Green construction monitoring method and system based on digital twinning

Abstract

This invention discloses a green construction monitoring method and system based on digital twins, belonging to the field of big data analysis technology. This invention achieves zero-sample, phased plug-and-play access of sensors without reconstructing the model topology by pre-setting standardized sensor extension mounting interfaces at the bottom layer of the BIM grid partition, solving the problem of cumbersome expansion and modification of closed models. It designs a two-layer split sensor feature and constructs steady-state convergence intervals according to different working conditions. It uses a bidirectional feature residual matrix to remove cold-start drift noise and combines multi-dimensional benchmark solutions to fuse weights, overcoming the deficiency of a single reference in distinguishing between equipment noise and actual construction anomalies. By isolating distorted data frame by frame, only valid samples are used to update the global feature benchmark, and time-series archived data continuously expands the samples and performs closed-loop monitoring, avoiding the dilution of existing sensor weights and missed anomaly detection. This breaks through the limitations of static benchmarks and takes into account both the inherent working condition constraints of buildings and the needs of full-cycle dynamic construction monitoring.

Description

Technical Field

[0001] This invention relates to the field of big data analytics, specifically to a green construction monitoring method and system based on digital twins. Background Technology

[0002] In a green construction monitoring system based on digital twins, as construction processes progress in stages, incremental addition of sensor nodes via model-reserved expansion interfaces is required to monitor local pollution and energy consumption. These new nodes are zero-sample cold-start devices lacking historical time-series samples. Existing solutions typically only use a single reference from the project's overall baseline features or current local neighborhood features to complete data verification and weight assignment. Due to the limitations of the reference dimension, it is difficult to accurately distinguish between the power-on drift noise of the new nodes and the actual discrete micro-pollution and instantaneous energy consumption anomalies on site. Cold-start distorted data is easily mixed into the overall feature fusion process, leading to problems such as the dilution of simulation weights of existing sensors and missed detection of local green anomalies. It cannot simultaneously meet the actual usage requirements of both the inherent constraints of the building's underlying working conditions and the real-time monitoring of dynamic construction. Therefore, a green construction monitoring method and system based on digital twins is urgently needed. Summary of the Invention

[0003] The purpose of this invention is to provide a green construction monitoring method and system based on digital twins to solve the problems raised in the prior art.

[0004] To achieve the above objectives, the present invention provides the following technical solution: a green construction monitoring method based on digital twins, the green construction monitoring method comprising the following steps:

[0005] Step S1: Construct a scalable digital twin green construction monitoring base adapted to the entire construction sequence evolution of building projects; based on the analysis of engineering construction drawings and supporting construction technical documents, obtain the component geometric parameters, spatial topological relationships, site zoning boundaries, and full-cycle WBS construction sequence decomposition rules of the building under construction, and complete the construction of a standardized BIM spatial skeleton model. The BIM stands for Building Information Model, which is a three-dimensional digital carrier that integrates the geometric parameters of building components, spatial topological relationships, site zoning boundaries, and the full-cycle construction sequence. The engineering construction drawings include architectural construction drawings, structural construction drawings, and site layout drawings. The supporting construction technical documents are the corresponding specialized construction organization design drawings. The construction of the standardized BIM spatial skeleton model is represented by establishing a three-dimensional digital base framework that maps one-to-one with the physical construction site and can support layered expansion interface embedding logic, based on the static entity information such as site boundaries, component topology, and zoning range output from the analysis of engineering drawings. Based on the hierarchical construction sequence logic obtained from the drawing analysis, the entire construction space is divided into independent BIM grid partitions that correspond one-to-one with each construction stage of foundation, main structure, electromechanical installation, and decoration. In response to the incremental environmental monitoring, energy consumption monitoring, and dust and noise monitoring expansion needs generated by each grid partition at different construction stages, standardized sensor extension mounting interfaces are natively preset at the bottom layer of the model of each BIM grid partition. The sensor extension mounting interface is specifically an open extension channel natively integrated into the BIM spatial skeleton model architecture layer. It pre-encapsulates communication protocol adaptation, automatic spatial component mapping, incremental data isolation caching, and neighborhood feature parameter migration subroutines. Each sensor extension mounting interface is uniquely bound to the corresponding BIM grid partition and independently configured with a data diversion link. In existing digital twin green construction monitoring technologies, the monitoring models are mostly fixed and closed architectures, without reserving standardized expansion access channels for the phased expansion monitoring needs of construction, and cannot adapt to the plug-and-play access of dynamically added zero-sample sensor nodes during construction; the models lack partitioned time-series adaptation architecture and built-in data processing subroutines, and the model topology needs to be reconstructed when adding new sensor nodes. By enabling plug-and-play access to various newly added sensor nodes during subsequent construction processes without altering or reconstructing the BIM spatial skeleton model topology, and through pre-defined standardized sensor extension interfaces for each construction stage, this approach provides a fixed model architecture to support data isolation, feature alignment, and weight initialization for incremental zero-sample sensor nodes during the cold start phase. This constructs a scalable digital twin green construction monitoring foundation with construction sequence adaptability and monitoring capabilities. Furthermore, it analyzes construction drawings to build a standardized BIM spatial skeleton model that is fully time-adaptive, divides independent BIM grid partitions according to construction processes, and natively pre-defines standardized sensor extension interfaces at the model's bottom layer. This allows for plug-and-play time-series expansion and access of newly added sensor nodes at different construction stages without requiring modifications, all without reconstructing the model topology. Step S2: Combining engineering construction drawings, supporting construction technical documents and real-time construction progress on site, complete the timing matching and access of incremental sensor nodes and the standardization preprocessing of raw monitoring data. The incremental sensor nodes are monitoring devices added and accessed through the BIM grid partition bottom layer preset standardized sensor extension mounting interface at each construction stage. Based on the engineering construction drawings and supporting construction technical documents, the interface configuration parameters, standard construction sequence nodes, and zone monitoring boundary range of the BIM grid partition are defined as the benchmark constraints for incremental sensor access and data processing. Real-time collection of dynamic construction process progress information at the construction site; timing verification and matching of the real-time construction progress at the site with the standard construction sequence nodes marked in the engineering construction drawings and supporting construction technical documents; determining the BIM grid partition construction stage in which the on-site construction process is located. When the on-site construction progresses to the target construction stage corresponding to any standardized sensor extension mounting interface, the preset sensor extension mounting interface in the corresponding BIM grid partition is activated and called; through the activated sensor extension mounting interface, the device identification, communication protocol adaptation and BIM space component binding mapping of the newly added incremental sensor nodes in the current construction stage are completed. According to the unified interface input parameter specification preset in the engineering construction drawings and supporting construction technical documents, the unified interface input parameter specification is a standardized access constraint defined in advance according to the data format, sampling sequence, parameter range and dimension defined in the engineering construction drawings and supporting construction technical documents. In existing monitoring technologies, there is no standardized timing access mechanism or unified data constraint specification for the sensor nodes added in stages during the construction process. Most of the new devices are temporarily connected and cannot be accurately bound to BIM grid partitions and construction sequence. The device protocol adaptation is chaotic and the monitoring boundaries are unclear. At the same time, the raw data output by the new cold start sensor devices is messy in format, inconsistent in range, and disordered in timing. There is a lack of unified preprocessing methods, which makes the raw data of the new nodes unable to be directly used for feature comparison and deviation correction. By using a unified interface input parameter specification, the raw data collected by the newly added sensor nodes are sequentially processed through parameter reduction and correction, sampling timing alignment, data format unification, and structured encapsulation. The raw data includes environmental, energy consumption, and dust and noise monitoring parameters. The standardized preprocessing of all incremental sensor raw data is completed to generate a structured incremental monitoring dataset with unified format, regular timing, and unified dimensions. Step S3: Based on the global multi-interface feature linkage comparison mechanism, complete the adaptive feature correction and dynamic weight solution of incremental sensing data; perform two-layer differential feature analysis and quantify the time-stationary features and local discrete mutation features of the structured incremental monitoring dataset after preprocessing with unified interface input parameter specifications; The mean, time series standard deviation, periodic fluctuation amplitude, and long-term benchmark offset of the monitoring indicators are calculated window by window using a sliding time series window as time series stationary features; the instantaneous peak jump amplitude, short-term mutation cumulative frequency, cross-regional sensor data difference, and sampling breakpoint interval are solved by point-by-point difference operation as local discrete mutation features. Based on the deployment data of incremental sensor nodes, a steady-state convergence interval is adaptively constructed under three operating conditions: In the first working case, when the BIM grid partition to which the incremental sensing node belongs has the same type of existing sensing interface deployed, the steady-state convergence interval is generated by self-learning based on the historical time-series characteristics statistical distribution of the existing sensing interface of the partition. In the second scenario, when the BIM grid partition to which the incremental sensor node belongs has no similar existing sensor interfaces, but there are similar sensor interfaces in the entire building area, the global feature distribution clustering algorithm is used to screen and match the current construction process and the site environment to fit and construct a steady-state convergence interval. In the third working condition, when there are no similar sensor interfaces in the entire building area and the incremental sensor nodes are brand new zero-sample monitoring devices, the steady-state convergence interval is initialized and constructed based on the design working condition parameters and green construction specification thresholds specified in the engineering construction drawings and supporting construction technical documents. For each operating condition, the mean confidence interval and standard deviation fluctuation range of the benchmark data are selected as the steady-state convergence boundary. The sliding window Markov state inference algorithm is used to iteratively determine the evolution trend of the real-time feature sequence. If the fluctuation of the stable feature is within the steady-state convergence interval for a consecutive preset number of frames, it is determined that the monitoring data is converging towards the steady-state operating condition. If the discrete mutation feature breaks through the boundary of the steady-state convergence interval for multiple consecutive frames and the mutation frequency exceeds the statistical normal, it is determined that the monitoring data has a collection offset or an abnormal operating condition. The boundary of the steady-state convergence interval is determined according to the three Sigma principle. Read the historical structured feature dataset of standardized sensor extension mounting interfaces for all BIM grid partitions of the building site, construct a global multi-interface feature benchmark library, and perform dimension-by-dimensional residual operations on the real-time feature sequences of incremental sensor nodes with the sensor features of their respective partitions and the cross-partition sensor features of the entire domain to generate a bidirectional feature residual matrix. The expression of the bidirectional feature residual matrix is ​​as follows: ; In the formula, R represents the two-way characteristic residual matrix; matrix subscripts... The first subscript i represents the monitoring feature number, i takes the values ​​1, 2, 3, and 4, which correspond to the time-series stationary feature of dust mean, the time-series stationary feature of energy consumption standard deviation, the instantaneous noise mutation amplitude, and the sampling breakpoint interval mutation feature, respectively; the second subscript j represents the residual type, j=1 corresponds to the residual of the partition neighborhood feature, j=2 corresponds to the residual of the global clustering feature. Let be the local stock sensor deviation value corresponding to the i-th monitoring feature. This represents the global similar sensing deviation value corresponding to the i-th monitoring feature; The sudden shift component and steady-state deviation component of the incremental sensing node are offset hierarchically based on the bidirectional feature residual matrix to filter out cold-start random drift noise. The long-term feature stability, data efficiency, and working condition adaptability of all sensing interfaces are quantified to obtain a benchmark reliability score for each sensing interface. Feature fusion weights are adaptively allocated based on these scores. The real-time features of the incremental sensing node, the regional steady-state features, and the global high-matching benchmark features are weighted and coupled together to output a standardized feature vector that eliminates cold-start drift and matches the on-site construction conditions. This standardized feature vector is generated by the feature distribution and weight distribution of mature existing sensing interfaces across the entire domain and is used to classify the data drift caused by cold starts in the incremental sensing node. The calculation formula for the benchmark reliability score of each sensing interface is as follows: ; In the formula, This is represented as the reliability score of the s-th standardized sensor extension mount interface; Represented as a long-term characteristic stability metric; This is expressed as a quantified value representing the percentage of valid data. This is represented as a quantified value for working condition adaptation and matching; , and These represent the fixed weighted adjustment coefficients corresponding to stability, efficiency, and matching degree, respectively, which are preset constants for the system. The feature fusion weight ratio is adaptively allocated based on the credibility score. The real-time features of incremental sensing nodes, regional steady-state features, and global high-matching benchmark features are weighted and coupled together to output a standardized feature vector that eliminates cold start drift and matches the on-site construction conditions. The existing monitoring system can only select a single dimension feature as a data reference for newly connected zero-sample cold start sensor nodes. It does not distinguish between device deployment scenarios and adapt to multiple benchmark construction methods. At the same time, it does not perform layered analysis on the stable and abrupt features of the monitoring data, and cannot layeredly remove the multi-level cold start offset noise brought by the newly added devices. For newly connected incremental sensor nodes, three types of working conditions are divided and steady-state convergence intervals are built respectively. Multiple reference dimensions are integrated, including partitioned existing data, global similar data, and construction design specifications. The time-series stable and local mutation features are split and quantitatively calculated. Data deviations are offset by multi-interface feature residual calculations across the entire domain and cold start drift interference is filtered out. Step S4: Using the global weight-constrained standardized feature vector output in step S3 as the discrimination benchmark, perform frame-by-frame feature matching verification on the raw monitoring data collected in real time by the incremental sensing nodes; compare and match the dimension parameters, amplitude range, and temporal fluctuation range of the raw data features with the standardized feature vector, retain the raw sampling frames with consistent feature distribution as reliable and valid data, and perform local isolation caching processing on distorted data frames that deviate from the benchmark fluctuation range and correspond to cold start random drift, restricting distorted data from participating in global feature statistics and weight calculation; The full-domain multi-interface feature benchmark library is incrementally updated using the filtered reliable and valid data. Based on the updated full-domain multi-interface feature benchmark library, the long-term feature stability, data efficiency, and working condition adaptability of all standardized sensor extension mount interfaces are requantified, and the benchmark reliability score of each sensor interface is iteratively generated. Normalize the credibility scores of each interface and update the feature fusion weight ratio of existing and incremental sensing interfaces in the whole domain simultaneously. Write the updated whole domain fusion weight parameters into the digital twin BIM grid partitioning simulation calculation operator, perform green construction multi-index coupling simulation calculation based on the updated weights, and output the whole domain green construction working condition simulation results. The existing monitoring does not perform data screening and filtering for newly added cold start sensor nodes. It directly uses the raw data mixed with power-on drift noise to participate in feature fusion and weight calculation. It relies on a single dimension to assign weights, which can easily cause the weight distribution of the entire domain to shift. This results in the weight of the existing sensor interface being diluted by the instantaneously distorted data. At the same time, it cannot effectively distinguish between equipment drift and real construction micro-anomalies. The standardized feature vectors generated by the joint constraints of the entire domain's existing interfaces enable accurate frame-by-frame verification. Within a single acquisition cycle, it quickly removes cold-start random drift and distortion data, using only reliable and valid data to participate in weight iteration. In a single round of calculation, it immediately avoids the weight crowding and weight offset problems caused by the addition of zero-sample nodes. Through multi-dimensional global benchmark matching, it distinguishes the authenticity of data, effectively differentiates instantaneous operating condition anomalies from equipment noise, and improves the accuracy of single-frame green construction monitoring and judgment. Step S5: Establish a temporal correlation binding relationship for the simulation results of the green construction conditions across the entire domain, the reliable and valid original sampling frames, the standardized feature vectors, the updated fusion weight parameters of each sensor interface, the reliability score, and the partition anomaly identifier, and write them into the partition event storage area of ​​the corresponding BIM grid partition to complete persistent storage. Read all historical data that has been archived in the partitioned event storage area and add it to the global multi-interface feature benchmark library. Expand the working condition feature samples of each construction stage and each BIM grid partition. After the global multi-interface feature benchmark library is updated, use the newly added historical feature samples in the global multi-interface feature benchmark library as the reference data source for the subsequent construction of steady-state convergence interval and residual correction calculation of newly added incremental sensor nodes. Execute the data correction, data filtering, weight iteration and simulation calculation process from step S3 to step S4 in a loop to monitor the construction of the entire building area. The simulation results of the full-domain green construction conditions are divided into multi-dimensional time-series features by BIM grid partition. Four types of simulation-derived features are extracted: dust time-series fluctuation features, energy consumption cumulative change features, noise instantaneous change features, and sensor benchmark reliability time-series drift features. A sliding time-series window is selected to extract the feature sequence of continuous simulation outputs. The year-on-year change amplitude, month-on-month offset difference, and periodic fluctuation attenuation coefficient are calculated for each feature dimension to complete the quantitative analysis of feature changes. Synchronously extract the original temporal features of the distorted sampling frames stored in the local isolation cache, perform a global distribution matching comparison between the cache temporal features and the historical steady-state features archived by the other standardized sensor interfaces of the partition, and combine the matching results of the simulation features and the steady-state convergence interval of the partition to comprehensively determine whether to start the weight update process; If only a single incremental sensor cache feature deviates independently, the historical steady-state features of other interfaces in the partition do not deviate synchronously, and all simulation-derived features fall within the steady-state convergence interval of the corresponding partition, and the temporal fluctuation amplitude is lower than the preset threshold, the deviation is determined to be only the sensor's own sampling noise, and the weight update process is allowed to be started. The contribution of each sensor interface in each dimension of the current round's bidirectional feature residual matrix is ​​retrieved, and the fusion weight of the high-drift sensor interface is reduced in reverse according to the residual contribution ratio, while the weight ratio of the high-confidence existing sensor interface in the same partition is increased synchronously, thus completing the second adaptive feedback adjustment of the global feature fusion weight. If the local isolation cache features deviate from the historical steady-state features of multiple interfaces in the partition in the same direction and synchronously, or if any type of simulation-derived feature breaks through the boundary of the steady-state convergence interval or the temporal fluctuation amplitude exceeds the standard, it is determined that the partition has an abnormal real green working condition, and the weight update process of all sensor interfaces in the partition is directly blocked to maintain the current weight ratio unchanged. Until the local isolated cache features regress to the historical steady-state distribution range of the whole domain interface in subsequent cyclic monitoring, and all simulation-derived features fall back to the steady-state convergence range, the weight iteration permission is reopened, and the new fusion weights after feedback adjustment are written into the corresponding BIM grid partition simulation calculation operator for the next round of green construction multi-index coupling simulation calculation. Existing solutions rely solely on initial global or local single features as a reference for cold-start sensor node verification. Without a dynamic sample accumulation mechanism, it is difficult to distinguish between equipment drift noise and actual construction anomalies. Distorted data continuously participates in global calculations, which dilutes the weight of existing sensors over a long period, causing local pollution and energy consumption anomalies to go undetected. Static fixed benchmarks cannot adapt to phased dynamic construction and cannot take into account both the inherent working conditions of buildings and the needs of real-time monitoring. By continuously expanding the full-domain feature benchmark library with reliable operating condition data, we enrich the reference samples for multiple zones and construction stages, breaking the limitations of a single reference. Distorted data is isolated from the database throughout the entire process, preventing weight dilution from the cyclical process. The benchmark library accumulates various real abnormal features, accurately distinguishing between equipment noise and construction anomalies, and reducing missed detections. The benchmark samples are continuously updated with the process, while also bearing the inherent operating condition constraints of the building and dynamic construction characteristics, adapting to full-cycle monitoring.

[0006] Furthermore, a green construction monitoring system based on digital twins is provided, which includes a digital twin base construction module, an incremental sensor access processing module, a feature correction weight solution module, and a full-domain simulation cyclic monitoring module. The digital twin base construction module is used to build a standardized BIM spatial skeleton model and configure standardized sensor extension mounting interfaces; the incremental sensor access processing module is used to time-series match and access incremental sensor nodes and normalize the original monitoring data; the feature correction weight solution module is used to analyze sensor features, construct a steady-state convergence interval, and generate a bidirectional feature residual matrix to solve for standardized feature vectors; the full-domain simulation cyclic monitoring module is used to verify monitoring data, execute green construction simulation, and archive data to achieve cyclic monitoring. The digital twin base construction module includes a BIM skeleton building unit and a sensor interface pre-configuration unit; the BIM skeleton building unit is used to parse engineering drawings to build a standardized BIM spatial skeleton model and divide BIM grid partitions; the sensor interface pre-configuration unit is used to integrate standardized sensor extension mounting interfaces at the bottom layer of BIM grid partitions and configure supporting subroutines and routing links. The incremental sensor access processing module includes a timing matching and binding unit and a monitoring data normalization unit; the timing matching and binding unit is used to match the construction timing to activate the standardized sensor extension mounting interface and complete the incremental sensor node binding mapping; the monitoring data normalization unit is used to process the original monitoring data according to the unified interface input parameter specification and generate a structured incremental monitoring dataset. The feature correction weight solution module includes a working condition interval construction unit and a bidirectional residual calculation unit. The working condition interval construction unit is used to calculate time-series stationary features and local discrete abrupt change features, construct steady-state convergence intervals for different working conditions and determine data status. The bidirectional residual calculation unit is used to construct a global multi-interface feature benchmark library, calculate and generate a bidirectional feature residual matrix and solve for the sensor interface fusion weights. The full-domain simulation cycle monitoring module includes a data verification simulation unit and a time-series cycle archiving unit. The data verification simulation unit is used to verify the original monitoring data based on standardized feature vectors, update the feature benchmark library, and calculate and output the simulation results of the full-domain green construction conditions. The time-series cycle archiving unit is used to persistently store the simulation results and valid monitoring data, and call the archived samples to perform full-process monitoring calculations in a loop.

[0007] Compared with the prior art, the beneficial effects of the present invention are: 1. This invention builds an scalable digital twin monitoring base by pre-setting standardized sensor extension mounting interfaces at the bottom layer of the BIM grid partition. It enables plug-and-play access of zero-sample sensor nodes at each construction stage without reconstructing the model topology, solving the problem that the existing closed model architecture requires topology modification for expansion and cannot adapt to phased incremental monitoring.

[0008] 2. This invention is based on splitting sensing features into two layers and constructing steady-state convergence intervals according to different operating conditions. It uses a two-way feature residual matrix to peel off multi-layer drift noise during cold start and adaptively solves the sensing fusion weights based on multi-dimensional benchmarks. This solves the problem that existing methods only use a single feature reference and cannot distinguish between equipment power-on drift and actual pollution or minor energy consumption anomalies on site.

[0009] 3. This invention isolates distorted data frame by frame, iteratively updates the full-domain feature benchmark using only valid samples, continuously expands the reference samples by archiving full-scale working condition data in time series, and monitors in a closed loop. This avoids the dilution of existing sensor weights by distorted data. At the same time, it distinguishes between equipment noise and actual on-site exceedances based on joint verification of simulation and cached features, and controls weight updates accordingly. This not only optimizes the accuracy of monitoring simulation in a normal way, but also prevents real construction anomalies from being covered by weighting, taking into account both the inherent working condition constraints of buildings and the needs of dynamic construction monitoring throughout the entire cycle. Attached Figure Description

[0010] Figure 1 This is a flowchart illustrating a green construction monitoring method based on digital twins according to the present invention. Detailed Implementation

[0011] 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.

[0012] Example 1: As Figure 1 As shown, this invention provides a technical solution: a green construction monitoring method based on digital twins. The green construction monitoring method includes the following steps: Analyze engineering drawings to build a standardized BIM spatial skeleton model, and pre-set standardized sensor extension mounting interfaces based on the construction phase; Based on the hierarchical construction sequence logic obtained from the analysis of engineering drawings, the entire construction space is divided into independent BIM grid partitions that match the foundation, main structure, electromechanical installation, and decoration and finishing construction stages; An open extension channel is natively integrated at the bottom layer of each BIM grid partition model as a standardized sensor extension mounting interface. The standardized sensor extension mounting interface has built-in communication protocol adaptation, automatic spatial component mapping, incremental data isolation caching, and neighborhood feature parameter migration subroutines. Each standardized sensor extension mounting interface is uniquely bound to its corresponding BIM grid partition and is configured with an independent data routing link. In practical implementation, taking the green monitoring of the entire construction cycle of a high-rise residential building as an example, we integrate the static parameters of components, site, and time sequence from the complete set of construction drawings to construct a standardized BIM spatial skeleton model that corresponds one-to-one with the actual construction site. We divide the building into independent BIM grid zones according to the four major construction processes and natively integrate standardized sensor extension mounting interfaces at the bottom layer of each zone. We reserve access channels that do not require modification for the phased addition of sensors. The standardized sensor extension mounting interfaces are bound one-to-one with their respective BIM grid zones. Independent data distribution links must not communicate across zones. The interfaces have a complete set of built-in data processing subroutines that cannot be deleted or replaced.

[0013] By matching the on-site construction progress and activating the corresponding standardized sensor extension mounting interface, the incremental sensor node binding is completed, and the original monitoring data is preprocessed according to the unified interface input parameter specification to generate a structured incremental monitoring dataset. The partition interface parameters, standard timing, and monitoring boundaries are defined as access constraints based on the engineering construction drawings and supporting construction technical documents. Collect on-site construction progress data and match it with time sequence nodes to determine the corresponding BIM grid partition construction stage; when the target stage is reached, activate the corresponding standardized sensor extension mounting interface to complete incremental sensor node identification, protocol adaptation and BIM component binding. In practical implementation, taking the green monitoring of the entire construction cycle of a high-rise residential building as an example, the zoning interface parameters, standard construction sequence, and monitoring boundaries are pre-defined according to the engineering drawings as unified constraints. The on-site construction progress and the timing node are matched in real time to determine the currently effective BIM grid zoning. The corresponding standardized sensor extension mounting interface is activated to complete the binding of the newly added sensor device. The original monitoring data is organized according to the unified interface input parameter specification to obtain a structured incremental monitoring dataset with a unified format. The incremental sensor node is only allowed to access the standardized sensor extension mounting interface matched in the current construction stage. The original data with misaligned timing or uncorrected parameters is prohibited from flowing into the structured incremental monitoring dataset.

[0014] Feature extraction and analysis were performed on the structured incremental monitoring dataset to construct a steady-state convergence interval and a global multi-interface feature benchmark library. The standardized feature vector of the incremental sensing node was obtained by combining the residual matrix output. Based on the global multi-interface feature linkage comparison mechanism, the incremental sensing data adaptive feature correction and dynamic weight solution are completed; the structured incremental monitoring dataset, which has been preprocessed by the unified interface input parameter specification, is subjected to two-layer differential feature analysis and quantitative calculation of time-stationary features and local discrete mutation features. The mean, time series standard deviation, periodic fluctuation amplitude, and long-term benchmark offset of the monitoring indicators are calculated window by window using a sliding time series window as time series stationary features; the instantaneous peak jump amplitude, short-term mutation cumulative frequency, cross-regional sensor data difference, and sampling breakpoint interval are solved by point-by-point difference operation as local discrete mutation features. Based on the deployment data of incremental sensor nodes, a steady-state convergence interval is adaptively constructed under three operating conditions: In the first working case, when the BIM grid partition to which the incremental sensing node belongs has the same type of existing sensing interface deployed, the steady-state convergence interval is generated by self-learning based on the historical time-series characteristics statistical distribution of the existing sensing interface of the partition. In the second scenario, when the BIM grid partition to which the incremental sensor node belongs has no similar existing sensor interfaces, but there are similar sensor interfaces in the entire building area, the global feature distribution clustering algorithm is used to screen and match the current construction process and the site environment to fit and construct a steady-state convergence interval. In the third working condition, when there are no similar sensor interfaces in the entire building area and the incremental sensor nodes are brand new zero-sample monitoring devices, the steady-state convergence interval is initialized and constructed based on the design working condition parameters and green construction specification thresholds specified in the engineering construction drawings and supporting construction technical documents. For each operating condition, the mean confidence interval and standard deviation fluctuation range of the benchmark data are selected as the steady-state convergence boundary. The sliding window Markov state inference algorithm is used to iteratively determine the evolution trend of the real-time feature sequence. If the fluctuation of the stable feature for a consecutive preset number of frames is within the steady-state convergence interval, it is determined that the monitoring data is converging towards the steady-state operating condition. If the discrete abrupt feature breaks through the boundary of the steady-state convergence interval for multiple consecutive frames and the abrupt frequency exceeds the statistical normal, it is determined that the monitoring data has a collection offset or an abnormal operating condition. Read the historical structured feature dataset of standardized sensor extension mounting interfaces of all BIM grid partitions of the building site, construct a global multi-interface feature benchmark library, and perform dimension-by-dimensional residual calculations on the real-time feature sequences of incremental sensor nodes with the sensor features of their respective partitions and the cross-partition sensor features of the entire domain to generate a bidirectional feature residual matrix. Based on the bidirectional feature residual matrix, hierarchical component cancellation processing is performed on the abrupt offset component and steady-state deviation component of the incremental sensing node; the long-term feature stability, data efficiency, and working condition adaptation matching degree of all sensing interfaces are quantified to generate a benchmark credibility score for each sensing interface; the fusion weight ratio corresponding to the features of each interface is adaptively allocated according to the credibility score; according to the allocated fusion weight ratio, weighted coupling calculation is performed on the real-time features, partition steady-state features, and global high-matching benchmark features of the incremental sensing node, and a standardized feature vector is output. In practical implementation, taking the green monitoring of the entire construction cycle of high-rise residential buildings as an example, the structured incremental monitoring dataset is split into two types of features: time-series stationary features and local discrete abrupt change features. Based on the sensor deployment environment, three types of working conditions are adaptively constructed to form a steady-state convergence interval. The residual vectors of the neighboring sensor in the partition and the clustering residual vectors of the same type of sensor in the whole domain are calculated respectively. The two sets of vectors are concatenated column by column to generate a bidirectional feature residual matrix. Each row of the matrix corresponds to a single monitoring feature dimension. The first column is the local residual of the partition and the second column is the clustering residual of the whole domain. Based on the matrix layering, the two types of cold start noise, steady-state deviation and abrupt shift, are offset. Combined with multi-dimensional benchmark weighted coupling, standardized feature vectors are generated. The bidirectional feature residual matrix must completely retain the two columns of independent residual data. It is not allowed to merge the two types of residual components in advance and then perform deviation offset calculation.

[0015] The original data is verified by standardized feature vectors and the global multi-interface feature benchmark library is updated. The global working condition simulation results are calculated and output based on the updated results. Based on the standardized feature vector as the discrimination benchmark, a frame-by-frame feature matching verification is performed on the raw monitoring data collected in real time by the incremental sensing nodes; the features of the raw data are compared and matched with the dimension parameters, amplitude range and temporal fluctuation range of the standardized feature vector. Create an incremental sensor node baseline set, determine the original sampling frames with consistent feature distribution as valid data frames and store them in the incremental sensor node baseline set; determine the original sampling frames with inconsistent feature distribution as distorted data frames and perform local isolation caching processing. Valid data frames from the incremental sensor node benchmark set are selected to incrementally update the global multi-interface feature benchmark library. Based on the updated global multi-interface feature benchmark library, the long-term feature stability, data efficiency and working condition adaptability of all standardized sensor extension mount interfaces are requantified, and the benchmark credibility score of each sensor interface is iteratively generated. Normalize the credibility scores of each interface and update the feature fusion weight ratio of existing and incremental sensing interfaces in the whole domain simultaneously. Write the updated whole domain fusion weight parameters into the digital twin BIM grid partitioning simulation calculation operator, perform green construction multi-index coupling simulation calculation based on the updated weights, and output the whole domain green construction working condition simulation results. In specific implementation, taking the green monitoring of the entire construction cycle of a high-rise residential building as an example, the standardized feature vector output in step three is used as a unified judgment standard to compare the original monitoring data frame by frame. The reliable and valid sampling frames are separated from the cold start distorted sampling frames and the distorted data is isolated. After updating the full-domain multi-interface feature benchmark library with valid samples, the reliability score and fusion weight of each sensor interface are recalculated. The updated weights are substituted into the BIM grid partitioning simulation operator to perform multi-index coupling simulation and output the full-domain working condition simulation results. The distorted data stored locally is prohibited from being written into the full-domain multi-interface feature benchmark library throughout the process. The weight normalization calculation must cover all standardized sensor extension mounting interfaces in the project.

[0016] Store the simulation results of the entire working condition and effective monitoring data, and monitor the construction in a loop based on the stored data; The simulation results of green construction conditions across the entire domain, reliable and valid sampling frames, standardized feature vectors, sensor interface fusion weight parameters, credibility scores, and partition anomaly identifiers are sequentially bound and persistently stored in the partition event storage area of ​​the corresponding BIM grid partition; archived historical data is retrieved to expand the multi-interface feature benchmark library across the entire domain; with the updated benchmark library samples as a reference, feature correction, data filtering, weight iteration, and simulation calculations are performed iteratively to achieve continuous monitoring of the entire domain; In specific implementation, taking the green monitoring of the entire construction cycle of a high-rise residential building as an example, the simulation results, effective sampling frames, feature vectors, weight parameters, confidence scores and partition anomaly identifiers are linked and bound in a time sequence and persistently stored in the partition event storage area of ​​the corresponding BIM grid partition. Archived historical samples are retrieved to supplement and expand the full-domain multi-interface feature benchmark library. The updated benchmark library is used as the reference for newly added sensing devices. The entire process of feature parsing and correction, data identification and screening, weight iterative update, and multi-index simulation calculation is executed cyclically to achieve uninterrupted full-domain construction monitoring. The historical data in the partition event storage area is only used to supplement the full-domain multi-interface feature benchmark library. During the main construction phase, a new batch of zero-sample dust sensors were added. These sensors exhibited temperature drift upon power-up. Half of the samples closely matched the actual on-site conditions, while the other half drifted randomly. The system first screened frame by frame, updating the global baseline only with valid samples. Initial weights were generated based on sensor stability, efficiency, and condition matching to complete the simulation. Simultaneously, the simulation time-series fluctuations, isolated cached distorted data, and historical curves of existing sensors were compared. If only a single newly added sensor deviated independently while the data from other devices remained stable, it was determined to be hardware noise. The weights were fine-tuned based on the deviation contribution using a bidirectional feature residual matrix to optimize simulation accuracy. If multiple sensors in a zone simultaneously exceeded the steady-state range and the simulation curves continuously exceeded the limits, it was determined to be an abnormality in actual dust from earthwork excavation. All weight update permissions for this zone were directly locked, maintaining the original ratio to prevent diluting the excessive signal. Weight iteration was automatically unlocked after construction ended and all features returned to the steady-state convergence range, entering the next simulation cycle. The design of this invention avoids three core problems: First, zero-sample sensing cold start distortion data mixing dilutes the existing sensing weights and simulation results; second, indiscriminate uniform weight adjustment masks the actual exceedance of construction site dust and energy consumption, causing missed risk detection; third, the weight iteration lacks closed-loop control, and the abnormality is continuously weakened during the exceedance stage, requiring manual intervention to restore parameters. Simultaneously, it achieves pre-isolation distortion sampling to purify the full-domain feature benchmark from the source, eliminating the need for manual screening of abnormal sensors; it accurately distinguishes two types of offset sources through dual verification of simulation time series and cross-device historical features; it adaptively adjusts parameters in normal noise scenarios to improve simulation fit, and locks weights to fully retain risk signals in real excessive scenarios; it autonomously unlocks and iterates after the working condition is restored, adapting to dynamic monitoring throughout the entire construction phase without the need for manual intervention.

[0017] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the invention can be implemented in other specific forms without departing from the spirit or essential characteristics of the invention. Therefore, the embodiments should be considered in all respects as exemplary and non-limiting, and the scope of the invention is defined by the appended claims rather than the foregoing description. Thus, it is intended that all variations falling within the meaning and scope of equivalents of the claims be included within the present invention.

Claims

1. A green construction monitoring method based on digital twinning, characterized in that: The green construction monitoring method includes: Analyze engineering drawings to build a standardized BIM spatial skeleton model, and pre-set standardized sensor extension mounting interfaces based on the construction phase; By matching the on-site construction progress and activating the corresponding standardized sensor extension mounting interface, the incremental sensor node binding is completed, and the original monitoring data is preprocessed according to the unified interface input parameter specification to generate a structured incremental monitoring dataset. Feature extraction and analysis were performed on the structured incremental monitoring dataset to construct a steady-state convergence interval and a global multi-interface feature benchmark library. The standardized feature vector of the incremental sensing node was obtained by combining the residual matrix output. The original data is verified by standardized feature vectors and the global multi-interface feature benchmark library is updated. The global working condition simulation results are calculated and output based on the updated results. It stores the simulation results of the entire operating conditions and effective monitoring data, and monitors the construction in a loop based on the stored data.

2. The green construction monitoring method based on digital twinning according to claim 1, characterized in that: The aforementioned standardized sensor extension mounting interface, pre-set during the construction phase, includes: Based on the hierarchical construction sequence logic obtained from the analysis of engineering drawings, the entire construction space is divided into independent BIM grid partitions that match the foundation, main structure, electromechanical installation, and decoration and finishing construction stages; An open extension channel is natively integrated at the bottom layer of each BIM grid partition model as a standardized sensor extension mounting interface. The standardized sensor extension mounting interface has built-in communication protocol adaptation, automatic spatial component mapping, incremental data isolation caching, and neighborhood feature parameter migration subroutines. Each standardized sensor extension mounting interface is uniquely bound to its corresponding BIM grid partition and is configured with an independent data routing link.

3. The green construction monitoring method based on digital twins according to claim 1, characterized in that: The generated structured incremental monitoring dataset includes: The partition interface parameters, standard timing, and monitoring boundaries are defined as access constraints based on the engineering construction drawings and supporting construction technical documents. Collect on-site construction progress data and match it with time sequence nodes to determine the corresponding BIM grid partition construction stage; when the target stage is reached, activate the corresponding standardized sensor extension mounting interface to complete incremental sensor node identification, protocol adaptation and BIM component binding.

4. The green construction monitoring method based on digital twin according to claim 1, characterized in that: The standardized feature vector of the incremental sensing node is obtained by combining the residual matrix output, including: Based on the global multi-interface feature linkage comparison mechanism, the incremental sensing data adaptive feature correction and dynamic weight solution are completed; the structured incremental monitoring dataset, which has been preprocessed by the unified interface input parameter specification, is subjected to two-layer differential feature analysis and quantitative calculation of time-stationary features and local discrete mutation features. The mean, time series standard deviation, periodic fluctuation amplitude, and long-term benchmark offset of the monitoring indicators are calculated window by window using a sliding time series window as time series stationary features; the instantaneous peak jump amplitude, short-term mutation cumulative frequency, cross-regional sensor data difference, and sampling breakpoint interval are solved by point-by-point difference operation as local discrete mutation features. Based on the deployment data of incremental sensor nodes, a steady-state convergence interval is adaptively constructed under three operating conditions: In the first working case, when the BIM grid partition to which the incremental sensing node belongs has the same type of existing sensing interface deployed, the steady-state convergence interval is generated by self-learning based on the historical time-series characteristics statistical distribution of the existing sensing interface of the partition. In the second scenario, when the BIM grid partition to which the incremental sensor node belongs has no similar existing sensor interfaces, but there are similar sensor interfaces in the entire building area, the global feature distribution clustering algorithm is used to screen and match the current construction process and the site environment to fit and construct a steady-state convergence interval. In the third operating condition, when there are no similar sensor interfaces in the entire building area and the incremental sensor nodes are brand new zero-sample monitoring devices, the steady-state convergence interval is initialized and constructed based on the design operating condition parameters and green construction specification thresholds specified in the engineering construction drawings and supporting construction technical documents.

5. The green construction monitoring method based on digital twins according to claim 4, characterized in that: The process of combining the residual matrix output to obtain the standardized feature vector of the incremental sensing node also includes: For each operating condition, the mean confidence interval and standard deviation fluctuation range of the benchmark data are selected as the steady-state convergence boundary. The sliding window Markov state inference algorithm is used to iteratively determine the evolution trend of the real-time feature sequence. If the fluctuation of the stable feature for a consecutive preset number of frames is within the steady-state convergence interval, it is determined that the monitoring data is converging towards the steady-state operating condition. If the discrete abrupt feature breaks through the boundary of the steady-state convergence interval for multiple consecutive frames and the abrupt frequency exceeds the statistical normal, it is determined that the monitoring data has a collection offset or an abnormal operating condition. Read the historical structured feature dataset of standardized sensor extension mounting interfaces of all BIM grid partitions of the building site, construct a global multi-interface feature benchmark library, and perform dimension-by-dimensional residual calculations on the real-time feature sequences of incremental sensor nodes with the sensor features of their respective partitions and the cross-partition sensor features of the entire domain to generate a bidirectional feature residual matrix. Based on the bidirectional feature residual matrix, hierarchical component cancellation processing is performed on the abrupt offset component and steady-state deviation component of the incremental sensing node; the long-term feature stability, data efficiency, and operating condition adaptation matching degree of all sensing interfaces are quantified to generate a benchmark credibility score for each sensing interface, and the fusion weight ratio corresponding to each interface feature is adaptively allocated according to the credibility score; according to the allocated fusion weight ratio, weighted coupling calculation is performed on the real-time features, partition steady-state features, and global high-matching benchmark features of the incremental sensing node, and a standardized feature vector is output.

6. The green construction monitoring method based on digital twin according to claim 1, characterized in that: The calculation and output of the global operating condition simulation results based on the updated results includes: Based on the standardized feature vector as the discrimination benchmark, a frame-by-frame feature matching verification is performed on the raw monitoring data collected in real time by the incremental sensing nodes; the features of the raw data are compared and matched with the dimension parameters, amplitude range and temporal fluctuation range of the standardized feature vector. Create an incremental sensor node baseline set, determine the original sampling frames with consistent feature distribution as valid data frames and store them in the incremental sensor node baseline set; determine the original sampling frames with inconsistent feature distribution as distorted data frames and perform local isolation caching processing. Valid data frames from the incremental sensor node benchmark set are selected to incrementally update the global multi-interface feature benchmark library. Based on the updated global multi-interface feature benchmark library, the long-term feature stability, data efficiency and working condition adaptability of all standardized sensor extension mount interfaces are requantified, and the benchmark credibility score of each sensor interface is iteratively generated. Normalize the credibility scores of each interface and update the feature fusion weight ratio of existing and incremental sensor interfaces across the entire domain simultaneously. Write the updated global fusion weight parameters into the digital twin BIM grid partitioning simulation calculation operator, perform multi-index coupled simulation calculation of green construction based on the updated weights, and output the simulation results of green construction conditions across the entire domain.

7. A green construction monitoring method based on digital twins according to claim 6, characterized in that: The method of monitoring construction by cyclically executing stored data also includes: Multi-dimensional time-series feature quantitative analysis is performed on the simulation results of the green construction conditions in the whole domain. The distorted sampling frames stored in the local isolated cache are compared with the historical steady-state features archived by the other standardized sensor extension mounting interfaces in the partition. The time-series feature analysis results and the historical steady-state feature matching results are combined to determine whether to iteratively adjust the feature fusion weights. When the offset is determined to originate solely from the sampling noise of the sensing device itself, the bidirectional feature residual matrix is ​​retrieved, and the feature fusion weights are iteratively adjusted adaptively based on the residual contribution ratio of each standardized sensing extension mounting interface. When a partition is determined to have an abnormality in real green working conditions, the feature fusion weight iterative update of all standardized sensor extension mounting interfaces of that partition is blocked, while the original weight ratio remains unchanged. After the distorted sampling frame features in the local isolation cache and the simulation time series features of the green construction working conditions of the whole domain have returned to the steady-state convergence interval of the corresponding partition, the blocking restriction is lifted, and the feature fusion weights that have completed iterative adjustment are written into the digital twin BIM grid partition simulation calculation operator to carry out the next round of green construction multi-index coupling simulation calculation.

8. A green construction monitoring system based on digital twins, applied to the green construction monitoring method based on digital twins as described in any one of claims 1-7, characterized in that: The green construction monitoring system includes a digital twin foundation construction module, an incremental sensor access processing module, a feature correction weight solution module, and a full-domain simulation cyclic monitoring module. The digital twin base construction module is used to build a standardized BIM spatial skeleton model and configure standardized sensor extension mounting interfaces; the incremental sensor access processing module is used to time-series match the access of incremental sensor nodes and normalize the original monitoring data; the feature correction weight solution module is used to analyze sensor features, construct a steady-state convergence interval, and generate a bidirectional feature residual matrix to solve for standardized feature vectors; the global simulation cyclic monitoring module is used to verify monitoring data, execute green construction simulation, and archive data to achieve cyclic monitoring.

9. A green construction monitoring system based on digital twins according to claim 8, characterized in that: The digital twin base construction module includes a BIM skeleton building unit and a sensor interface pre-configuration unit; the BIM skeleton building unit is used to parse engineering drawings to build a standardized BIM spatial skeleton model and divide BIM grid partitions; the sensor interface pre-configuration unit is used to integrate standardized sensor extension mounting interfaces at the bottom layer of BIM grid partitions and configure supporting subroutines and routing links. The incremental sensor access processing module includes a timing matching and binding unit and a monitoring data normalization unit. The timing matching and binding unit is used to match the construction timing to activate the standardized sensor extension mounting interface and complete the incremental sensor node binding mapping. The monitoring data normalization unit is used to process the original monitoring data according to the unified interface input parameter specification and generate a structured incremental monitoring dataset.

10. A green construction monitoring system based on digital twins according to claim 8, characterized in that: The feature correction weight solution module includes a working condition interval construction unit and a bidirectional residual calculation unit. The working condition interval construction unit is used to calculate time-series stationary features and local discrete abrupt change features, construct steady-state convergence intervals for different working conditions and determine data status. The bidirectional residual calculation unit is used to construct a global multi-interface feature benchmark library, calculate and generate a bidirectional feature residual matrix and solve for the sensor interface fusion weights. The full-domain simulation cycle monitoring module includes a data verification simulation unit and a time-series cycle archiving unit. The data verification simulation unit is used to verify the original monitoring data based on standardized feature vectors, update the feature benchmark library, and calculate and output the simulation results of the full-domain green construction conditions. The time-series cycle archiving unit is used to persistently store the simulation results and valid monitoring data, and call the archived samples to perform full-process monitoring calculations in a loop.

Images