128-dimensional wavelet features, the information retention rate is 98.5%, the processing delay is <5 ms, the feature discrimination degree (inter-class distance / intra-class distance ratio under different settling states) is >5, and the feature stability (relative standard deviation RSD of repeatability test) is <2%, which provides high-quality, low-dimensional, and strong feature data for anomaly detection and prediction analysis.

[0077] The abnormality detection unit adopts X-R control chart (mean-range control chart) for abnormality detection, and establishes control limits based on the 3σ criterion: upper control limit UCL = μ + 3σ and lower control limit LCL = μ - 3σ. Since the probability of data falling within the range of μ ± 3σ under normal distribution is 99.73%, the false alarm rate of normal data exceeding the limit is only 0.27%, which has high reliability. A dynamic baseline updating mechanism is realized: the initial baseline is calculated using the first 1000 sampling points to calculate the initial mean μ0 and standard deviation σ0, and the exponential weighted moving average (EWMA) algorithm is used to update the baseline once every 100 new data points, and the update formula is μ new = α · μ old + (1 - α) · x new, weight coefficient a = 0.95 keeps the memory of historical data; at the same time, the seasonality cycle of the data is identified, a seasonal baseline u(season) is established, and the de-seasoned data is used for control chart analysis to adapt to the non-stationary characteristics of the settlement process. At the same time, six types of control chart discrimination criteria are realized: (1) a single point exceeds the 3s control limit, (2) 9 consecutive points fall on the same side of the center line, (3) 6 consecutive points monotonically increase or decrease, (4) 14 consecutive points alternate up and down, (5) 2 of the 3 consecutive points fall in the 2s-3s area, and (6) 15 consecutive points fall within the u±s range (overly stable), any criterion meets the abnormality. The Mahalanobis distance is used to measure the degree of abnormality in abnormal pattern recognition, the baseline mean vector and covariance matrix are established based on the 36-month historical database, the Mahalanobis distance between the current data and the baseline is calculated, and 12 types of abnormal patterns are classified according to the Mahalanobis distance threshold: normal fluctuation (MD < 2), trend drift (2 < MD < 4), periodic anomaly (4 < MD < 6), sudden jump (MD > 6), slow decay, oscillation enhancement, step change, noise surge, seasonal deviation, nonlinear drift, intermittent anomaly and composite anomaly, an abnormal pattern database is established to store typical abnormal samples for fast matching, the recognition time is < 5 seconds, and the accuracy rate is 96.8%. The early warning trigger mechanism adopts a continuous over-limit trigger rule: detect the last 3 sampling points, and trigger the early warning when 3 consecutive points exceed the control limit; automatically adjust the detection sensitivity according to the settlement development stage: in the initial stage (data variance > 1.0), use high sensitivity to adjust the control limit to 2.5s, in the stable stage (variance 0.5-1.0), use medium sensitivity to keep 3.0s, and in the convergence stage (variance < 0.5), use low sensitivity to relax to 3.5s, to ensure timely detection of abnormalities and avoid excessive alarms. A hierarchical early warning mechanism is realized: I (low risk) normal monitoring, II (lower risk) intensive observation, III (medium risk) key attention, IV (higher risk) early warning response, and V (high risk) immediate disposal. After the early warning is triggered, the comprehensive decision support information is automatically generated, including the position and time of the abnormal point, the type and severity of the abnormality, the possible cause analysis (based on the similarity matching of the historical case library), the recommended disposal measures (divided into immediate response 0-24 hours, short-term disposal 1-7 days, medium-term improvement 1-3 months, and long-term prevention 3-12 months four levels), impact range assessment, etc. The decision support information is pushed to the management terminal in real time, the response time is < 1 second, and the multi-terminal synchronous and confirmation feedback mechanism is supported. The performance indicators of the abnormality detection unit are: detection accuracy 96.8%, false alarm rate < 0.5%, false negative rate < 1.0%, response time < 1 second, and sensitivity adjustment period is real-time dynamic, which provides timely and reliable abnormal early warning and decision support for building safety monitoring.

[0078] The data processing and analysis module realizes end-to-end processing from raw sensor data to abnormal detection results through the pipeline parallel processing architecture of the preprocessing unit (10 ms), the feature extraction unit (5 ms), and the abnormal detection unit (10 ms), with a total processing delay controlled within 25 ms and a data throughput of 1.2M per day, meeting the real-time monitoring requirements. A perfect quality control and monitoring mechanism is established to monitor the running state of each unit in real time, record key performance indicators such as processing success rate, delay distribution, error type, establish a data quality scoring mechanism to comprehensively evaluate signal-to-noise ratio, integrity, consistency, and timeliness, automatically label and isolate low-quality data, and generate module performance reports regularly to provide the basis for optimization. The module realizes adaptive optimization mechanism, automatically optimizes algorithm parameters including Kalman filter Q / R parameters, wavelet decomposition layers, PCA principal component quantity, control chart sensitivity, etc. according to the changes of long-term monitoring data statistical characteristics, uses reinforcement learning strategy to detect accuracy and false alarm rate as optimization objective, realizes intelligent adaptive adjustment of parameters; at the same time, parameter evolution history record is established to track parameter change trend, provide data support for algorithm improvement and upgrade. The seamless connection between each processing unit is realized through standardized data interface: the preprocessing unit outputs clean time series data, the feature extraction unit outputs dimensionality reduced feature vector, the abnormal detection unit outputs abnormal judgment result and risk level, all data are equipped with complete time stamp, sensor ID, data quality label, etc. meta information, which is convenient for tracing and auditing. The fault tolerance processing mechanism is realized under abnormal conditions, the backup algorithm or degradation processing strategy is automatically started when a unit fails to ensure continuous availability; the data caching mechanism is established to cache data to be processed during network interruption or maintenance, automatically compensate processing after recovery to ensure data continuity and integrity. The comprehensive performance indicators of the module are: total processing delay 25 ms, data processing success rate 99.2%, feature extraction accuracy 98.5%, abnormal detection accuracy 96.8%, availability 99.9%, data throughput 1.2M per day, through the cooperative work of the three-level architecture of preprocessing, feature extraction and abnormal detection, high-quality, high-efficiency and high-reliability settlement monitoring data processing is realized, which provides a solid data foundation and technical support for deep learning prediction module, multi-factor analysis module and risk assessment and early warning module, and constitutes the core data processing engine of the building ground real-time settlement monitoring device.

[0079] From the above, through the organic combination of preprocessing, feature extraction and anomaly detection, efficient processing and accurate identification of settlement data are realized. The data preprocessing unit can effectively filter out noise and environmental interference to ensure data reliability; the feature extraction unit uses multi-frequency domain analysis and dimension reduction method to extract key parameters to improve the expression ability of structural deformation characteristics; the anomaly detection unit compares the dynamic control limit with the historical data to identify abnormal settlement in time and has the function of self-adaptive sensitivity adjustment. Overall, the data processing and analysis module significantly improves the monitoring accuracy, real-time performance and reliability of abnormal warning.

[0080] Further, the deep learning prediction module adopts a hybrid architecture combining long short-term memory neural network and attention mechanism, which captures the time series dependence of settlement data by constructing a multi-layer LSTM network, and the attention mechanism is used to adaptively allocate the weight of different time step data, thereby improving the prediction accuracy of key time nodes; the deep learning prediction module includes a data normalization layer, a feature encoding layer, a time series modeling layer, and an output decoding layer, wherein: the data normalization layer standardizes the input multi-dimensional time series data, the feature encoding layer converts the sensor data into a high-dimensional feature vector, the time series modeling layer learns the dynamic pattern of settlement evolution through a bidirectional long short-term memory neural network, and the output decoding layer generates the settlement probability distribution prediction result in the future time window.

[0081] From the above, the deep learning prediction module effectively improves the modeling and prediction ability of the time series features of building settlement by combining long short-term memory network and attention mechanism. Its hierarchical structure can first normalize and high-dimensional feature encode the multi-dimensional sensor data, then use bidirectional LSTM to capture long-term and short-term dynamic rules, and highlight the contribution of key time nodes to the prediction result through the attention mechanism, and finally output the probability distribution of future settlement in the decoding layer. This process not only improves the accuracy and stability of the prediction, but also provides a reliable basis for early identification of settlement risks and trend analysis.

[0082] Further, the multi-factor analysis module includes an environmental factor influence evaluation unit, a structure characteristic analysis unit, and a settlement mechanism identification unit, wherein: the environmental factor influence evaluation unit is used to quantitatively analyze the action relationship of various environmental factors on settlement, and establish its response function and sensitivity coefficient; the structure characteristic analysis unit evaluates the mechanical state and safety reserve level of the building structure through structural mechanics modeling and measured data correction; the settlement mechanism identification unit analyzes the spatio-temporal evolution characteristics of settlement through data mining and pattern recognition, identifies the dominant factors and triggering conditions, extracts typical settlement patterns and establishes discrimination criteria.

[0083] From the above, the multi-factor analysis module realizes in-depth analysis of the causes and evolution law of building settlement through comprehensive modeling of environment, structure and mechanism. The environmental factor evaluation unit can quantify the influence degree of external conditions on settlement and reveal the sensitive factors and their response relationship; the structure characteristic analysis unit accurately reflects the mechanical state and safety margin of the building by combining theoretical modeling and measured correction; the settlement mechanism identification unit extracts typical settlement patterns and establishes discrimination criteria through data mining and pattern recognition, thereby identifying the dominant factors and triggering conditions.

[0084] Further, the multi-factor analysis module further includes a dynamic weight distribution mechanism implementation unit for adaptively adjusting the weight coefficients of each influencing factor according to different stages of settlement evolution, which includes the following steps.

[0085] By calculating the covariance matrix between each influencing factor and the settlement response, the real-time influence strength of the factor is dynamically evaluated.

[0086] The sliding window mechanism is used to perform real-time statistics on the relationship between the factors and the settlement in the last 72 hours, and the contribution variance and stability index of each factor are calculated. When the contribution variance of a factor exceeds a certain threshold, the weight coefficient of the factor is automatically increased to ensure that key factors can be given priority at critical moments.

[0087] By calculating the marginal contribution of each factor to the reduction of settlement prediction error, the importance ranking of the factors is updated in real time, and the weight of the top three factors is increased by 20-30%, while the weight of the last three factors is reduced accordingly, to realize dynamic redistribution of weight resources.

[0088] Considering the prediction accuracy, calculation efficiency and model stability, the Pareto optimal solution set is used to select the best weight allocation scheme to ensure that the key factor weight is increased while the overall performance is not affected.

[0089] From the above, by introducing the dynamic weight distribution mechanism, the adaptive adjustment of influencing factors at different stages is realized, and the accuracy and stability of settlement prediction are improved. The core is to dynamically evaluate the contribution of factors based on real-time covariance analysis and sliding window statistics, and to preferentially increase the weight of important factors and reduce the influence of irrelevant factors at critical stages. At the same time, the factor ranking is optimized based on the marginal contribution of prediction error, and the best weight allocation is selected through the Pareto optimal strategy to ensure that the key factor is enhanced while maintaining the overall calculation efficiency and prediction performance, thereby realizing accurate modeling and risk warning of the settlement evolution process.

[0090] Further, the settlement mechanism identification unit further includes a multi-scale spatio-temporal coupling modeling sub-unit for deep analysis of the coupling mechanism of influencing factors at different spatio-temporal scales, which includes the following implementation methods.

[0091] Micro-scale coupling modeling: For single sensor monitoring points, a local factor coupling model is established, and the direct coupling relationship between environmental factors such as groundwater level, soil moisture content, and temperature is analyzed by partial least squares regression. The coupling strength is quantified by the weighted combination of Pearson correlation coefficient and mutual information, and when the coupling strength exceeds 0.7, it is marked as a strong coupling relationship.

[0092] Mesoscale coupling modeling: Based on structural partition, a regional factor coupling network is established, and the spatial transmission effect between adjacent monitoring points is captured by graph convolutional neural network. At the same time, a time delay function is introduced to model the spatial propagation characteristics of factor action, realizing dynamic modeling of cross-regional factor action.

[0093] Macro-scale coupling modeling: A factor coupling matrix of the whole building system is constructed, and principal component analysis is used to extract the dominant factor combination at the system level. Through eigenvalue decomposition, the dominant coupling mode between factors is identified, and the time evolution equation of factor coupling strength is established.

[0094] Through cluster analysis, the historical coupling mode is divided into four types: stable, fluctuating, mutating, and mixed. For each type of mode, a feature template is established to monitor the matching degree of the current coupling state with the historical mode in real time. When the matching degree is less than 0.8, the new mode learning program is triggered to automatically update the coupling mode library.

[0095] As can be seen from the above, through multi-scale spatio-temporal coupling modeling, the factor action relationship in the settlement mechanism is deeply analyzed. At the micro level, the strong coupling characteristics between single point factors are identified, at the meso level, the spatial transmission and time delay effect of regional factors are captured, and at the macro level, the dominant factor combination at the system level is extracted and an evolution equation is constructed, fully presenting the synergistic action of factors at different scales. At the same time, combined with historical mode clustering and real-time matching, the coupling mode library is dynamically updated to ensure the adaptability of the model to environmental and structural changes, thereby improving the settlement prediction accuracy and risk identification ability.

[0096] As shown in Figure 2 , a building ground real-time settlement monitoring device usage method includes the following steps.

[0097] According to the initial state of the settlement monitoring device, the key positions of the building, and the preset monitoring accuracy requirements, the sampling frequency configuration of each sensor is determined, and the multi-time scale data acquisition scheme of the hierarchical sensor network is generated in combination with the key positions of the building and the sampling frequency configuration.

[0098] The hierarchical sensor network is controlled to collect building structure deformation data and environmental parameter data in real time according to the multi-time scale data acquisition scheme.

[0099] When the hierarchical sensor network completes a round of data collection, the current collected structural deformation data is marked as target monitoring data, and the anomaly detection parameters of the data processing analysis module are configured based on the target monitoring data and a preset adaptive threshold to perform real-time anomaly identification on the target monitoring data to identify the instantaneous abnormal deformation of the building structure.

[0100] In the real-time anomaly identification process, the duration of the instantaneous abnormal deformation is analyzed to obtain the duration characteristics of the instantaneous abnormal deformation, and the instantaneous abnormal deformation and the duration characteristics are combined to classify and mark the abnormal deformation, and corresponding feature fingerprint data is extracted.

[0101] If the duration of the instantaneous abnormal deformation does not reach a preset classification standard, the real-time anomaly identification is maintained until the classification of the abnormal deformation is completed.

[0102] If the instantaneous abnormal deformation has been classified and marked, the anomaly identification process of the target monitoring data is ended, and the classification result is transmitted to the deep learning prediction module.

[0103] By fusing feature data of different time windows and introducing an attention mechanism, the probability state quantitative prediction of the evolution of building abnormal deformation to continuous settlement is realized.

[0104] Based on multivariate time series analysis, the contribution weight of each influencing factor to the abnormal evolution is calculated, and the analysis initial parameters of the multi-factor analysis module are obtained.

[0105] According to the analysis initial parameters, the environmental parameter data, and the probability state quantitative prediction result, spatial correlation analysis is performed to realize the multi-factor coupling relationship identification of environmental factors, structural characteristics, and abnormal evolution process.

[0106] The abnormal classification result and the probability state quantitative prediction result are mapped to a two-dimensional risk matrix, and the evaluation benchmark parameters of the risk assessment and early warning module are obtained.

[0107] Based on the evaluation benchmark parameters and the two-dimensional risk matrix, the automatic assessment and hierarchical early warning of the evolution risk of the building structure are realized, and comprehensive decision support information including a settlement trend curve, a risk analysis report, and an emergency disposal suggestion is generated.

[0108] In the process of generating the comprehensive decision support information, a warning notification is pushed in real time, and the risk analysis report and the emergency disposal suggestion are combined to guide the on-site emergency response.

[0109] If the evolution risk level of the building structure does not reach the warning standard, the normal monitoring state is maintained and the data collection is continued.

[0110] If the building structure evolution risk level has reached the early warning standard, the emergency plan is immediately started, and the building ground real-time settlement monitoring device is controlled to enter the high-frequency monitoring mode.

[0111] The building ground real-time settlement monitoring device uses a set of systematic and hierarchical processes to achieve closed-loop management from data collection, anomaly identification, trend prediction to risk assessment and emergency response. First, in the data collection stage, according to the key positions of the building and the preset accuracy requirements, the sampling frequency of the sensor is reasonably configured, and a multi-time scale data collection scheme is formed. This way not only ensures that the monitoring needs of different parts and different stages are taken into account, but also effectively balances the monitoring accuracy and data processing efficiency, avoiding the monitoring blind area or data redundancy caused by a single sampling frequency. Secondly, the hierarchical sensor network acquires structural deformation and environmental parameters in real time according to the collection scheme, providing rich and multi-dimensional data support for subsequent analysis. In this way, the settlement state of the building can be captured at the millimeter level or even smaller, and external environmental factors are also included in the monitoring range, making the data more comprehensive and relevant.

[0112] In the data processing and anomaly identification stage, the adaptive threshold configuration and real-time anomaly detection mechanism are fully utilized to analyze and identify the target data collected in real time, so that the building structure can be quickly responded when it appears instantaneous abnormal deformation. Compared with traditional offline analysis or periodic sampling, this real-time processing method significantly shortens the time from anomaly occurrence to identification, effectively avoiding potential risks caused by reaction lag. At the same time, not only single abnormal points are identified, but also the evolution characteristics of the anomaly are analyzed and counted, and different types of anomalies are distinguished through classification marking and feature fingerprint extraction. This refined anomaly classification not only improves the accuracy of anomaly identification, but also provides more accurate input conditions for subsequent prediction and factor analysis.

[0113] In the prediction link, a hybrid architecture of deep learning is introduced, which combines long short-term memory network and attention mechanism to fuse feature data in different time windows. In this way, the model can not only capture long-term settlement trends, but also highlight the important influence of key time nodes on settlement development, thus generating more accurate and dynamic probability prediction results. Compared with traditional linear prediction methods, this prediction method can better cope with non-linear changes and uncertainties in the settlement process.

[0114] Meanwhile, by multivariate time series analysis, the contribution weight of each influencing factor to abnormal evolution is calculated, and combined with the initial analysis parameters, the spatial correlation is identified, which can reveal the coupling relationship between environmental factors, structural characteristics and settlement evolution. This analysis not only stays at the result judgment level, but also can deeply analyze the mechanism of settlement formation, so as to help managers accurately judge whether it is caused by geological conditions, water level fluctuation or structural design problem, and then provide a solid basis for formulating targeted control measures.

[0115] In the risk assessment and early warning link, by mapping the abnormal classification results and probability prediction results to a two-dimensional risk matrix, an intuitive and scientific risk classification system is established. The evaluation results can automatically trigger a graded warning, generating comprehensive decision support information including settlement trend curve, risk analysis report and emergency disposal suggestion. Such decision information is not just a warning prompt, but an action plan with all-round guiding significance, which can provide practical basis for on-site management and emergency response. For example, when the risk level is low, it can be recommended to strengthen daily monitoring and local detection; when the risk level is high, emergency disposal suggestions will be pushed and the on-site will be guided to start the corresponding emergency plan, ensuring that the risk is controllable first and the disposal is timely and effective.

[0116] In addition, when the risk level does not reach the warning standard, normal monitoring state can be maintained, so as to avoid excessive response and resource waste. Once the risk level reaches or exceeds the warning standard, the emergency plan is started immediately, and the high-frequency monitoring mode is automatically switched to, so as to continuously track the settlement state with higher sampling density and real-time performance. This phased dynamic management strategy not only ensures the economy and stability of the monitoring process, but also ensures quick response at critical moments, significantly improving the initiative and safety of building settlement management.

[0117] Overall, this process integrating collection, analysis, prediction, evaluation and response makes building settlement monitoring no longer limited to single-point detection or passive response, but develops into a dynamic, intelligent and full-cycle comprehensive management system. Not only can it significantly improve the real-time performance and reliability of settlement monitoring, but also can effectively reduce the safety risks and economic losses in the operation process of buildings.

[0118] Further, the implementation of the adaptive threshold configuration includes the following steps.

[0119] The settlement data and related environmental parameters of the last 30 days are divided into a statistical analysis window in chronological order.

[0120] Based on the statistical analysis window data, the key statistical indicators of building settlement and related environmental parameters are calculated to quantify the fluctuation characteristics of the current state of the building.

[0121] The abnormality detection threshold is set according to the calculated key statistical indicators to realize the hierarchical monitoring of abnormal deformations of different degrees.

[0122] The threshold range set initially is dynamically adjusted in combination with the current settlement state of the building and the seasonal variation characteristics, so that it can adapt to the changes of the building under different operating environments and time conditions.

[0123] The historical abnormal events are statistically analyzed to identify abnormal patterns and threshold adaptability problems, and the threshold configuration is optimized accordingly.

[0124] As can be seen from the above, the embodiment dynamically extracts key indicators to set abnormality detection thresholds through statistical analysis of settlement and environmental data in the past 30 days, and realizes hierarchical monitoring. On the basis of initial threshold setting, the threshold is dynamically adjusted in combination with the real-time state and seasonal variation characteristics of the building settlement, so that the threshold is more adaptable and flexible. At the same time, the threshold configuration is continuously optimized through retrospective analysis of historical abnormal events to avoid false positives and false negatives, thereby effectively improving the accuracy and robustness of abnormal monitoring, and maintaining stable and reliable monitoring effect under different environmental and time conditions.

[0125] Further, the duration analysis includes the following steps.

[0126] The instantaneous abnormality is continuously tracked, and the minimum duration threshold is set to 10 minutes and the maximum tracking time is set to 24 hours.

[0127] The deviation degree of the abnormal deformation amplitude relative to the normal deformation range is calculated to evaluate the abnormal intensity and determine the abnormal level.

[0128] The abnormal deformation data is fitted to analyze its development trend and determine whether it is in an increasing, decreasing or stable state to assist in the determination of potential risks.

[0129] The fluctuation degree of the abnormal deformation is evaluated through variance analysis to determine the stability and persistence of the abnormal signal.

[0130] When the duration of the abnormal deformation exceeds 30 minutes and shows an increasing trend, it is automatically marked as a persistent abnormality, triggering a deep analysis program, and increasing the sampling frequency of the related area sensors to obtain more detailed deformation information.

[0131] From the above, the embodiment realizes the comprehensive evaluation of the intensity, development trend and stability of abnormal deformation by continuous tracking and duration limitation of instantaneous abnormality, combining with abnormal amplitude deviation, trend fitting and fluctuation variance analysis. When the abnormality lasts more than the set threshold and shows an increasing trend, it can be automatically marked as a persistent abnormality and trigger in-depth analysis, while the local sensor sampling frequency is increased to obtain more detailed data, thereby effectively avoiding misjudgment caused by short-term disturbance and improving the accuracy of abnormality identification and the foresight of risk research.

[0132] Further, the feature fingerprint data extraction includes the following steps.

[0133] The multi-dimensional features of the building abnormal deformation data are fused, and the time domain, frequency domain and spatial distribution features are extracted, wherein: the time domain features include abnormal peak value, duration and change rate parameters; the frequency domain features obtain the frequency spectrum characteristics of abnormal deformation through fast Fourier transform; and the spatial distribution features analyze the propagation mode of the abnormality in each part of the building.

[0134] The extracted multi-dimensional features are integrated into a 128-dimensional feature vector to realize the structured representation and subsequent processing of abnormal deformation.

[0135] The principal component analysis method is used to reduce the dimension of the 128-dimensional feature vector to extract a 32-dimensional core feature.

[0136] The reduced core feature vector is stored in the abnormal deformation feature library, and historical abnormal events are classified and stored based on feature similarity.

[0137] The feature library and classification results are used to provide reference for abnormal deformation cause analysis and development trend prediction, and realize accurate monitoring of the abnormal state of the building.

[0138] From the above, the embodiment realizes efficient and structured representation of abnormal deformation by fusing time domain, frequency domain and spatial distribution features, constructing a 128-dimensional feature vector, and reducing the dimension to a 32-dimensional core feature through principal component analysis. The core feature is stored in the feature library and classified based on similarity to archive historical events, thereby forming a systematic abnormal information management, which not only improves the expression and identification ability of abnormal deformation, but also provides reliable data support for cause analysis and trend prediction, significantly enhancing the accuracy and traceability of building abnormal state monitoring.

[0139] Further, the duration of the instantaneous abnormal deformation is analyzed to obtain the duration feature, and the instantaneous abnormal deformation and the duration feature are combined to classify and mark the abnormal deformation, including the following steps.

[0140] A time window is established for each instantaneous abnormal deformation event to track its starting time and duration in real time, thereby quantifying the duration feature of the abnormal deformation.

[0141] Key duration parameters of abnormal deformation events are calculated, including duration, cumulative deformation amplitude, and trend change.

[0142] According to the deviation of abnormal deformation key duration parameters from the normal range, the abnormal intensity is evaluated and a reference is provided for classification.

[0143] Combined with the duration characteristics and intensity evaluation results, the abnormal deformation is classified, and it is determined whether the preset threshold is reached.

[0144] The classification results are marked in a structured manner, recording the abnormal type and characteristic parameters, and are synchronized to the deep learning prediction module and the multi-factor analysis module to update the abnormal deformation feature library.

[0145] As can be seen from the above, the embodiment can accurately quantify the duration characteristics of abnormal deformation by establishing a time window for instantaneous abnormal events and tracking in real time, while combining key parameters such as duration, cumulative amplitude, and trend change to evaluate the abnormal intensity. On this basis, the deviation of the abnormal state from the normal state is taken as the classification basis to ensure the scientificity and comparability of the classification results. After classification, the abnormal type and characteristic parameters are recorded in a structured manner, and are synchronized to the prediction and multi-factor analysis module to realize the dynamic improvement of the feature library.

[0146] Further, the implementation of the attention mechanism includes the following steps.

[0147] Map data from different time windows and different sensors to a unified feature space to achieve feature alignment.

[0148] Self-attention calculation is performed on a single sequence in the time dimension, and weights are assigned according to the correlation between different time points in the sequence to highlight the influence of key time nodes.

[0149] Cross-attention calculation is performed on different sensor features in the spatial dimension to analyze the correlation between sensors and identify spatial correlation patterns.

[0150] Attention weights are generated through matrix operations of query vectors, key vectors, and value vectors, and the results are normalized.

[0151] Based on the normalized attention weights, the importance of each feature is adjusted, and a weighted aggregation method is used to generate the fused feature representation.

[0152] As can be seen from the above, the embodiment realizes dynamic weighting and fusion of key features by performing self-attention and cross-attention calculation on sensor data in time and space dimensions. In the time dimension, the key time nodes that have the greatest impact on settlement prediction in the sequence are highlighted; in the space dimension, the correlation between sensors is analyzed, and important spatial patterns are identified. Normalized weights are generated through matrix operations on the query, key, and value vectors, and the importance of each feature is adjusted accordingly. Finally, the fused feature representation is generated in a weighted aggregation manner, thereby significantly enhancing the sensitivity and expression ability of the prediction model to key spatio-temporal information, and improving the accuracy and reliability of settlement trend prediction.

[0153] Further, the probability state quantitative prediction implementation step.

[0154] The Monte Carlo method is used to randomly sample the prediction model parameters to generate multiple potential parameter combinations to construct different prediction trajectories and quantify the prediction uncertainty.

[0155] The randomly sampled parameters are input into the prediction model to generate multiple settlement prediction trajectories covering different future states.

[0156] Statistical analysis is performed on the prediction trajectories to calculate the occurrence probability of different settlement levels, form a probability distribution, and extract key indicators, including mean prediction, confidence interval, and extreme event probability.

[0157] Model uncertainty, parameter uncertainty, and observation uncertainty are considered comprehensively during the prediction process, and the posterior distribution of the model parameters is updated through Bayesian inference.

[0158] The prediction results in the form of probability distribution are output as a quantitative reference for decision-making, including the average state, confidence interval range, and extreme settlement event probability.

[0159] As can be seen from the above, the embodiment generates multiple sets of prediction model parameters through Monte Carlo random sampling to construct multiple potential settlement trajectories, thereby quantifying the uncertainty of future states. Through statistical analysis of the trajectories, the occurrence probability of different settlement levels is calculated, a probability distribution is formed, and key indicators such as mean prediction, confidence interval, and extreme event probability are extracted. In this process, model, parameter, and observation uncertainties are considered simultaneously, and the posterior distribution of the model parameters is updated using Bayesian inference. The final output of the probabilistic prediction results provides a quantifiable reference for building settlement risk assessment and decision-making.

[0160] Further, the multivariate time series analysis includes the following steps.

[0161] The augmented Dickey-Fuller test is used to perform stationarity test on the time series data of each influencing factor to ensure that the data meets the stationarity requirement.

[0162] Cointegration test is performed on the stationary or differenced time series to identify the long-term equilibrium relationship between variables.

[0163] A multivariate time series model is constructed to analyze the dynamic influence of factors on the settlement response, and Granger causality test is used to analyze the strength of the causal relationship to identify the dynamic influence of each factor on the settlement change.

[0164] Maximum likelihood method is used to estimate the parameters of the multivariate time series model to obtain the regression coefficients of each factor and the residual characteristics of the model.

[0165] Variance decomposition method is used to calculate the explanation proportion of each factor to the settlement variation to comprehensively reflect the direct and indirect influence of the factor.

[0166] The dynamic causal relationship, contribution weight and settlement prediction result of each factor are output to provide quantitative basis for subsequent settlement risk assessment and management.

[0167] As can be seen from the above, the embodiment ensures that the factor data meets the modeling requirements through stationarity test and cointegration test, and identifies the long-term equilibrium relationship between variables. On this basis, a multivariate time series model is constructed, the dynamic influence of each factor on the settlement is analyzed by combining Granger causality test, the factor regression coefficients and residual characteristics are obtained by using maximum likelihood method for parameter estimation, the direct and indirect contribution of each factor to the settlement variation is calculated by variance decomposition, and finally the dynamic causal relationship, contribution weight and settlement prediction result are output, thereby providing quantitative basis for settlement risk assessment, realizing systematic analysis of the action mechanism of multiple factors, and improving the accuracy and reliability of monitoring and management.

[0168] Further, after multivariate time series analysis is completed, intelligent factor screening and weight optimization are further implemented, specifically including.

[0169] The information entropy and conditional information entropy of each factor are calculated, the information contribution of the factor to the settlement is quantified by mutual information, redundant factors with mutual information less than 0.1 are automatically removed, and model overcomplication is avoided.

[0170] A three-layer factor importance evaluation system is established, the first layer is direct influence evaluation, the direct action strength is quantified by linear regression, the second layer is indirect influence evaluation, the indirect effect of the factor through other variables is identified by path analysis, and the third layer is interactive influence evaluation, the nonlinear interaction mode between factors is identified by regression tree method.

[0171] Improved particle swarm optimization algorithm is used to optimize the factor weight in real time, and the objective function is the weighted combination of minimum prediction error and optimal model complexity.

[0172] When the weight of a factor exceeds 1.5 times the average weight for 3 consecutive time steps, it is automatically marked as a current key factor, and the sampling frequency of the factor is increased to 2 times the normal frequency within the next 6 hours, while the factor is assigned an additional 20% weight gain in the coupling modeling, ensuring that key influences are monitored.

[0173] As can be seen from the above, through intelligent factor screening and weight optimization, the accuracy and model efficiency of settlement prediction are significantly improved. First, low-contribution factors are removed based on mutual information to reduce redundancy and avoid model over-complexity; then a three-layer importance evaluation system is constructed to comprehensively analyze the direct, indirect and interactive effects of factors, ensuring the comprehensiveness of influence relationship identification. Finally, improved particle swarm optimization is used to realize dynamic adjustment of weights, and after identifying key factors, the sampling frequency and coupling weight of the key factors are increased to realize key monitoring of important factors, thereby balancing prediction accuracy, real-time performance and system stability.

[0174] Further, the spatial correlation analysis includes the following steps.

[0175] A spatial weight matrix is constructed based on the spatial positions and structural connectivity of each part of the building to describe the spatial relationship between the parts.

[0176] Moran's I index is used to perform spatial autocorrelation analysis on the settlement data to quantify the spatial aggregation degree of the settlement and identify hot and cold areas of the settlement.

[0177] A regression model considering spatial effects is established to analyze the relationship between the influencing factors and the settlement, and to eliminate the interference of spatial correlation on parameter estimation.

[0178] The independent influence of a single factor on the settlement is evaluated through main effect analysis, and the synergistic or antagonistic effects between factors are identified through interaction effect analysis, thereby revealing the complex coupling relationship of multiple factors.

[0179] The results of spatial autocorrelation analysis, spatial regression and multi-factor coupling analysis are integrated to provide a basis for building settlement mechanism analysis, risk identification and control strategy formulation.

[0180] As can be seen from the above, the embodiment quantifies the spatial aggregation characteristics of the settlement of each part of the building by constructing a spatial weight matrix and applying Moran's I index to identify hot and cold areas of the settlement. On this basis, the relationship between the influencing factors and the settlement is analyzed by combining a spatial effect regression model to eliminate the interference of spatial correlation on parameter estimation, and the independent action and synergistic or antagonistic action between factors are revealed through main effect and interaction effect analysis to comprehensively depict the coupling relationship of multiple factors, thereby enabling systematic analysis of the settlement mechanism and clear spatial distribution characteristics to provide a scientific basis for risk identification and control strategy formulation.

[0181] Further, based on the spatial correlation analysis results, the key factor action mechanism is established, and the identification and modeling of important influencing factors are strengthened through the following means.

[0182] The recursive feature elimination algorithm combined with cross-validation is used to dynamically evaluate the importance of each factor to the settlement prediction, calculate the factor significance score, mark the top 30% of the factors with the highest scores as key factors, and increase their weights in the coupled modeling.

[0183] The kernel principal component analysis and manifold learning algorithm are introduced to mine the nonlinear coupling relationship between factors, map the factor space to a high-dimensional feature space through Gaussian radial basis function kernel mapping, and identify complex coupling patterns that cannot be found by linear analysis.

[0184] An adaptive coupling strength adjustment mechanism is established to dynamically adjust the coupling coefficients between factors according to the settlement development stage (initial, development, stable, and acceleration), automatically increase the coupling weight of key environmental factors by 40-60% in the settlement acceleration period, and ensure that the key factors play a full role in the key stage.

[0185] The variance decomposition method is used to quantify the independent action, two-way interaction and high-order interaction of each factor, and establish a synergistic effect matrix. When the synergistic effect exceeds 50% of the individual effect, a factor combination model is automatically established and the combination weight is increased.

[0186] As can be seen from the above, the identification and expression of key factors are strengthened through spatial correlation analysis and advanced modeling methods. First, the highest contribution factors are selected by recursive feature elimination and cross-validation, and their weights are increased in modeling; then the nonlinear coupling relationship is mined through kernel principal component analysis and manifold learning to find complex patterns that cannot be identified by conventional methods; at the same time, the coupling coefficients between factors are dynamically adjusted according to the settlement stage to highlight the role of environmental factors in the key stage; finally, the independent and synergistic effects between factors are quantified through variance decomposition, and the combination weight is increased for the combination with significant synergistic effect, ensuring that the model fully and accurately reflects the settlement mechanism.

[0187] Further, according to the analysis of initial parameters, environmental parameter data and probability state quantitative prediction results, spatial correlation analysis is performed to identify the multi-factor coupling relationship between environmental factors, structural characteristics and abnormal evolution process, including the following steps.

[0188] According to the spatial position and structural connectivity of each part of the building, a spatial weight matrix is constructed to describe the spatial relationship between different monitoring points.

[0189] The analysis of initial parameters, environmental parameter data and probability state prediction results is unified and coded to facilitate the unified processing of multiple factors in spatial correlation analysis.

[0190] The spatial autocorrelation index is used to quantify the aggregation characteristics of the settlement of each part of the building, and to identify abnormal settlement hot and cold spot areas, thereby providing a reference for causal relationship analysis.

[0191] Based on the constructed spatial weight matrix, a regression model considering spatial effects is established to analyze the influence of environmental factors and structural characteristics on the evolution of abnormal settlement, while eliminating the interference of spatial dependence on parameter estimation.

[0192] Through main effect analysis and interaction effect analysis, the direct action and synergistic action of each environmental factor and structural characteristic on the abnormal evolution process are quantitatively evaluated, and the multi-factor coupling mechanism is revealed.

[0193] The spatial correlation analysis and coupling relationship identification results are structured and output, thereby providing quantitative basis for the multi-factor analysis module and risk assessment module, and updating the analysis parameters of the monitoring device.

[0194] As can be seen from the above, the embodiment integrates and analyzes the initial parameters, environmental data and probability state prediction results, constructs a spatial weight matrix in combination with the spatial positions and structural connectivity of each part of the building, and realizes the description of the spatial relationship between the monitoring points. First, all kinds of data are uniformly arranged and coded, so as to realize unified processing of multiple factors in spatial correlation analysis, and then the spatial autocorrelation index is used to quantify the settlement aggregation characteristics, identify abnormal hot and cold spot areas, and provide a reference for causal relationship analysis. Based on the spatial weight matrix, a regression model considering spatial effects is established to analyze the influence of environmental factors and structural characteristics on the evolution of abnormal settlement, while eliminating the interference of spatial dependence on parameter estimation. Further, through main effect and interaction effect analysis, the direct action and synergistic or antagonistic action of each factor are quantitatively evaluated, and the multi-factor coupling mechanism in the abnormal evolution process is revealed. Finally, the spatial correlation and coupling analysis results are structured and output, thereby providing reliable quantitative basis for the multi-factor analysis and risk assessment module, and dynamically updating the analysis parameters of the monitoring device, realizing fine analysis and scientific management of the building settlement state.

[0195] Further, the two-dimensional risk matrix is constructed with the settlement occurrence probability as the horizontal axis and the settlement consequence severity as the vertical axis, wherein the horizontal axis is divided into five levels of extremely low, low, medium, high and extremely high, and the vertical axis is divided into five levels of slight, general, serious, major and super major, thereby forming 25 risk units; the risk mapping process converts the quantitative prediction results into risk levels through a fuzzy membership function, wherein the probability level is determined according to the cumulative probability distribution of the prediction, and the consequence level is determined according to the influence degree of the settlement on the safety, use function and economic value of the building structure.

[0196] As can be seen from the above, the embodiment divides the building settlement risk into 25 units by taking the settlement occurrence probability as the horizontal axis and the settlement consequence severity as the vertical axis, to realize quantitative risk grading management. The horizontal axis is divided into five levels of extremely low, low, medium, high and extremely high, and the vertical axis is divided into five levels of slight, general, serious, major and super major, to form a clear risk grid. Through the fuzzy membership function, the probability distribution result predicted by deep learning is mapped to the corresponding risk level, and the consequence level is evaluated according to the influence of settlement on the structural safety, use function and economic value, so as to effectively convert the quantitative prediction result into an operable risk level.

[0197] Further, the evaluation benchmark parameters include four core parameters of a settlement rate benchmark value, a cumulative settlement benchmark value, a prediction trend coefficient and an environmental influence coefficient, wherein: the settlement rate benchmark value is obtained from the specification standard according to the building type and the geological condition; the cumulative settlement benchmark value considers the design allowable deformation value and the service life of the building; the prediction trend coefficient reflects the acceleration characteristics of settlement development; and the environmental influence coefficient quantifies the amplification effect of external environmental factors on settlement.

[0198] As can be seen from the above, the embodiment realizes scientific quantitative management of the building settlement risk through the four core indexes of the settlement rate benchmark value, the cumulative settlement benchmark value, the prediction trend coefficient and the environmental influence coefficient. The settlement rate benchmark value is formulated according to the building type and the geological condition, to ensure that the evaluation conforms to the actual engineering characteristics; the cumulative settlement benchmark value reflects the total settlement capacity that the structure can withstand in combination with the design allowable deformation and the service life; the prediction trend coefficient depicts the acceleration change of settlement development, to reveal the potential aggravating risk; and the environmental influence coefficient quantifies the amplification effect of external factors on settlement, to realize comprehensive consideration of environmental coupling influence. The overall parameter system provides a scientific and quantifiable judgment basis for risk evaluation and early warning, and improves the accuracy and reliability of building settlement management.

[0199] Further, the settlement trend curve is displayed by using a multi-scale time axis, including a real-time curve, a daily trend curve, a weekly trend curve and a monthly trend curve, wherein: the real-time curve displays the settlement change in the last 24 hours, the daily trend curve displays the daily average settlement amount in the last 30 days, the weekly trend curve displays the weekly average settlement amount in the last 12 weeks, and the monthly trend curve displays the monthly average settlement amount in the last 12 months; the Y-axis scale is automatically adjusted according to the data change range, to highlight the detail characteristics of the settlement change, and the uncertainty range of the prediction result is indicated by a shaded area, and the prediction accuracy index is marked in numerical form on the chart; the risk analysis report adopts a structured format, including six parts of monitoring overview, abnormal analysis, risk evaluation, development trend, influence factor analysis and suggested measures, and the report content is automatically generated according to the current settlement state and the risk level, to ensure the comprehensiveness and practicality of the decision support information.

[0200] From the above, the embodiment realizes all-round visual management of building settlement through multi-scale time axis. Real-time curve presents the settlement change in the last 24 hours, and daily, weekly and monthly trend curves reflect the average settlement in the last 30 days, 12 weeks and 12 months respectively, realizing the combination of short-term dynamic monitoring and long-term evolution analysis. The Y axis is automatically adjusted according to the data range, and the details of the change are refined. The shadow area represents the prediction uncertainty, and the prediction accuracy index is labeled, providing a quantitative reference for the monitoring results. The risk analysis report adopts a structured format, including six modules of monitoring overview, abnormal analysis, risk assessment, development trend, influence factor analysis and response suggestion. The content is automatically generated according to the current settlement state and risk level, realizing the visualization of settlement situation, the scientization of risk analysis and the comprehensiveness and practicality of decision support information.

[0201] Further, the generation of the emergency treatment suggestion comprises the following steps.

[0202] An expert knowledge base is established, covering standard treatment procedures and experience practices under different settlement types and risk levels; at the same time, a case reasoning system is constructed for storing historical emergency treatment cases for retrieval and comparison.

[0203] Based on the similarity matching method, the most similar treatment scheme to the current settlement risk is retrieved from the historical case library, and the candidate treatment scheme set is formed by combining the standard procedures in the knowledge base.

[0204] Considering the building structure characteristics, use function, surrounding environment and available resources, a multi-criteria decision analysis method is used to optimize the candidate scheme, and a feasible and effective treatment suggestion is selected.

[0205] The optimized emergency treatment suggestion is output to form a hierarchical and operable decision support result, providing quantitative guidance for emergency response and risk control.

[0206] From the above, the embodiment realizes scientific response to building settlement risk through the organic combination of expert knowledge base and historical case base. First, standard treatment procedures and experience practices under different settlement types and risk levels are established, and historical emergency cases are stored for retrieval and comparison. Through the similarity matching method, the most similar treatment scheme to the current risk is selected from the historical cases, and the candidate scheme set is generated by combining the standard procedures in the knowledge base. Then, considering the building structure characteristics, use function, surrounding environment and available resources, a multi-criteria decision analysis method is used to optimize the candidate scheme, forming a hierarchical and operable emergency treatment suggestion. The output suggestion is not only scientific and quantitative, but also operable, and can provide clear guidance for on-site emergency response and risk control, improving the timeliness, accuracy and decision reliability of building settlement management.

[0207] Further, based on the evaluation benchmark parameter and the two-dimensional risk matrix, automatic evaluation and grading early warning of the building structure evolution risk are realized, and comprehensive decision support information including the settlement trend curve, risk analysis report and emergency disposal suggestion is generated, including the following steps.

[0208] In combination with the evaluation benchmark parameter, each risk unit in the two-dimensional risk matrix is automatically evaluated to determine the corresponding risk grade.

[0209] According to the monitoring data and the prediction result, a multi-scale settlement trend curve is drawn to reflect the building structure evolution dynamics and potential risk changes.

[0210] The risk evaluation result, abnormal event record and trend information are integrated to automatically generate a structured risk analysis report.

[0211] Based on the expert knowledge base and historical case reasoning, a hierarchical and operable emergency disposal suggestion is generated according to the current risk grade and structure state.

[0212] The settlement trend curve, risk analysis report and emergency disposal suggestion are integrated into comprehensive decision support information, which is pushed to the management terminal in real time to realize the risk early warning and decision assistance closed loop.

[0213] As can be seen from the above, the embodiment realizes risk quantification and grade division through the evaluation benchmark parameter and the two-dimensional risk matrix. The risk units are automatically evaluated to determine the risk grade, and a multi-scale settlement trend curve is drawn based on the monitoring data and the prediction result to comprehensively reflect the building structure evolution dynamics and potential risk changes. Meanwhile, the risk evaluation result, abnormal event record and trend information are integrated to generate a structured risk analysis report. In combination with the expert knowledge base and historical case reasoning, a hierarchical and operable emergency disposal suggestion is generated according to the current risk grade and structure state to ensure the feasibility and effectiveness of the emergency response scheme. Finally, the settlement trend curve, risk analysis report and emergency disposal suggestion are integrated into comprehensive decision support information, which is pushed to the management terminal in real time to realize the closed loop management from monitoring, analysis, early warning to emergency disposal.

[0214] Finally, it should be pointed out that: the above embodiments are only used to illustrate the technical solutions of the present application, but not to limit them. Although the present application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that they can still modify the technical solutions recorded in the foregoing embodiments, or make equivalent replacement to part of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A device for real-time monitoring of settlement of a building ground, characterized in that, The system comprises the following modules: A hierarchical sensor network for real-time data collection at multiple time scales, comprehensive acquisition of building structure deformation data and related environmental parameters; A data processing and analysis module for real-time processing and intelligent analysis of collected data, timely identification and classification of abnormal deformation characteristics of building structures; A deep learning prediction module for quantitative prediction of the probability state and evolution trend of building settlement by constructing a multi-scale time series prediction model; A multi-factor analysis module for coupled modeling and comprehensive analysis of environmental factors, structural characteristics and abnormal evolution processes to identify key influencing factors and their mechanisms; A risk assessment and early warning module for grading assessment of building settlement risk based on prediction and analysis results, and generation of corresponding warning information and decision support schemes.

2. The real-time settlement monitoring device for building ground according to claim 1, characterized in that, The hierarchical sensor network comprises a basic layer sensor array, a structural layer sensor node and an environmental monitoring sensor group, wherein: the basic layer sensor array adopts a combination of high-precision laser displacement sensors and digital tilt sensors, and is installed at key parts of the building foundation according to the grid arrangement principle; the structural layer sensor node adopts wireless strain sensors and three-axis acceleration sensors, and is arranged along the main load-bearing structure of the building to realize real-time monitoring of structural stress state changes and dynamic responses; the environmental monitoring sensor group includes soil moisture sensors, underground water level monitoring sensors, temperature and humidity sensors and ground vibration sensors for obtaining external environmental parameters affecting building settlement.

3. The real-time settlement monitoring device for building ground according to claim 2, characterized in that, The data processing and analysis module comprises a data preprocessing unit, a feature extraction unit and an anomaly detection unit, wherein: the data preprocessing unit is used for noise filtering and signal enhancement processing of original sensor data, and simultaneously corrects sensor drift and suppresses environmental interference; the feature extraction unit extracts multi-frequency domain features based on wavelet transform, and reduces data dimension by combining principal component analysis method, thereby extracting key feature parameters representing building deformation state; the anomaly detection unit establishes dynamic control limits using statistical process control method, identifies abnormal deformation patterns by comparing with historical baseline data, triggers abnormal warning when three consecutive sampling points exceed the control limits, and automatically adjusts detection sensitivity according to different settlement development stages.

4. The real-time settlement monitoring device for building ground according to claim 3, characterized in that, The deep learning prediction module adopts a hybrid architecture combining long short-term memory neural network and attention mechanism, captures the time sequence dependence of settlement data by constructing a multi-layer LSTM network, and the attention mechanism is used to adaptively allocate weights to data at different time steps, thereby improving the prediction accuracy of key time nodes; the deep learning prediction module comprises a data normalization layer, a feature encoding layer, a time sequence modeling layer and an output decoding layer, wherein: the data normalization layer performs standardization processing on the input multi-dimensional time series data, the feature encoding layer converts sensor data into a high-dimensional feature vector, the time sequence modeling layer learns the dynamic pattern of settlement evolution through a bidirectional long short-term memory neural network, and the output decoding layer generates the prediction results of settlement probability distribution in the future time window.

5. The real-time settlement monitoring device for building ground according to claim 4, characterized in that, The multi-factor analysis module comprises an environmental factor influence evaluation unit, a structural property analysis unit and a settlement mechanism identification unit, wherein: the environmental factor influence evaluation unit is used for quantitatively analyzing the action relationship of various environmental factors on settlement and establishing a response function and a sensitivity coefficient thereof; the structural property analysis unit evaluates the mechanical state and safety reserve level of the building structure through structural mechanics modeling and measured data correction; and the settlement mechanism identification unit identifies the dominant factor and triggering condition by data mining and pattern recognition analysis of the spatio-temporal evolution characteristics of settlement, extracts a typical settlement mode and establishes a discrimination criterion.

6. The method of using the real-time building ground settlement monitoring device, based on the real-time building ground settlement monitoring device of claim 5, wherein, The method comprises the following steps: According to the initial state of the settlement monitoring device, the key positions of the building and the preset monitoring accuracy requirements, the sampling frequency configuration of each sensor is determined, and a multi-time scale data acquisition scheme of the hierarchical sensor network is generated in combination with the key positions of the building and the sampling frequency configuration; The hierarchical sensor network is controlled to collect building structure deformation data and environmental parameter data in real time according to the multi-time scale data acquisition scheme; Each time the hierarchical sensor network completes a round of data acquisition, the currently collected structure deformation data is marked as target monitoring data, the abnormal detection parameters of the data processing and analysis module are configured based on the target monitoring data and a preset adaptive threshold value, so as to perform real-time abnormal identification on the target monitoring data and identify the instantaneous abnormal deformation of the building structure; In the real-time abnormal identification process, the duration characteristics of the instantaneous abnormal deformation are statistically analyzed in real time, the abnormal deformation is classified and marked in combination with the instantaneous abnormal deformation and the duration characteristics, and the corresponding feature fingerprint data are extracted; If the duration of the instantaneous abnormal deformation does not reach a preset classification standard, the real-time abnormal identification is maintained until the abnormal classification is completed; If the instantaneous abnormal deformation has been classified and marked, the abnormal identification process of the target monitoring data is ended, and the classification result is transmitted to the deep learning prediction module; By fusing feature data of different time windows and introducing an attention mechanism, the probability state quantitative prediction of the evolution of building abnormal deformation to continuous settlement is realized; Based on multivariate time series analysis, the contribution weight of each influence factor to abnormal evolution is calculated, and the analysis initial parameters of the multi-factor analysis module are obtained; According to the analysis initial parameters, the environmental parameter data and the probability state quantitative prediction result, spatial correlation analysis is performed to realize multi-factor coupling relationship identification of environmental factors, structural properties and abnormal evolution process; The abnormal classification result and the probability state quantitative prediction result are mapped to a two-dimensional risk matrix, and the evaluation benchmark parameters of the risk assessment and early warning module are obtained; Based on the evaluation benchmark parameters and the two-dimensional risk matrix, the automatic assessment and hierarchical early warning of the evolution risk of the building structure are realized, and comprehensive decision support information including a settlement trend curve, a risk analysis report and an emergency disposal suggestion is generated; In the process of generating the comprehensive decision support information, a warning notification is pushed in real time, and the risk analysis report and the emergency disposal suggestion are combined to guide the on-site emergency response. If the building structure evolution risk level does not reach the early warning standard, the normal monitoring state is maintained and data collection is continued; If the building structure evolution risk level has reached the early warning standard, an emergency plan is immediately started, and the building ground real-time settlement monitoring device is controlled to enter a high-frequency monitoring mode.

7. The method of using a device for monitoring real-time settlement of a building ground according to claim 6, wherein, The duration analysis is used to statistically analyze the duration characteristics of the instantaneous abnormal deformation in real time, and the abnormal deformation is classified and labeled according to the instantaneous abnormal deformation and the duration characteristics, including the following steps: A time window is established for each instantaneous abnormal deformation event, and the starting time and duration are tracked in real time, so as to quantify the duration characteristics of the abnormal deformation; Key duration parameters of the abnormal deformation event are calculated, including the duration, cumulative deformation amplitude and trend change; According to the deviation of the key duration parameters of the abnormal deformation from the normal range, the abnormal intensity is evaluated and a reference is provided for classification; The abnormal deformation is classified according to the duration characteristics and the evaluation results of the intensity, and it is determined whether the preset threshold is reached; The classification results are labeled in a structured manner, the abnormal type and characteristic parameters are recorded, and the deep learning prediction module and the multi-factor analysis module are synchronized to update the abnormal deformation feature library.

8. The method of using a device for monitoring real-time settlement of a building ground according to claim 6, wherein, The probability state quantitative prediction implementation steps are as follows: Monte Carlo method is used to randomly sample the prediction model parameters to generate multiple potential parameter combinations, to construct different prediction trajectories and quantify the prediction uncertainty; The parameters obtained by random sampling are input into the prediction model to generate multiple settlement prediction trajectories covering different future states; Statistical analysis is performed on the prediction trajectories to calculate the occurrence probability of different settlement levels, form a probability distribution and extract key indicators, including mean prediction, confidence interval and extreme event probability; In the prediction process, the model uncertainty, parameter uncertainty and observation uncertainty are considered comprehensively, and the posterior distribution of the model parameters is updated through Bayesian inference; The prediction results in the form of probability distribution are output as a quantitative reference for decision-making, including the average state, confidence interval range and extreme settlement event probability.

9. The method of using a device for monitoring real-time settlement of a building ground according to claim 6, wherein, According to the analysis of the initial parameters, the environmental parameter data and the probability state quantitative prediction results, spatial correlation analysis is performed to identify the multi-factor coupling relationship between environmental factors, structural characteristics and abnormal evolution process, including the following steps: A spatial weight matrix is constructed according to the spatial positions and structural connectivity of different parts of the building, which is used to describe the spatial relationship between different monitoring points; The analysis initial parameters, environmental parameter data and probability state prediction results are uniformly arranged and coded to facilitate the unified processing of multiple factors in the spatial correlation analysis; The spatial autocorrelation index is used to quantify the aggregation characteristics of the settlement of each part of the building, to identify abnormal settlement hot and cold point areas, and to provide a reference for causal relationship analysis; Based on the constructed spatial weight matrix, a regression model considering spatial effects is established to analyze the influence of environmental factors and structural characteristics on abnormal settlement evolution, while eliminating the interference of spatial dependence on parameter estimation; Through main effect analysis and interaction effect analysis, the direct effect and synergistic effect of each environmental factor and structural characteristic on the abnormal evolution process are quantitatively evaluated, and the multi-factor coupling mechanism is revealed. The spatial correlation analysis and coupling relationship recognition results are structured and output, providing quantitative basis for the multi-factor analysis module and the risk assessment module, and updating the analysis parameters of the monitoring device.

10. The method of using a device for monitoring real-time settlement of a building ground according to claim 6, wherein, Based on the evaluation benchmark parameters and the two-dimensional risk matrix, automatic evaluation and hierarchical early warning of the building structure evolution risk are realized, and comprehensive decision support information including the settlement trend curve, risk analysis report and emergency disposal suggestion is generated, including the following steps: According to the evaluation benchmark parameters, each risk unit in the two-dimensional risk matrix is automatically evaluated to determine the corresponding risk level; According to the monitoring data and the prediction results, a multi-scale settlement trend curve is drawn to reflect the building structure evolution dynamics and potential risk changes; The risk assessment results, abnormal event records and trend information are integrated to automatically generate a structured risk analysis report; Based on the expert knowledge base and historical case reasoning, hierarchical and operable emergency disposal suggestions are generated according to the current risk level and structure state; The settlement trend curve, risk analysis report and emergency disposal suggestion are integrated into comprehensive decision support information, which is pushed to the management terminal in real time to realize the risk early warning and decision assistance closed loop.

Images