An intelligent building monitoring management method based on multi-dimensional data and data analysis

Abstract

The application relates to the technical field of intelligent building monitoring and discloses an intelligent building monitoring management method based on multidimensional data and data analysis. The method synchronously collects original heterogeneous data of multiple sensing networks in a building, carries out data fusion and structure restoration processing, and constructs a building digital mirror body with spatiotemporal consistency. On this basis, dynamic response features and static intrinsic features of the structure are separated and extracted, and two types of data sets, namely time-varying and steady-state data sets, are obtained. By progressively matching the time-varying features with historical failure feature maps in a knowledge base, an abnormal evolution mode is identified, the performance decay trajectory of the building structure in multiple maintenance cycles in the future is deduced, and an active intervention scheme accurately corresponding to a time point is generated. The method realizes the whole process from unified modeling of multiple source heterogeneous data to intelligent prediction of structure performance decay, and supports the change of building monitoring from post-alarm to pre-prediction and accurate maintenance.

Description

Technical Field

[0001] This invention relates to the field of intelligent building monitoring technology, specifically to an intelligent building monitoring and management method based on multidimensional data and data analysis. Background Technology

[0002] Current building structural health monitoring systems generally rely on deploying multiple sensor networks, such as vibration, strain, temperature and humidity sensors, and video surveillance, to collect data on the building's operational status. The primary operational model of these systems remains focused on real-time display and storage of independent data from each sensor network, along with alarms based on simple thresholds. This model treats data from different physical quantities, sampling rates, and spatiotemporal references as isolated information streams, only capable of capturing transient anomalies in specific parameters.

[0003] This conventional technical approach has its flaws. Due to the lack of deep fusion of multi-source heterogeneous data and the construction of a unified spatiotemporal benchmark, monitoring results are fragmented, making it difficult to comprehensively and accurately reflect the true state of the building structure as a complete system. Its analysis remains superficial, failing to reveal the underlying structural behavior patterns and evolutionary laws hidden behind the data. Such systems are essentially passively responsive, only issuing alarms after anomalies or failures occur or reach a critical level, unable to identify potential risk evolution trends in advance, and unable to provide a basis for predictive maintenance decisions. The core issues that need to be addressed are: how to construct a unified model from discrete heterogeneous data that faithfully reflects the entire physical entity of the building, and how to achieve early identification and forward-looking prediction of structural performance degradation processes based on this model. Summary of the Invention

[0004] The purpose of this invention is to provide an intelligent building monitoring and management method based on multidimensional data and data analysis to solve the problems mentioned in the background art.

[0005] To achieve the above objectives, the present invention provides an intelligent building monitoring and management method based on multidimensional data and data analysis, the method comprising:

[0006] Simultaneously acquire raw heterogeneous data streams within a monitoring time window from multiple designated sensor networks of the building;

[0007] Data fusion and structure restoration processing are performed on the original heterogeneous data stream to construct a digital mirror of the building with complete spatiotemporal consistency;

[0008] Based on the digital mirror of the building, the dynamic load response characteristics and static structural intrinsic characteristics are separated and extracted to separate the time-varying feature dataset and the steady-state feature dataset.

[0009] The time-varying feature dataset is progressively pattern matched with a historical failure feature map in a knowledge base to identify abnormal evolution patterns of structural response.

[0010] Based on the abnormal evolution pattern and the steady-state feature dataset, the performance degradation trajectory of the building structure in multiple future maintenance cycles can be deduced.

[0011] Based on the performance degradation trajectory, an active intervention plan is generated that precisely matches future time points.

[0012] Preferably, the step of performing data fusion and structural restoration processing on the original heterogeneous data stream to construct a digital mirror of the building with complete spatiotemporal consistency includes the following steps:

[0013] For sensor data from different sources in the original heterogeneous data stream, a unified timestamp alignment and spatial registration benchmark is established to map all data to the same spatiotemporal coordinate system of the building information model.

[0014] Based on the physical properties and spatial topological relationships of the sensor data, a multi-source data correlation matrix is ​​constructed. This multi-source data correlation matrix is ​​used to describe the coupling and transmission relationships between vibration, stress, temperature, and humidity data.

[0015] Based on the multi-source data association matrix, missing value imputation and outlier correction are performed on the data with spatiotemporal correlation to generate cleaned continuous spatiotemporal field data.

[0016] Using the continuous spatiotemporal field data as boundary and initial conditions, a preset building physical property calculation model is driven to solve the distribution of structural parameters inside the building that cannot be directly measured in reverse.

[0017] The structural parameter distribution obtained by the solution is deeply integrated with the building information model to form a three-dimensional computable model that can reflect the internal state of the structure in real time. The three-dimensional computable model is the digital mirror of the building.

[0018] Preferably, the separation and extraction process of dynamic load response characteristics and static structural intrinsic characteristics includes the following steps:

[0019] A set of preset virtual excitation signals are applied to the digital image of the building, and the theoretical response field of the digital image of the building under the virtual excitation signals is calculated.

[0020] Extract the actual response field data corresponding to the actual environmental load from the continuous spatiotemporal field data;

[0021] Calculate the residual field between the theoretical response field and the actual response field data, wherein the residual field characterizes the response deviation caused by changes in the local performance of the structure;

[0022] The residual field is decomposed into a multi-scale spatial decomposition to separate the macro-scale residual components related to the overall structural stiffness and the micro-scale residual components related to local damage and connection state.

[0023] The macro-scale residual components and the micro-scale residual components are subjected to time-domain feature extraction to form the time-varying feature dataset describing the overall time-varying characteristics of the structure.

[0024] From the structural parameter distribution of the building digital mirror, material property parameters and geometric connection parameters that remain relatively stable within the monitoring time window are extracted to form the steady-state feature dataset.

[0025] Preferably, the progressive pattern matching of the time-varying feature dataset with a historical failure feature map in a knowledge base includes the following steps:

[0026] Historical monitoring cases with the same or similar structural type as the target building are retrieved from the knowledge base, and the complete feature evolution sequence from normal state to failure state in each case is extracted to form the historical failure feature map.

[0027] The feature vectors in the time-varying feature dataset are arranged in chronological order to form a feature evolution sequence to be identified;

[0028] The dynamic time warping algorithm is used to calculate the similarity distance between the feature evolution sequence to be identified and the early, middle and late stage features of each historical failure feature map in the knowledge base;

[0029] Based on the similarity distance, select several target historical failure feature maps that have the highest matching degree with the feature evolution sequence to be identified, and extract their subsequent evolution paths;

[0030] Based on the difference between the evolution path of the target's historical failure feature map and the current feature evolution sequence to be identified, the abnormal evolution mode that the current structure may exhibit in the future is predicted through interpolation and extrapolation calculations. The abnormal evolution mode includes damage type, location, and diffusion direction.

[0031] Preferably, the step of deducing the performance degradation trajectory of the building structure over multiple future maintenance cycles based on the abnormal evolution pattern and the steady-state feature dataset includes the following steps:

[0032] Establish a performance degradation prediction model with the steady-state feature dataset as the initial state and the abnormal evolution mode as the driving rule;

[0033] In the performance degradation prediction model, the time axis is divided into multiple consecutive small time steps. Within each small time step, the parameters of the corresponding structural parts are updated according to the damage propagation law described by the abnormal evolution mode.

[0034] By combining the statistical distribution of environmental loads, random load disturbances conforming to probability distributions are introduced into the performance degradation prediction model to simulate the uncertainties in actual use.

[0035] Run the performance degradation prediction model and iteratively calculate until the preset performance failure threshold is reached. Record the complete curve of the overall structural performance index changing over time during the entire simulation process.

[0036] From the complete curve, the predicted performance index values ​​corresponding to different future maintenance cycle time points are extracted, and the time points when the performance index first falls below the warning threshold of each level are marked. The performance degradation trajectory is defined by these time points and the corresponding predicted performance index values.

[0037] Preferably, generating an active intervention plan that precisely matches future time points based on the performance degradation trajectory includes the following steps:

[0038] Analyze the performance degradation trajectory to determine the specific future time points when the predicted value of the performance index falls below the warning threshold for each level;

[0039] For each specific future time point, a reverse analysis is performed on the set of key components and their parameter degradation amounts that lead to performance degradation in the performance degradation prediction model.

[0040] Based on the set of key components and their parameter degradation, executable intervention measures are matched from the measure library. These intervention measures include, but are not limited to, applying prestress, adding temporary supports, replacing local components, and applying protective coatings.

[0041] For each set of matched intervention measures, evaluate its implementation effect in the performance degradation prediction model, that is, the degree of recovery and maintenance time of structural performance indicators after implementation;

[0042] By combining engineering costs and building usage constraints, multi-objective optimization is performed on intervention measures that can effectively maintain performance at various future time points. For each optimal intervention measure, a precise implementation time window, specific operating parameters, and expected performance maintenance period are specified, forming a series of time-triggered proactive intervention schemes.

[0043] Preferably, the step of driving a preset building physics property calculation model to inversely solve for the distribution of structural parameters inside the building that cannot be directly measured includes the following steps:

[0044] Construct an inverse physical equation model with the structural parameter distribution as the variable to be solved and the continuous spatiotemporal field data as the known observations;

[0045] In the inverse problem model of the physical equations, a regularization constraint term reflecting the smoothness of the spatial distribution of structural parameters is introduced;

[0046] An iterative optimization algorithm is used to minimize the difference between the predicted field data generated by the building physical property calculation model and the continuous spatiotemporal field data, while satisfying the regularization constraint.

[0047] In each iteration, based on the difference between the prediction and the actual measurement, the gradient of the objective function with respect to the variable to be solved is calculated using the adjoint state method, and the distribution of the structural parameters is updated along the gradient descent direction;

[0048] When the difference between the predicted field data and the measured continuous spatiotemporal field data is less than the preset tolerance, or when the number of iterations reaches the upper limit, the calculation is terminated, and the final optimized parameter distribution is used as the structural parameter distribution.

[0049] Preferably, the construction of the inverse physical equation model, using the structural parameter distribution as the variable to be solved and the continuous spatiotemporal field data as the known observations, includes the following steps:

[0050] The building digital mirror is geometrically discretized into a finite number of elements, each element is assigned a set of structural parameters to be identified, and the structural parameters of all elements constitute the variables to be solved;

[0051] Establish control equations describing the mechanical response of the building, and use the variables to be solved as spatial distribution coefficients in the control equations;

[0052] Define an objective function, which is the norm of the difference between the theoretical response calculated by the governing equations under given boundary conditions and the observed values ​​of the corresponding physical quantities in the continuous spatiotemporal field data;

[0053] The regularization constraint term is added to the objective function. The regularization constraint term is the gradient of the variable to be solved in space or the norm of the Laplacian operator, which is used to suppress the ill-posedness of the solution.

[0054] The inverse problem model of the physical equation is expressed as: under the regularization constraint, finding the distribution of structural parameters that minimizes the objective function.

[0055] Preferably, the method further includes incremental learning and updating of the knowledge base, specifically including:

[0056] In the subsequent monitoring of the target building structure, the actual structural state evolution data and the effect feedback data of the implemented active intervention scheme are recorded;

[0057] When the actual structural state evolution deviates significantly from the predicted abnormal evolution pattern or the performance degradation trajectory, this complete monitoring, prediction, intervention, and feedback sequence will be treated as a new case.

[0058] The new case is subjected to feature extraction and standardization to generate a new feature evolution sequence and the corresponding actual evolution endpoint label;

[0059] The processed new cases are stored in the knowledge base, and together with the original historical failure feature map, they form an updated knowledge base.

[0060] The updated knowledge base is used in the next pattern matching execution to achieve continuous evolution of the model's predictive capabilities.

[0061] Preferably, the step of employing a dynamic time warping algorithm to calculate the similarity distance between the feature evolution sequence to be identified and the early, middle, and late stage features of each historical failure feature map in the knowledge base includes the following steps:

[0062] From each historical failure feature map in the knowledge base, three feature subsequences representing the early, middle, and late stages of failure development are extracted according to a preset time division ratio.

[0063] The feature evolution sequence to be identified is used as the matching sequence, and the early, middle and late feature subsequences of the historical failure feature map are used as three reference sequences respectively.

[0064] For the sequence to be matched and each of the reference sequences, a dynamic time warping algorithm is executed. By constructing a cumulative cost matrix and finding the optimal warping path, the minimum cumulative cost required to nonlinearly stretch or compress the sequence to be matched on the time axis to best match the reference sequence is calculated.

[0065] The minimum cumulative cost is quantified as the similarity distance between the two, wherein the sequence to be matched and the early, middle and late feature subsequences of the historical failure feature map are calculated respectively, and finally three similarity distance values ​​representing the degree of matching with the early, middle and late stages are obtained respectively.

[0066] Compared with the prior art, the beneficial effects of the present invention are:

[0067] Data fusion and structural reconstruction are performed on the original heterogeneous data streams to construct a digital mirror of the building with complete spatiotemporal consistency. Original stream data collected from different sensor networks, differing in data format, spatiotemporal reference, and physical meaning, are integrated into a unified and precisely corresponding digital model. This process eliminates the isolation and inconsistency between multi-source data, enabling the correlation and analysis of various state parameters of the building, such as vibration, strain, and temperature, within the same spatiotemporal coordinate system. This provides an internally logically consistent and globally traceable data entity foundation for all subsequent advanced analyses, replacing the scattered and fragmented data views found in traditional monitoring.

[0068] This method employs progressive pattern matching between time-varying feature datasets and historical failure feature maps in a knowledge base to identify abnormal evolution patterns and infer performance degradation trajectories over multiple future maintenance cycles. This surpasses instantaneous anomaly detection based on fixed thresholds by continuously comparing the current structural dynamic response with pre-stored failure development process characteristics to identify weak early signs and their evolution patterns representing specific damage types or degradation mechanisms. Based on the identified patterns and structural steady-state characteristics, the system can simulate and infer the degradation path and rate of structural performance indicators, thus transforming the monitoring output from current state alerts to quantitative predictions of future potential risks and their development timelines. Attached Figure Description

[0069] Figure 1 This is a schematic diagram illustrating the working principle of the intelligent building monitoring and management method based on multidimensional data and data analysis described in this invention.

[0070] Figure 2 A flowchart for constructing a digital mirror of a building;

[0071] Figure 3 The flowchart for separating and extracting dynamic and static features;

[0072] Figure 4 A diagram showing the decay trajectory and early warning node prediction of the remaining bearing capacity of a steel box girder bridge deck due to fatigue damage.

[0073] Figure 5 A heat map showing the spatiotemporal distribution of performance degradation risk for key components of a steel box girder bridge. Detailed Implementation

[0074] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0075] Please see Figure 1 This invention provides an intelligent building monitoring and management method based on multidimensional data and data analysis. The method includes: synchronously collecting raw data streams within a predetermined monitoring time window from multiple heterogeneous sensor networks deployed within the building. These data streams typically contain information on multiple physical quantities such as vibration, stress, temperature, and humidity, and have different sampling rates and formats. Subsequently, data fusion and structural reconstruction processing are performed to integrate and map these raw heterogeneous data into a unified, spatiotemporally consistent digital mirror of the building. This mirror is a precise digital mapping of the building's physical state in the information space. Based on this digital mirror, dynamic load response characteristics and static structural intrinsic characteristics are further separated and extracted, thereby separating a time-varying feature dataset reflecting the structure's time-varying characteristics and a steady-state feature dataset reflecting its inherent characteristics. Next, the extracted time-varying feature dataset is progressively pattern-matched with historical failure feature maps stored in a pre-built knowledge base to identify abnormal evolution patterns inherent in the structure's current response. Based on the identified abnormal evolution patterns and the steady-state feature dataset, a mathematical model is used to deduce the performance degradation trajectory of the building structure over several predetermined maintenance cycles in the future. Ultimately, based on the predicted performance degradation trajectory, a series of proactive intervention plans are automatically generated that are precisely correlated with specific future time points to guide preventative maintenance.

[0076] In one embodiment of the present invention, see [reference] Figure 2 First, a unified timestamp alignment and spatial registration benchmark is established for the raw heterogeneous data streams from different sensor networks. Using the precise geometric and spatial coordinate system provided by Building Information Modeling (BIM), the spatiotemporal information of all sensor data is mapped to this unified coordinate system. Based on the physical properties of the sensor data and their spatial topological relationships within the building structure, a multi-source data correlation matrix is ​​constructed. This matrix is ​​used to quantify the coupling and transmission relationships, such as between vibration data and stress data, or between temperature field and structural response. Based on this multi-source data correlation matrix, missing value imputation and outlier correction are performed on data with spatiotemporal correlations, generating a set of cleaned, temporally and spatially continuous field data, i.e., continuous spatiotemporal field data.

[0077] Using the obtained continuous spatiotemporal field data as known boundary and initial conditions, a pre-defined calculation model for building physics properties is driven. This model is typically based on the finite element method or other numerical methods, and identifies the distribution of structural parameters inside the building that cannot be directly measured by solving an inverse problem. Specifically, this inverse problem process involves constructing a physical equation inverse problem model with the desired structural parameter distribution as the solution variable and the continuous spatiotemporal field data as the known observations. In constructing this model, the digital mirror of the building is geometrically discretized into a finite number of elements, and a set of structural parameters to be identified is assigned to each element. The structural parameters of all elements constitute the solution variables. Simultaneously, control equations describing the building's mechanical or thermodynamic response are established, and the solution variables are used as spatial distribution coefficients in the control equations. The objective function is defined as the norm of the difference between the theoretical response field calculated by the control equations and the corresponding observations in the continuous spatiotemporal field data. A regularization constraint term reflecting the smoothness of the spatial distribution of structural parameters is added to the objective function. This constraint term is typically the spatial gradient of the solution variable or the norm of the Laplace operator. The physical equation inverse problem model is thus expressed as finding the structural parameter distribution that minimizes the objective function under the regularization constraint.

[0078] This model is solved using an iterative optimization algorithm, minimizing the difference between the predicted field data generated by the building physics property calculation model and the continuous spatiotemporal field data. In each iteration, the gradient of the objective function for the variable to be solved is calculated using the adjoint state method based on the difference between the prediction and the measurement, and the structural parameter distribution is updated along the gradient descent direction. The calculation terminates when the difference between the predicted field data and the measured continuous spatiotemporal field data is less than a preset tolerance or the number of iterations reaches the upper limit. Finally, the optimized structural parameter distribution is deeply integrated with the original building information model to form a three-dimensional computable model that can reflect the internal state of the structure in real time. This model is the required building digital mirror.

[0079] In practical implementation, the following describes the implementation method in detail with a specific bridge structural health monitoring scenario. The monitoring object is a long-span steel box girder bridge. Multiple sensor networks are deployed at several key sections of the bridge, including a strain gauge network for monitoring local strain, an accelerometer network for monitoring overall vibration, and a temperature and humidity sensor network for monitoring environmental influences.

[0080] In practical implementation, a raw heterogeneous data stream within a 24-hour monitoring time window is synchronously acquired from the aforementioned sensor network. This raw data stream includes strain data sampled per second, vibration acceleration data sampled per second, and temperature and humidity data sampled per minute. A unified timestamp alignment and spatial registration benchmark is established, using GPS time as the unified timestamp source to align the timestamps of all sensor data to the millisecond level. Simultaneously, based on a pre-established bridge building information model, the spatial coordinates of each sensor are mapped to a unified three-dimensional coordinate system within the building information model, ensuring consistent spatiotemporal reference for sensor data from different locations and of different types. A multi-source data association matrix is ​​constructed based on the physical properties and spatial topological relationships of the sensor data. This multi-source data association matrix is ​​a mathematical expression describing the coupling relationships between data. For example, there is a linear relationship between the bending moment of a specific section of the bridge main beam and the readings of multiple strain gauges arranged at that section; there is a correlation between the modal frequencies of the bridge and the spectral peaks of multiple accelerometers; and there is a thermal effect relationship between changes in ambient temperature and the strain readings of the steel beam. The multi-source data association matrix quantifies these relationships into weighting coefficients. Based on the multi-source data association matrix, missing value imputation and outlier correction are performed on data with spatiotemporal correlation. In specific implementation, when a strain gauge loses ten consecutive minutes of data due to a temporary malfunction, the missing value is filled by using the contemporaneous data of other strain gauges that are spatially adjacent and mechanically correlated with the strain gauge in the multi-source data association matrix and performing collaborative estimation through spatial interpolation and mechanical relationship. For abnormally fluctuating data points that deviate significantly from the normal relationship described by the multi-source data association matrix, weighted smoothing correction is performed using the associated data. Finally, cleaned strain field, vibration field and temperature field data that are continuous in time and space are generated, i.e., continuous spatiotemporal field data.

[0081] Using continuous spatiotemporal field data as known boundary and initial conditions, a pre-defined finite element model of a bridge structure is driven to inversely solve for the distribution of structural parameters inside the bridge that cannot be directly measured, such as the spatial distribution of the material's elastic modulus. An inverse physical equation model is constructed, with the structural parameter distribution as the variable to be solved and the continuous spatiotemporal field data as the known observations. The specific method for constructing this inverse physical equation model involves geometrically discretizing the digital mirror image of the building into a finite number of elements, assigning a set of structural parameters to be identified to each element, and the structural parameters of all elements collectively constituting the variable to be solved. Subsequently, a governing equation describing the mechanical response of the building is established, and the variable to be solved is used as the spatial distribution coefficient in the governing equation, thereby relating the physical response to the structural parameters. When defining the objective function, it is expressed as the norm of the difference between the theoretical response calculated by the governing equations under given boundary conditions and the observed values ​​of the corresponding physical quantities in continuous spatiotemporal field data. To suppress ill-posedness of the solution, a regularization constraint term is introduced into the objective function. This term is typically based on the gradient of the variable to be solved in space or the norm of the Laplace operator. In some embodiments, the digital image of the bridge is geometrically discretized into 100,000 finite element elements, each element is assigned an elastic modulus parameter to be identified, and the elastic modulus parameters of all elements constitute the variable to be solved. The governing equations describing the mechanical response of the bridge structure, i.e., the structural dynamics equations, are established, with the variable to be solved as material property coefficients in the governing equations. The objective function is defined as the sum of squares of the L2 norms of the difference between the theoretical response calculated by the governing equations under given measured load boundary conditions and the observed values ​​of the corresponding physical quantities in continuous spatiotemporal field data. A regularization constraint term is added to the objective function, which is the L2 norm of the gradient of the variable to be solved in space, to suppress ill-posedness of the solution and ensure the spatial smoothness of the solution.

[0082] The inverse problem model of the physical equations is expressed as finding the structural parameter distribution that minimizes the objective function under regularization constraints. An iterative optimization algorithm is used to solve this problem, minimizing the difference between the predicted field data generated by the building physics characteristic calculation model and the continuous spatiotemporal field data, while simultaneously satisfying the regularization constraints. In each iteration, based on the difference between the predicted response and the measured continuous spatiotemporal field data, the gradient of the objective function for the variable to be solved is calculated using the adjoint state method. The adjoint state method efficiently obtains gradient information by solving an adjoint equation. Subsequently, the structural parameter distribution is updated along the gradient descent direction. The calculation terminates when the difference between the predicted field data and the measured continuous spatiotemporal field data is less than a preset tolerance. It can be understood that the above inverse problem solution process can be summarized by a mathematical formula for its objective function form:

[0083]

[0084] in: Represents the objective function value. Let represent the vector of variables to be solved, consisting of the elastic modulus parameters of all elements. Represents the continuous spatiotemporal field data vector of the observation. This indicates that the calculation model is based on the current parameters using the building's physical properties. The calculated predicted response vector, Represents the regularization parameter. Indicates parameters Spatial gradient operator, The L2 norm of the vector is represented by the square. The final optimized distribution of elastic modulus parameters is deeply fused with the original bridge building information model. The fusion process involves assigning the solved parameter values ​​to the corresponding geometric components and regions in the building information model, thereby forming a three-dimensional computable model that can reflect the internal material state of the structure in real time. This three-dimensional computable model is the digital mirror image of the bridge under the current monitoring time window.

[0085] In one embodiment of the present invention, see [reference] Figure 3 A set of preset virtual excitation signals is applied to the constructed digital mirror image of the building. These signals typically simulate typical environmental loads such as wind loads or harmonic excitations. The theoretical response field of the digital mirror image under these virtual excitation signals is obtained by solving a built-in building physics characteristic calculation model. Simultaneously, actual response field data corresponding to the environmental loads during the actual monitoring period are extracted from the continuous spatiotemporal field data generated through data fusion processing. The residual field between the theoretical and actual response field data is calculated. This residual field characterizes the deviation between the response of the actual structure caused by local performance changes such as damage or stiffness degradation and the response of the ideal digital mirror image.

[0086] The calculated residual field is decomposed into multi-scale spatial components. Using spatial filtering or wavelet transform, macroscopic residual components related to the overall structural stiffness characteristics and microscopic residual components related to local damage, loose connections, and other conditions are separated. Time-domain features are extracted from the separated macroscopic and microscopic residual components, including statistical, frequency-domain, or time-frequency-domain features, thus forming a time-varying feature dataset describing the overall time-varying characteristics of the structure. On the other hand, parameters that remain relatively stable within the monitoring time window are extracted from the structural parameter distribution contained in the building's digital mirror image. These parameters include the elastic modulus density distribution of the material or the design geometric parameters of key connection nodes. These parameters constitute a steady-state feature dataset reflecting the inherent properties of the structure.

[0087] In practical implementation, the following description uses a specific health monitoring scenario of a large-scale spatial grid structure roof as an example. The monitoring object is the welded ball-node grid structure of a stadium. In the implementation, a set of preset virtual excitation signals are applied to the building's digital mirror image. The virtual excitation signals are in the form of a set of white noise-based excitations applied to all nodes of the finite element model of the building's digital mirror image, with a frequency range covering the first 50 natural frequencies of the structure. Frequency response analysis is performed using a built-in building physics characteristic calculation model to calculate the theoretical acceleration response spectral density of each node of the building's digital mirror image under this virtual excitation signal, thus forming the theoretical response field. From the continuous spatiotemporal field data generated after data fusion processing, the actual acceleration response time history data, corresponding to the environmental wind load and crowd activity load during the actual monitoring period, collected by 50 acceleration sensors deployed on the roof, are extracted. The actual acceleration response time history data is converted into actual acceleration response spectral density through spectrum analysis, forming the actual response field data. The residual field between the theoretical response field and the actual response field data is calculated. The residual field is obtained by calculating the difference between the theoretical spectral density and the actual spectral density at each frequency point and each sensor location. The residual field characterizes the deviation between the dynamic response of the actual space frame structure and the idealized digital mirror body response caused by local performance changes such as fatigue of welded nodes and changes in the initial stress of the members.

[0088] The calculated spectral density residual field is decomposed into multi-scale spatial components. In some embodiments, a two-dimensional wavelet transform is used to decompose the spatial distribution of the residual at each frequency. This decomposition process separates the residual field into approximation coefficients and detail coefficients. The approximation coefficients correspond to the low-frequency spatial components of the residual field, reflecting the macroscopic residual components related to the overall structural stiffness characteristics. The detail coefficients correspond to the high-frequency spatial components of the residual field, reflecting the microscopic residual components related to local node damage and bolt connection status. Time-domain features are extracted from the separated macroscopic and microscopic residual components. For the macroscopic residual components, their energy integral values ​​in different frequency bands are arranged in a time series, and their root mean square value, skewness, and kurtosis are calculated to form a macroscopic time-varying feature vector describing the time-varying characteristics of the overall structural stiffness. For the microscopic residual components, the time-varying sequence of their energy concentration coefficient in a specific high-frequency band and the evolution sequence of their spatial distribution entropy are extracted to form a microscopic time-varying feature vector describing the time-varying characteristics of local states. The macroscopic and microscopic time-varying feature vectors together constitute a time-varying feature dataset. On the other hand, parameters that remain relatively stable within the 24-hour monitoring window are extracted from the structural parameter distribution contained in the building's digital mirror image. These parameters include the average elastic modulus of each member material obtained through inversion, the cross-sectional geometric parameters of the members, and the initial design strength parameters of the joint welds. These parameters, which do not change during the monitoring period, constitute the steady-state characteristic dataset. The spatial decomposition process can be formally represented as:

[0089]

[0090] in: Indicates frequency and spatial coordinates Spectral density residual at that location This refers to the low-frequency approximation component after wavelet decomposition, which is the macroscopic scale residual component. Indicates the first High-frequency detail components at each scale are microscale residual components. This indicates the total number of levels in the decomposition.

[0091] In one embodiment of the present invention, the pattern matching process begins by retrieving historical monitoring cases from a pre-built knowledge base that are the same as or similar to the target building structure type. A complete feature evolution sequence from a normal state to a failure state is extracted from each case; these sequences constitute a historical failure feature map for comparison. The feature vectors in the real-time extracted time-varying feature dataset are arranged in chronological order to form the feature evolution sequence to be identified.

[0092] A dynamic time warping algorithm is used to calculate the matching degree between the current feature evolution sequence to be identified and each historical failure feature map in the knowledge base. First, three feature sub-sequences representing the early, middle, and late stages of failure development are extracted from each historical failure feature map according to a preset time division ratio as reference sequences. The feature evolution sequence to be identified is used as the matching sequence, and the dynamic time warping algorithm is applied to each reference sequence to calculate the similarity distance. This process constructs a cumulative cost matrix and finds the optimal warping path, calculating the minimum cumulative cost required to nonlinearly stretch or compress the matching sequence on the time axis to best match the reference sequence. This minimum cumulative cost is quantified as the similarity distance between the two. The three similarity distance values ​​are finally obtained by calculating the early, middle, and late feature sub-sequences of the historical failure feature map respectively. Based on the calculated similarity distance with the features of each stage, several target historical failure feature maps with the highest matching degree with the current feature evolution sequence are selected, and their subsequent evolution paths are extracted. Based on the difference between the evolution path of the target historical failure feature map and the current feature evolution sequence, interpolation and extrapolation are used to predict the abnormal evolution mode of the current structure in the future. This mode includes the damage type, location, and diffusion direction.

[0093] During the continuous operation of the system, the knowledge base needs to be incrementally learned and updated. In the subsequent monitoring of the target building structure, data on the actual structural state evolution and the feedback data on the effects of implemented proactive intervention schemes are recorded. When the actual structural state evolution deviates significantly from the previously predicted abnormal evolution pattern or performance degradation trajectory, this complete monitoring, prediction, intervention, and feedback sequence is treated as a new case. Feature extraction and standardization are performed on this new case to generate a new feature evolution sequence and corresponding actual evolution endpoint labels. The processed new case is stored in the knowledge base, forming an updated knowledge base together with the existing historical failure feature map. This updated knowledge base is used during the next pattern matching execution to achieve continuous evolution of the model's predictive capabilities.

[0094] In practical implementation, the following description is based on a specific health monitoring scenario of a reinforced concrete frame structure office building. In this scenario, a time-varying feature dataset describing the concentration coefficient of micro-strain energy in the beam-end shear wall node area has been extracted from the building's digital mirror over the past three months, on a weekly basis. The goal is to identify potential node shear crack evolution patterns.

[0095] In specific implementation, historical monitoring cases of the same type as the target reinforced concrete frame structure are retrieved from a pre-built knowledge base. The knowledge base stores complete monitoring records of multiple similar office building structures from a healthy state to partial failure due to node damage. A complete feature evolution sequence, characterized by the same microscopic strain energy concentration coefficient, is extracted from each case, representing the progression from a normal state to a failure state. These sequences constitute historical failure feature maps for comparison. The real-time extracted feature vectors, each twelve weeks long, are arranged chronologically to form the feature evolution sequence to be identified. A dynamic time warping algorithm is used to calculate the matching degree between the feature evolution sequence to be identified and each historical failure feature map in the knowledge base. In some embodiments, from each historical failure feature map in the knowledge base, three feature sub-sequences representing the early, middle, and late stages of failure development are extracted as reference sequences according to a preset time division ratio. The twelve-week feature evolution sequence to be identified is used as the matching sequence, and the dynamic time warping algorithm is executed with the early, middle, and late feature sub-sequences of each historical failure feature map. For each reference sequence and the sequence to be matched, a dynamic time warping algorithm is executed. By constructing a cumulative cost matrix and finding the optimal warping path, the minimum cumulative cost required to nonlinearly stretch or compress the sequence to be matched on the time axis to best match the reference sequence is calculated. The minimum cumulative cost is quantified as the similarity distance between the two sequences. Specifically, the similarity distance is calculated separately for the sequence to be matched and the early, middle, and late feature subsequences of the historical failure feature map, resulting in three similarity distance values ​​representing the degree of matching with the early, middle, and late stages, respectively.

[0096] It can be understood that the process of the Dynamic Time Warping algorithm finding the optimal warped path is to find a point-to-point mapping that minimizes the total alignment cost. The total path cost calculation can be expressed as:

[0097]

[0098] in: This represents the minimum cumulative cost, i.e., the similarity distance, obtained through final calculation. This represents a regular path from the start to the end of a sequence. Indicates a pair of points on the path. Indicates the sequence to be matched In the Feature values ​​at each time point Represents the reference sequence In the Feature values ​​at each time point This represents a function for calculating the local distance between two points, such as Euclidean distance. Based on the calculated similarity distance values ​​with the features of each stage, several target historical failure feature maps with the highest matching degree in the early or middle stages of the current feature evolution sequence to be identified are selected, and the evolution paths of these target historical failure feature maps to the later stages are extracted. Based on the difference between the evolution paths of the target historical failure feature maps and the current feature evolution sequence to be identified, the abnormal evolution mode of the current reinforced concrete node is predicted through interpolation and extrapolation calculations. The abnormal evolution mode specifically includes predicting that cracks will mainly develop in a shear manner, predicting that the potential crack location is located on the southeast side of the node core area, and predicting that the crack propagation direction is along the diagonal direction.

[0099] In practical implementation, the knowledge base needs to be incrementally learned and updated for continuous optimization. During the subsequent continuous monitoring of the target reinforced concrete frame structure, the actual nodal strain state evolution data and the effect feedback data of the subsequent active intervention schemes are recorded. When the actual structural state evolution deviates significantly from the previously predicted abnormal evolution pattern or performance degradation trajectory—for example, the actual crack propagation rate is much slower than predicted—this complete monitoring, prediction, intervention, and feedback sequence is treated as a new case. Feature extraction and standardization are performed on the new case to generate a new feature evolution sequence consistent with the knowledge base format. The new feature evolution sequence is then labeled with its actual evolution endpoint, such as "stable without development" or "recovered after repair." The processed new case is stored in the knowledge base, forming an updated knowledge base together with the original historical failure feature map.

[0100] In one embodiment of the present invention, a performance degradation prediction model is first established, using the component material and connection parameters described by a steady-state feature dataset as the initial state, and the damage propagation law described by the identified abnormal evolution pattern as the driving rule. This model is typically constructed based on physical degradation equations or empirical degradation rate functions. In the performance degradation prediction model, the entire prediction time axis is divided into multiple consecutive small time steps. Within each small time step, the parameters of the corresponding structural parts, such as the local stiffness reduction factor or crack propagation length, are updated according to the damage propagation law and rate described by the abnormal evolution pattern.

[0101] To simulate uncertainties in real-world usage environments, random load disturbances conforming to a probability distribution are introduced into the performance degradation prediction model, incorporating the statistical distribution of historical environmental loads. This performance degradation prediction model is run iteratively until the overall performance index of the structure reaches a preset performance failure threshold. The complete curve of the overall structural performance index changing over time is recorded throughout the simulation. Predicted performance index values ​​corresponding to different preset maintenance cycle times are extracted from this complete performance change curve, and the times when the performance index first falls below each warning threshold are marked. These key time points, along with the corresponding predicted performance index values, collectively define the performance degradation trajectory of the building structure over the future timeframe.

[0102] In specific implementation, the following description is based on a specific scenario of fatigue damage development in the orthotropic steel bridge deck of a steel box girder bridge. In this scenario, the steady-state feature dataset includes the elastic modulus, yield strength, initial fatigue crack size, and fracture toughness parameters of the bridge deck steel. The identified abnormal evolution pattern is the propagation of fatigue cracks at the weld details between the U-rib and the top plate. The pattern describes the crack type as type I dominant fatigue cracks, located at the end of the U-rib weld in the middle of the third span, and the propagation direction is along the weld fusion line and extending towards the top plate matrix.

[0103] In practical implementation, a performance degradation prediction model is established, using a steady-state feature dataset as the initial state and anomaly evolution patterns as the driving rules. The performance degradation prediction model is built upon fracture mechanics theory, with the Paris formula describing fatigue crack propagation as the core driving rule. In practice, the initial state of the performance degradation prediction model consists of the initial crack length, material constants, and stress state at key details of the bridge deck. The driving rule is a physical equation describing the relationship between crack propagation rate and the stress intensity factor range. In the performance degradation prediction model, the entire prediction time axis is divided into multiple consecutive micro-time steps, each representing a fixed number of load cycles the bridge experiences, for example, 10,000 vehicle load cycles. Within each micro-time step, based on the fatigue crack propagation pattern described by the anomaly evolution pattern, the parameters of the corresponding structural parts are updated. Specifically, based on the stress intensity factor range calculated in the current micro-time step, the crack propagation amount within this micro-time step is calculated using the crack propagation formula, and the crack length parameter is updated. Combining the statistical distribution of environmental loads, random load disturbances conforming to a probability distribution are introduced into the performance degradation prediction model to simulate uncertainties in actual use.

[0104] In some embodiments, random load disturbances are implemented using the Monte Carlo method. At each tiny time step, a load amplitude is randomly selected based on the vehicle load spectrum obtained from long-term monitoring and applied to the model. The probability distribution of the load amplitude conforms to the Weibull distribution obtained from monitoring statistics. The performance degradation prediction model is run and iteratively calculated until the crack length reaches the preset performance failure threshold, i.e., penetrating the plate thickness or causing critical instability. The complete curve of the overall structural performance index as a function of time, expressed as the fatigue life consumption index of structural details or the remaining bearing capacity coefficient, is recorded throughout the simulation process. From the complete curve, the predicted values ​​of the performance index corresponding to different future maintenance cycle time points are extracted, such as the predicted values ​​at the annual inspection time, and the time point when the performance index first falls below the warning threshold of each level is marked. The performance degradation trajectory is defined by these key time points and the corresponding predicted values ​​of the performance index. It can be understood that the crack propagation driving rule is the core of the performance degradation prediction model, and its expression is:

[0105]

[0106] in: This represents the increment of fatigue crack propagation within a tiny time step. and These are material constants provided by the steady-state characteristic dataset. This represents the range of stress intensity factors determined by the current crack size, structural geometry, and random load disturbances. This represents the number of load cycles represented by a tiny time step. See Table 1.

[0107] Table 1: Key Points for Predicting the Performance Degradation Trajectory of Bridge Key Sections

[0108] Maintenance cycle Forecast timeline (operational year) Predicted performance index (remaining bearing capacity coefficient) Warning Level Next cycle 8.5 years 0.85 Yellow alert threshold The next two cycles Year 12.1 0.70 Orange alert threshold The next three cycles Year 15.3 0.55 Red alert threshold

[0109] In practical implementation, due to the introduction of random load disturbances in the performance degradation prediction model, the trajectory obtained from a single simulation has randomness. Optionally, to assess the uncertainty of performance degradation, the performance degradation prediction model can be run hundreds or thousands of times to obtain the probability distribution of time points when performance indicators fall below each warning threshold, such as the average time and confidence interval for fatigue life to reach the orange warning threshold. In some embodiments, the performance degradation prediction model can consider the coupling of multiple damage modes. For example, when fatigue cracks extend to a certain length, they induce accelerated corrosion of the steel structure. At this time, the driving rule of the performance degradation prediction model will switch or superimpose a corrosion rate model. The process of extracting predicted values ​​from the complete performance change curve is automated. The system performs interpolation calculations on the simulated time-performance curve based on preset future maintenance plan time points to obtain accurate predicted values ​​of performance indicators. Marking the time points when performance indicators first fall below each level of warning threshold is done by comparing performance indicator values ​​with preset thresholds in real time during the simulation. When the performance indicator curve crosses a certain threshold line, the time point is recorded.

[0110] See Figure 4 In the prediction of fatigue damage performance degradation trajectory of steel box girder bridge decks, this figure uses operational time as the horizontal axis and the remaining bearing capacity coefficient as the vertical axis to present the evolution law of structural performance. The blue curve in the figure represents the average remaining bearing capacity: initially (approximately 0-8 years), performance remains stable (coefficient ≈ 1.0), then enters a rapid degradation stage, approaching 0 by 20 years. The light blue area represents the 90% confidence interval, reflecting the uncertainty range of performance degradation under random load disturbances. The figure marks key nodes using three sets of warning systems: yellow, orange, and red. The yellow warning key point (7.5 years of operation, coefficient 0.85) corresponds to the yellow warning threshold; the orange warning key point (12.5 years of operation, coefficient 0.70) corresponds to the orange warning threshold; and the red warning key point (15 years of operation, coefficient 0.55) corresponds to the red warning threshold. The intersection of each key point with its corresponding warning time point (yellow dashed line, orange dashed line, red dashed line) clarifies the time node when performance first falls below the threshold of each level. The core value of the graph lies in combining the output of the performance degradation prediction model with a multi-level early warning mechanism, which intuitively presents the performance trajectory from the stable stage to the failure stage, and provides a quantitative basis for the time window planning of proactive intervention schemes.

[0111] In one embodiment of the invention, the performance degradation trajectory is analyzed to determine specific future time points when the predicted values ​​of the overall or local performance indicators of the structure fall below the warning thresholds for each level. For each determined future time point, a reverse analysis is performed to identify the set of key components and their specific parameter degradation amounts that lead to performance decline at that time point in the performance degradation prediction model. Based on the identified set of key components and their parameter degradation amounts, executable technical measures are matched from a predefined intervention measure library, which includes various predefined maintenance methods such as applying prestress, adding temporary supports, replacing local components, and applying protective coatings. The implementation effect of each matched intervention measure in the performance degradation prediction model is evaluated, i.e., the degree of recovery of structural performance indicators and the performance maintenance time are simulated after the implementation of the measure. Combining engineering cost budget and building use function constraints, a multi-objective optimization analysis is performed on candidate intervention measures that can effectively maintain or improve performance at each future time point. For each finally selected optimal intervention measure, a precise implementation time window, specific operating parameters, and expected performance maintenance period are specified, thereby forming a series of proactive intervention schemes triggered by the predicted time points.

[0112] In practical implementation, the following description continues based on the previously simulated fatigue damage development scenario of orthotropic steel bridge decks in steel box girder bridges. In this scenario, a performance degradation trajectory has been obtained, defining key warning time points and predicted performance index values. In practical implementation, the performance degradation trajectory is analyzed to determine the specific future time points when the predicted performance index values ​​fall below the warning thresholds for each level. Based on the provided table data, the system determines that the performance index will fall below the yellow warning threshold in year 8.5, below the orange warning threshold in year 12.1, and below the red warning threshold in year 15.3. For each specific future time point, a reverse analysis is performed on the set of key components and their parameter degradation amounts that lead to performance degradation at that time point in the performance degradation prediction model. For the year 8.5 time point, the reverse analysis determines that the key component leading to performance degradation is the welding detail between the U-rib (numbered U3-R5) at the mid-span of the third span and the top plate, with a predicted fatigue crack length of 15 mm and remaining fatigue life reduced to 60% of the design life.

[0113] Based on the set of critical components and their parameter degradation, executable interventions are matched from a predefined intervention library. The intervention library is a database storing standardized maintenance operations and their applicable conditions. For the degradation state of "15 mm fatigue crack length in the U-rib-top plate weld detail," the executable interventions matched from the library include "ultrasonic impact treatment to improve weld toe geometry and introduce residual compressive stress," "drilling crack arrest holes to mediate the fracture propagation path and unload at the crack tip," "local grinding to eliminate the crack and re-welding," and "reinforcing the crack area with carbon fiber composite plates." For each matched intervention, its implementation effect in the performance degradation prediction model is evaluated. The evaluation is conducted by adjusting the parameters of the critical components in the performance degradation prediction model according to the expected effect of the intervention, and then rerunning the model to simulate the degree of recovery and maintenance time of the structural performance indicators after implementation. In some embodiments, for the measure of "drilling an arresting hole," the evaluation model replaces the crack tip with a circular hole to eliminate stress singularity, and updates the stress intensity factor calculation formula for this detail in the model. After recalculation, the remaining bearing capacity coefficient after implementing this measure can be restored to 0.92 and can remain above the yellow warning threshold for approximately four years. It can be understood that the effectiveness evaluation of the intervention measure can be formally expressed as:

[0114]

[0115] in: This represents the predicted value of the structural performance index after the implementation of intervention measures. This represents the predicted value of the performance index before the implementation of intervention measures. This represents a vector of changes in key component parameters resulting from intervention measures. This indicates a performance degradation prediction model that has been modified to reflect the physical effects of the intervention. This function represents a re-evaluation of performance based on the revised model. In practice, when assessing the effectiveness of intervention measures, the system first makes targeted corrections to the key component parameters in the performance degradation prediction model according to the matched intervention measures. The revised model inherits the steady-state characteristic dataset and anomalous evolution mode driving rules of the original model, but the initial state of the key parts has been updated to the parameter values ​​after the intervention. Subsequently, the function restarts the iterative calculation process of the performance degradation prediction model, starting from the current time point, and simulates the evolution trajectory of structural performance indicators over future time periods based on the same small time step division and random load disturbance introduction rules.

[0116] Combining engineering costs and building usage constraints, multi-objective optimization is performed on interventions that can effectively maintain performance at various future time points. The objective function typically includes the total cost of the intervention, the impact of implementation time on building operations, and the length of the performance maintenance period after implementation. For the 8.5-year time point, the optional intervention "local grinding and re-welding" is costly and requires closing a portion of the bridge deck for 3 days, but offers good performance recovery and a long maintenance period; "attaching carbon fiber panels" has moderate costs and does not require traffic interruption, but its long-term durability needs evaluation. A multi-objective optimization algorithm is used to weigh these trade-offs, specifying a precise implementation time window, specific operational parameters, and expected performance maintenance period for each optimal intervention. In some embodiments, the optimization result is: scheduling "ultrasonic impact treatment" at the end of the eighth year, with an implementation time window of 5 consecutive nights during low traffic hours, specific operational parameters including an impact intensity of 0.25 mmA amplitude and an impact coverage area of ​​20 mm wide regions on both sides of the weld, which is expected to postpone the yellow warning time point to the 11th year.

[0117] See Figure 5 A spatiotemporal evolution matrix of performance risk for multiple components was constructed, using structural component type as the vertical dimension and time nodes as the horizontal dimension. Specifically, the vertical dimension of the matrix covers core components of steel box girder bridges such as U3-R5 welds, U2-L4 welds, and end webs, while the horizontal dimension marks key time nodes from the present to 20 years. The numerical values ​​within the matrix, combined with the risk scores (1-10, low to high) on the right and the corresponding color gradients, quantify the performance degradation risk level of each component at different time nodes. In actual analysis, the matrix intuitively presents the spatiotemporal distribution characteristics of risk through the coupled mapping of color and numerical values: for example, the U3-R5 weld has a current risk score of 2 (low risk), which rises to 10 (extremely high risk) as time progresses to 20 years; while the risk score of the diaphragm bolts remains in the range of 1-5 throughout the entire cycle, reflecting the slow performance degradation. Meanwhile, the differences in risk evolution rates among different components can be reflected through numerical comparisons at the same time points. For example, at 12.1 years, the risk score of the U3-R5 weld is 8, while that of the bridge deck pavement is only 2, reflecting the differences in the performance degradation mechanisms of welding details and pavement structures. At the parameter level, the time points in the figure (8.5 years, 12.1 years, etc.) correspond to the warning threshold time points in the performance degradation trajectory, and the values ​​within the matrix represent the risk assessment scores of each component at the corresponding time points. The calculation is based on the iterative results of the performance degradation prediction model, integrating the coupled analysis of time-varying feature datasets and steady-state feature datasets.

[0118] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0119] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for intelligent building monitoring and management based on multidimensional data and data analysis, characterized in that, The following processing steps are included: Simultaneously acquire raw heterogeneous data streams within a monitoring time window from multiple designated sensor networks of the building; Data fusion and structure restoration processing are performed on the original heterogeneous data stream to construct a digital mirror of the building with complete spatiotemporal consistency; Based on the digital mirror image of the building, the dynamic load response characteristics and static structural intrinsic characteristics are separated and extracted to separate the time-varying feature dataset and the steady-state feature dataset, including the following steps: A set of preset virtual excitation signals are applied to the digital image of the building, and the theoretical response field of the digital image of the building under the virtual excitation signals is calculated. Extract the actual response field data corresponding to the actual environmental load from the continuous spatiotemporal field data; Calculate the residual field between the theoretical response field and the actual response field data, wherein the residual field characterizes the response deviation caused by changes in the local performance of the structure; The residual field is decomposed into a multi-scale spatial decomposition to separate the macro-scale residual components related to the overall structural stiffness and the micro-scale residual components related to local damage and connection state. The macro-scale residual components and the micro-scale residual components are subjected to time-domain feature extraction to form the time-varying feature dataset describing the overall time-varying characteristics of the structure. From the structural parameter distribution of the building digital mirror, material property parameters and geometric connection parameters that remain relatively stable within the monitoring time window are extracted to form the steady-state feature dataset; The time-varying feature dataset is progressively pattern-matched with a historical failure feature map in a knowledge base to identify abnormal evolution patterns of structural response, including the following steps: Historical monitoring cases with the same or similar structural type as the target building are retrieved from the knowledge base, and the complete feature evolution sequence from normal state to failure state in each case is extracted to form the historical failure feature map. The feature vectors in the time-varying feature dataset are arranged in chronological order to form a feature evolution sequence to be identified; The dynamic time warping algorithm is used to calculate the similarity distance between the feature evolution sequence to be identified and the early, middle and late stage features of each historical failure feature map in the knowledge base; Based on the similarity distance, select several target historical failure feature maps that have the highest matching degree with the feature evolution sequence to be identified, and extract their subsequent evolution paths; Based on the difference between the evolution path of the target historical failure feature map and the current feature evolution sequence to be identified, the abnormal evolution mode that may occur in the current structure in the future is predicted by interpolation and extrapolation calculation. The abnormal evolution mode includes damage type, location and diffusion direction. Based on the anomalous evolution pattern and the steady-state feature dataset, the performance degradation trajectory of the building structure over multiple future maintenance cycles is deduced, including the following steps: Establish a performance degradation prediction model with the steady-state feature dataset as the initial state and the abnormal evolution mode as the driving rule; In the performance degradation prediction model, the time axis is divided into multiple consecutive small time steps. Within each small time step, the parameters of the corresponding structural parts are updated according to the damage propagation law described by the abnormal evolution mode. By combining the statistical distribution of environmental loads, random load disturbances conforming to probability distributions are introduced into the performance degradation prediction model to simulate the uncertainties in actual use. Run the performance degradation prediction model and iteratively calculate until the preset performance failure threshold is reached. Record the complete curve of the overall structural performance index changing over time during the entire simulation process. From the complete curve, the predicted performance index values ​​corresponding to different future maintenance cycle time points are extracted, and the time points when the performance index first falls below the warning threshold of each level are marked. The performance degradation trajectory is defined by these time points and the corresponding predicted performance index values. Based on the performance degradation trajectory, an active intervention plan is generated that precisely matches future time points.

2. The intelligent building monitoring and management method based on multidimensional data and data analysis according to claim 1, characterized in that, The process of performing data fusion and structure restoration on the original heterogeneous data stream to construct a digital mirror of the building with complete spatiotemporal consistency includes the following steps: For sensor data from different sources in the original heterogeneous data stream, a unified timestamp alignment and spatial registration benchmark is established to map all data to the same spatiotemporal coordinate system of the building information model. Based on the physical properties and spatial topological relationships of the sensor data, a multi-source data correlation matrix is ​​constructed. This multi-source data correlation matrix is ​​used to describe the coupling and transmission relationships between vibration, stress, temperature, and humidity data. Based on the multi-source data association matrix, missing value imputation and outlier correction are performed on the data with spatiotemporal correlation to generate cleaned continuous spatiotemporal field data. Using the continuous spatiotemporal field data as boundary and initial conditions, a preset building physical property calculation model is driven to solve the distribution of structural parameters inside the building that cannot be directly measured in reverse. The structural parameter distribution obtained by the solution is deeply integrated with the building information model to form a three-dimensional computable model that can reflect the internal state of the structure in real time. The three-dimensional computable model is the digital mirror of the building.

3. The intelligent building monitoring and management method based on multidimensional data and data analysis according to claim 2, characterized in that, The step of generating an active intervention plan that precisely matches future time points based on the performance degradation trajectory includes the following steps: Analyze the performance degradation trajectory to determine the specific future time points when the predicted value of the performance index falls below the warning threshold for each level; For each specific future time point, a reverse analysis is performed on the set of key components and their parameter degradation amounts that lead to performance degradation in the performance degradation prediction model. Based on the set of key components and their parameter degradation, executable intervention measures are matched from the measure library. These intervention measures include, but are not limited to, applying prestress, adding temporary supports, replacing local components, and applying protective coatings. For each set of matched intervention measures, evaluate its implementation effect in the performance degradation prediction model, that is, the degree of recovery and maintenance time of structural performance indicators after implementation; By combining engineering costs and building usage constraints, multi-objective optimization is performed on intervention measures that can effectively maintain performance at various future time points. For each optimal intervention measure, a precise implementation time window, specific operating parameters, and expected performance maintenance period are specified, forming a series of time-triggered proactive intervention schemes.

4. The intelligent building monitoring and management method based on multidimensional data and data analysis according to claim 3, characterized in that, The process involves driving a pre-defined building physics property calculation model to inversely solve for the distribution of structural parameters inside the building that cannot be directly measured, including the following steps: Construct an inverse physical equation model with the structural parameter distribution as the variable to be solved and the continuous spatiotemporal field data as the known observations; In the inverse problem model of the physical equations, a regularization constraint term reflecting the smoothness of the spatial distribution of structural parameters is introduced; An iterative optimization algorithm is used to minimize the difference between the predicted field data generated by the building physical property calculation model and the continuous spatiotemporal field data, while satisfying the regularization constraint. In each iteration, based on the difference between the prediction and the actual measurement, the gradient of the objective function with respect to the variable to be solved is calculated using the adjoint state method, and the distribution of the structural parameters is updated along the gradient descent direction; When the difference between the predicted field data and the measured continuous spatiotemporal field data is less than the preset tolerance, or when the number of iterations reaches the upper limit, the calculation is terminated, and the final optimized parameter distribution is used as the structural parameter distribution.

5. The intelligent building monitoring and management method based on multidimensional data and data analysis according to claim 4, characterized in that, The construction of the inverse physical equation model, which uses the structural parameter distribution as the variable to be solved and the continuous spatiotemporal field data as the known observations, includes the following steps: The building digital mirror is geometrically discretized into a finite number of elements, each element is assigned a set of structural parameters to be identified, and the structural parameters of all elements constitute the variables to be solved; Establish control equations describing the mechanical response of the building, and use the variables to be solved as spatial distribution coefficients in the control equations; Define an objective function, which is the norm of the difference between the theoretical response calculated by the governing equations under given boundary conditions and the observed values ​​of the corresponding physical quantities in the continuous spatiotemporal field data; The regularization constraint term is added to the objective function. The regularization constraint term is the gradient of the variable to be solved in space or the norm of the Laplacian operator, which is used to suppress the ill-posedness of the solution. The inverse problem model of the physical equation is expressed as: under the regularization constraint, finding the distribution of structural parameters that minimizes the objective function.

6. The intelligent building monitoring and management method based on multidimensional data and data analysis according to claim 5, characterized in that, The method further includes incremental learning and updating of the knowledge base, specifically including: In the subsequent monitoring of the target building structure, the actual structural state evolution data and the effect feedback data of the implemented active intervention scheme are recorded; When the actual structural state evolution deviates significantly from the predicted abnormal evolution pattern or the performance degradation trajectory, this complete monitoring, prediction, intervention, and feedback sequence will be treated as a new case. The new case is subjected to feature extraction and standardization to generate a new feature evolution sequence and the corresponding actual evolution endpoint label; The processed new cases are stored in the knowledge base, and together with the original historical failure feature map, they form an updated knowledge base. The updated knowledge base is used in the next pattern matching execution to achieve continuous evolution of the model's predictive capabilities.

7. The intelligent building monitoring and management method based on multidimensional data and data analysis according to claim 6, characterized in that, The method employs a dynamic time warping algorithm to calculate the similarity distance between the feature evolution sequence to be identified and the early, middle, and late stage features of each historical failure feature map in the knowledge base, including the following steps: From each historical failure feature map in the knowledge base, three feature subsequences representing the early, middle, and late stages of failure development are extracted according to a preset time division ratio. The feature evolution sequence to be identified is used as the matching sequence, and the early, middle and late feature subsequences of the historical failure feature map are used as three reference sequences respectively. For the sequence to be matched and each of the reference sequences, a dynamic time warping algorithm is executed. By constructing a cumulative cost matrix and finding the optimal warping path, the minimum cumulative cost required to nonlinearly stretch or compress the sequence to be matched on the time axis to best match the reference sequence is calculated. The minimum cumulative cost is quantified as the similarity distance between the two, wherein the sequence to be matched and the early, middle and late feature subsequences of the historical failure feature map are calculated respectively, and finally three similarity distance values ​​representing the degree of matching with the early, middle and late stages are obtained respectively.

Images