An intelligent matching method and system based on artificial intelligence

Abstract

This invention belongs to the field of building construction and digital twin technology, and relates to an intelligent matching method and system based on artificial intelligence. This invention collects construction environment data in real time and identifies risk markers, calculates component deformation using a time-series prediction model, and corrects the virtual components generated by the BIM model for environmental compensation. It employs 3D point cloud registration technology for geometric matching analysis to determine hole alignment errors and interface gaps. Based on a dynamic decision tree model, it dynamically adjusts the installation sequence by comprehensively considering the degree of geometric mismatch, environmental risks, and construction logic. Furthermore, it incrementally updates the model based on actual installation deviations. This invention effectively solves the technical problem of component deformation and installation mismatch caused by environmental factors such as temperature and humidity during construction, achieving dynamic adaptation and precision control of the construction process, improving the matching success rate and construction efficiency of component assembly, and continuously optimizing the system's adaptability and reliability through continuous learning.

Description

Technical Field

[0001] This invention belongs to the field of building construction and digital twin technology, and relates to an intelligent matching method and system based on artificial intelligence. Background Technology

[0002] With the widespread application of Building Information Modeling (BIM) technology in the installation of large steel structures or precision curtain walls, changes in environmental factors such as temperature and humidity at the construction site cause physical deformation of components. This results in geometric deviations between BIM design parameters and actual components, leading to frequent problems such as excessive hole alignment errors and substandard interface fit during installation. Therefore, it is necessary to break through the limitations of traditional BIM static design and improve its dynamic adaptability to environmental deformation.

[0003] In existing technologies, some research has attempted to introduce artificial intelligence into BIM models. For example, Chinese invention patent publication number CN118468409B proposes an information processing method and system for BIM systems based on artificial intelligence. This method trains an artificial intelligence network to evaluate the functional weights of BIM components and formulates optimization schemes based on these weights, including strategies such as component accuracy optimization, component restoration, and multi-path backup. This method improves the integrity and robustness of BIM models to a certain extent.

[0004] However, the above method still has limitations. It only focuses on the optimization of BIM data and fails to consider the dynamic impact of environmental factors such as temperature and humidity at the construction site on the physical deformation of actual components. In environmentally sensitive construction scenarios such as steel structures and large prefabricated components, changes in temperature and humidity can cause deformations at the millimeter or even centimeter level. This environmentally induced component deformation and the deviation from BIM design parameters will directly lead to geometric mismatch between the model and the actual component, ultimately resulting in insufficient component installation matching accuracy and failure to meet precise docking requirements.

[0005] Furthermore, in actual construction, the alignment of holes and the surface fit of component interfaces directly determine the installation quality. However, existing technologies cannot accurately assess the degree of geometric mismatch between the actual components and the BIM model before installation, leading to frequent inefficient cycles of trial installation, adjustment, and re-trial installation on site. This severely reduces the efficiency of component installation and matching, violating the core requirements of intelligent matching.

[0006] Furthermore, in complex building structures, component installation follows a strict logical sequence. However, when environmental risks cause deformation of critical components, a fixed installation sequence can easily lead to matching conflicts. Existing technologies do not comprehensively consider dynamically adjusting the installation sequence based on real-time environmental risks and component deformation states, resulting in the superposition of matching conflicts under high-risk conditions. This can induce systemic installation matching failures, failing to achieve the goal of intelligent matching with dynamic adaptation. Summary of the Invention

[0007] In view of this, in order to solve the problems mentioned in the background technology, an intelligent matching method and system based on artificial intelligence is proposed.

[0008] The first aspect of the present invention proposes an intelligent matching method based on artificial intelligence, comprising the following steps: S1, real-time collection of temperature and humidity data and component location data in the construction environment, and triggering risk environment marking based on the dynamic change characteristics of the data.

[0009] S2. Input the temperature and humidity data, component material properties and BIM design dimensions corresponding to the risk environment markers into the pre-trained time series prediction model, and output the deformation prediction of the components.

[0010] S3. Correct the interface geometric parameters of the components to be matched in the BIM model based on the deformation prediction, and generate a virtual component model after environmental compensation.

[0011] S4. Based on the virtual component model, a geometric matching analysis is performed using a 3D point cloud registration algorithm. When the hole alignment error or interface gap value between the actual component and the virtual model exceeds the corresponding tolerance value, it is determined to be a geometric mismatch.

[0012] S5. Based on the geometric mismatch determination results, risk environment markings, and construction logic rules, the installation sequence of the components to be installed is adjusted through a dynamic decision tree model.

[0013] S6. After the component installation is completed, when the deviation between the actual installation state and the BIM benchmark model exceeds the historical average installation deviation value and the risk environment marker is triggered during the installation process, the time series prediction model and the dynamic decision tree model are incrementally updated.

[0014] The second aspect of the present invention proposes an intelligent matching system based on artificial intelligence, comprising the following modules: an environmental risk prediction module, used to collect temperature and humidity data and component location data in the construction environment in real time, and trigger risk environment marking based on the dynamic change characteristics of the data, inputting the temperature and humidity data, component material properties and BIM design dimensions corresponding to the risk environment marking into a pre-trained time-series prediction model, and outputting the deformation prediction amount of the component.

[0015] The model environment compensation module is used to correct the interface geometric parameters of the components to be matched in the BIM model based on the deformation prediction, and generate a virtual component model after environment compensation.

[0016] The geometric matching decision module is used to perform geometric matching analysis based on the virtual component model using a 3D point cloud registration algorithm. When the hole alignment error or interface gap value between the actual component and the virtual model exceeds the corresponding tolerance value, it is judged as geometric mismatch. Based on the geometric mismatch judgment result, risk environment markers and construction logic rules, the installation sequence of the components to be installed is adjusted through a dynamic decision tree model.

[0017] The incremental update module is used to incrementally update the time-series prediction model and the dynamic decision tree model when the deviation between the actual installation state and the BIM benchmark model exceeds the historical average installation deviation value and the risk environment marker is triggered during the installation process after the component installation is completed.

[0018] Compared with the prior art, the beneficial effects of the present invention are as follows: (1) The present invention directly outputs the quantified component deformation prediction by inputting the environmental data, component material and design dimensions corresponding to the risk marker into the pre-trained time series prediction model. This method establishes the mapping relationship between environmental disturbance-material response-deformation prediction, solves the problem of missing correlation between environmental variables and geometric deviation, realizes advanced digital simulation of deformation, and provides accurate data basis for subsequent geometric compensation.

[0019] (2) This invention generates a virtual component model after environmental compensation and uses a three-dimensional point cloud registration algorithm to calculate quantitative indicators such as hole alignment error and interface gap value. This method transforms abstract mismatch into measurable and comparable specific values, solves the problem of inaccurate prediction of geometric compatibility before installation, realizes the leap from qualitative judgment to quantitative analysis, and reduces the delay in construction period and waste of resources caused by blind trial installation.

[0020] (3) This invention constructs a dynamic decision tree model, which integrates multi-dimensional features such as the severity of geometric mismatch, risk frequency, component attributes, and construction logic, and outputs and adjusts the installation priority in real time. This method breaks through the rigid constraints of pre-fixed processes, solves the problem of rigid decision-making where installation strategies cannot be adaptively adjusted under risk disturbances, realizes real-time dynamic optimization of the installation sequence, and improves the robustness and success rate of the overall installation process under complex working conditions. Attached Figure Description

[0021] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0022] Figure 1 This is a flowchart illustrating the steps of an artificial intelligence-based intelligent matching method in this invention.

[0023] Figure 2 This is a flowchart of the method for obtaining the virtual component model in this invention.

[0024] Figure 3 This is a schematic diagram showing the connection of various modules in an artificial intelligence-based intelligent matching system according to the present invention. Detailed Implementation

[0025] 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.

[0026] Example 1

[0027] Please see Figure 1 As shown, the present invention proposes an intelligent matching method based on artificial intelligence, including the following steps: S1, real-time collection of temperature, humidity and component location data in the construction environment, and triggering risk environment marking based on the dynamic change characteristics of the data.

[0028] Considering that during the installation of large steel structures or precision curtain walls, the construction environment will create instantaneous temperature gradients and humidity differences, causing non-uniform thermal expansion and contraction deformation of metal components. This deformation will directly change the geometric dimensions and relative positions of the component interfaces, and directly affect the matching accuracy of subsequent component installation.

[0029] Therefore, in one specific embodiment, by deploying temperature and humidity sensors and GPS positioning modules or laser positioning devices in the construction area, and according to a preset sampling period, wherein the sampling period can be determined by on-site testing based on the construction rhythm, and is usually set to 5-15 minutes / time, real-time temperature and humidity data, including ambient temperature value and relative humidity value, and three-dimensional spatial position data of components, are collected.

[0030] Furthermore, the triggering of the risk environment marker specifically involves: sorting the collected temperature and humidity data by timestamp to form a temperature and humidity time series; extracting the horizontal plane projection coordinates from the collected component location data and sorting them by timestamp to form a location time series.

[0031] For local extremum detection of temperature and humidity time series, the detection window size, i.e., the number of reference sampling times before and after, should be determined based on the preset sampling period and the sensitivity to environmental changes. For example, when the sampling period is 5 minutes, detection can be performed using 3 sampling times before and after; if the environmental changes are drastic, the sampling period can be shortened or the window size reduced, such as 2 sampling times before and after, to improve the response speed.

[0032] If the temperature and humidity values ​​at a certain sampling time are greater than the values ​​at the adjacent sampling times within its window, it is determined to be a local maximum; otherwise, it is a local minimum. This indicates that a sudden change in environmental parameters may cause instantaneous deformation of the component.

[0033] Vector calculation is performed on the horizontal projection positions of two adjacent sampling periods in the position time sequence to obtain the component movement direction vector. If the angle between the movement direction vector of the current sampling period and the movement direction vector of the previous sampling period is greater than the preset critical angle, it is determined that the movement direction has reversed. This indicates that the hoisting trajectory is abnormal and that the installation reference has been offset.

[0034] The critical angle can be set according to the stability requirements of the component hoisting trajectory. For precision components that require strict straight-line hoisting, a strict threshold of 170°-175° can be used; for ordinary components that allow small adjustments, the threshold can be relaxed to 160°-170°.

[0035] When a local maximum or local minimum value is detected in the temperature and humidity data in the time series, or when the direction of movement of the component's horizontal projection position in two adjacent sampling periods is reversed, a risk environment marker is triggered.

[0036] S2. Input the temperature and humidity data, component material properties and BIM design dimensions corresponding to the risk environment markers into the pre-trained time series prediction model, and output the deformation prediction of the components.

[0037] If the identified risky environments, such as sudden changes in temperature and humidity or reversal of component movement direction, are only marked, it is impossible to quantify their specific impact on the geometric state of the components. This makes it impossible to predict the risk of geometric mismatch before installation, and also impossible to provide a basis for decision-making for dynamically adjusting the installation sequence.

[0038] Considering that the deformation of a component is a complex result of the combined effects of temperature and humidity changes, its own material properties, and the initial design dimensions, it is necessary to transform risk data into quantifiable deformation predictions through forecasting.

[0039] If predictions are based on a single factor, the accuracy will be insufficient due to ignoring the coupling effect of multiple factors. Consequently, it will be impossible to know the deformation of components in advance, resulting in geometric deviations between the BIM model and the actual components.

[0040] The time-series prediction model can learn the deformation patterns under the combined effect of the above-mentioned multiple factors using historical data, and achieve accurate prediction of deformation quantities such as interface displacement and hole size change of components in three-dimensional space.

[0041] Therefore, in one specific embodiment, the temperature and humidity time series data corresponding to the risk environment marker are first extracted, the mean μ and standard deviation σ of the time series data are calculated, and the normal data range is set as [μ-3σ, μ+3σ] according to the 3σ criterion. Temperature and humidity data points that exceed this range are identified as outliers and removed.

[0042] The material attribute parameters and BIM design dimensions of the components to be matched are extracted from the BIM benchmark model through the data interface. The material attribute parameters of the components are encoded and the BIM design dimensions are decomposed according to the three-dimensional coordinate system to extract key dimension parameters such as the X, Y, and Z axis coordinates of the interface center and the diameter of the holes.

[0043] It should be noted that the BIM benchmark model refers to the original BIM design model that has not undergone environmental compensation or modification, serving as a geometric and positional reference standard for construction and installation.

[0044] The three types of processed data are concatenated to form the model input vector, which is then input into the pre-trained temporal prediction model. The model extracts and maps features through its internal temporal convolutional network and fully connected layers, and outputs the deformation prediction quantity, which includes the displacement components of the interface center in the X-axis, Y-axis, and Z-axis directions in the three-dimensional spatial coordinate system.

[0045] Furthermore, the training method for the time-series prediction model is as follows: historical data from similar building construction projects are collected in advance, including temperature and humidity time-series data at different construction stages, corresponding component material properties, BIM design dimensions, and actual deformation data of components obtained through laser scanning. The above data are divided into training set, validation set, and test set in a ratio of 7:2:1 to form a complete training sample set.

[0046] A hybrid prediction model based on temporal convolutional networks and long short-term memory networks is constructed. The temporal prediction model is trained using a training sample set. The input layer features are temperature and humidity time series data, component material properties and BIM design dimensions, and the output layer labels are the three-dimensional displacement components of the actual deformation.

[0047] The early stopping method is adopted during training. By monitoring the prediction error on the validation set, training is stopped when the validation set error does not decrease for 10 consecutive rounds to determine the optimal number of training rounds. The mean absolute error (MAE) between the predicted deformation of the test set samples and the actual deformation is calculated. When the error value does not exceed the error threshold, the model is deemed to meet the accuracy requirements, thus obtaining the time series prediction model.

[0048] The error threshold is determined based on the design matching accuracy level of the component. For components such as large steel structures, the docking accuracy requirement is usually within ±5mm, and the threshold can be set to 2.0 mm. For components such as high-precision curtain wall units, the threshold should be increased to 1.0 mm or higher.

[0049] It should be noted that when there is no historical data for the first construction scenario, a general time series prediction model pre-trained based on historical data of similar projects is loaded as the initial model. This initial model has been trained with historical data from no less than 5 similar projects.

[0050] S3. Correct the interface geometric parameters of the components to be matched in the BIM model based on the deformation prediction, and generate a virtual component model after environmental compensation.

[0051] It should be noted that in this invention, BIM model is a general term that covers all types of models involved in the BIM modeling process, including the BIM benchmark model and its modified models.

[0052] The virtual component model specifically refers to the BIM model generated based on the BIM baseline model, after correcting its interface geometric parameters according to the environmental deformation prediction results.

[0053] Since the BIM benchmark model is designed based on standard environmental conditions and does not take into account the actual deformation of components caused by risky environments during construction, if the BIM benchmark model is used directly for geometric matching, the geometric differences between the model and the actual components will lead to an increase in matching error, which in turn will cause geometric mismatch problems.

[0054] The deformation prediction has quantified the three-dimensional displacement and dimensional changes of the component under risky conditions. Therefore, it is necessary to make targeted corrections to the key geometric parameters of the BIM model interface based on the prediction so that the corrected model can reflect the actual state of the component.

[0055] Please see Figure 2 As shown, in one specific embodiment, the method for obtaining the virtual component model is as follows: extracting the original coordinates of the interface center and the original diameter of the hole of the component to be matched from the BIM reference model, and extracting the deformation prediction amount.

[0056] Based on the set of boundary point coordinates of holes in the BIM model, the three-dimensional displacement component of each boundary point is projected onto the direction of the hole's central axis, which is calculated by a vector projection algorithm. The projection direction is the unit vector of the hole's central axis. The average displacement of all boundary points after projection is statistically analyzed and used as the change in hole diameter.

[0057] Adjust the X-axis value of the interface center coordinate of the component to be matched to the algebraic sum of the original X-axis value of the interface center coordinate and the X-axis displacement component. Similarly, adjust the Y-axis and Z-axis values. Adjust the hole diameter to the algebraic sum of the original hole diameter and the change in hole diameter.

[0058] The BIM model after the above adjustments is used as the virtual component model after environmental compensation.

[0059] S4. Based on the virtual component model, a geometric matching analysis is performed using a 3D point cloud registration algorithm. When the hole alignment error or interface gap value between the actual component and the virtual model exceeds the corresponding tolerance value, it is determined to be a geometric mismatch.

[0060] Since the virtual component model has compensated for the effects of environmental deformation, but the actual component may have additional errors during processing and transportation, the model prediction alone cannot fully reflect the actual state. Therefore, it is necessary to achieve accurate matching and judgment by directly comparing the measured data with the virtual model.

[0061] In one specific embodiment, the actual component to be matched is scanned from all angles using a laser scanning device deployed on-site to obtain high-precision three-dimensional point cloud data. A statistical filtering algorithm is used to remove noise points in the point cloud, and a voxel grid downsampling algorithm is used to simplify the point cloud data. This is existing technology and will not be described in detail here.

[0062] The 3D point cloud data and the virtual component model are registered using the Iterative Closest Point (ICP) algorithm to generate the registered point cloud alignment state.

[0063] In the alignment state of the registered point cloud, the coordinate differences between the center coordinates of the actual component interface and the center coordinates of the virtual component model interface in the X, Y, and Z axes are calculated, and the three-dimensional Euclidean distance is calculated based on the coordinate differences as the hole alignment error value.

[0064] Extract the point cloud of the interface surface of the actual component and the point cloud of the interface surface of the virtual component model. Calculate the maximum Euclidean distance between corresponding points in the interface area of ​​the two sets of point clouds as the interface gap value. When the hole alignment error value is greater than the hole allowable tolerance value, or the interface gap value is greater than the interface allowable tolerance value, it is judged as geometric mismatch.

[0065] The allowable tolerance value is determined through engineering tests based on the installation accuracy level of the component. The specific process includes: preparing test component samples with standard interfaces for different accuracy levels such as high-precision installation and ordinary installation; conducting a series of installation tests in a simulated construction environment, gradually adjusting the installation deviation and recording the structural connection performance and sealing effect after each installation; and determining the maximum allowable deviation value that can ensure structural safety and functional integrity as the tolerance threshold under that accuracy level.

[0066] S5. Based on the geometric mismatch determination results, risk environment markings, and construction logic rules, the installation sequence of the components to be installed is adjusted through a dynamic decision tree model.

[0067] The degree of geometric mismatch, environmental risk status, and the properties of the components themselves all affect the efficiency and safety of the installation operation. For example, if severely mismatched components are installed first, subsequent components may not be compatible. The safety of hoisting heavy components in high-risk environments is low. Furthermore, there are installation dependencies between adjacent components in the construction logic. If they are installed in a fixed order, it will be impossible to cope with the dynamic changes in complex construction scenarios.

[0068] Therefore, it is necessary to integrate multi-dimensional influencing factors and output the component installation priority through a dynamic decision tree model. In a specific embodiment, the specific content of S5 is: calculate the first difference between the hole alignment error value and the hole allowable tolerance value. If the first difference is greater than zero, it indicates that there is a risk of hole alignment installation interference. Then, the first difference is used as the hole mismatch amount. Otherwise, the hole mismatch amount is zero.

[0069] Calculate the second difference between the interface gap value and the allowable tolerance value of the interface. If the second difference is greater than zero, it indicates that there is a risk of interface sealing or connection strength failure. Then, the second difference is used as the interface mismatch amount. Otherwise, the interface mismatch amount is zero. Take the maximum value of the two mismatch amounts as the geometric mismatch severity.

[0070] The number of times a risk environment marker is triggered within one hour is taken as the risk frequency value. The maximum value of the single trigger duration of a risk environment marker, which is the time interval from marker generation to marker removal, is taken as the risk duration value.

[0071] The weight values ​​of each component to be installed are retrieved from the component attribute database. The current position coordinates of the hoisting equipment are obtained through the hoisting equipment positioning system. The horizontal distance to the component installation point is calculated. The working radius is determined according to the rated working range of the hoisting equipment and its absolute difference is calculated. The number of adjacent components that must be installed first but have not yet been installed is also calculated.

[0072] The above-mentioned geometric mismatch severity, risk frequency value, risk duration value, weight value, absolute difference value and number of not installed components are normalized and used as input feature vectors. The priority score of each component is output through the dynamic decision tree model, and the component order is rearranged from largest to smallest according to the priority score to generate the adjusted installation order.

[0073] When the system is deployed for the first time, the model needs to be initialized and trained. This process involves collecting successful installation records from historical construction projects, extracting the input features and the corresponding actual installation sequence as training samples, and splitting nodes by maximizing the information gain ratio to construct the initial decision tree structure and node thresholds.

[0074] The method for obtaining the information gain ratio is as follows: for each input feature in the training samples, the information gain and split information are calculated respectively. The information gain is obtained by calculating the reduction of the uncertainty of the actual installation order label by the feature, which is measured by the entropy value. The split information is calculated based on the value distribution of the feature to correct the bias of the information gain. The information gain of each feature is divided by its corresponding split information to obtain the information gain ratio of the feature.

[0075] Furthermore, the dynamic decision tree model also includes an update step: extracting the geometric mismatch severity, risk frequency value, risk duration value, component weight value, difference between the hoisting equipment's operating radius and the horizontal distance between the installation point, the number of adjacent components not installed, and the corresponding actual installation sequence records from recent historical installation records.

[0076] A sliding window mechanism is used to filter historical data to obtain effective updated data. Based on this data, the information gain ratio of each feature parameter is calculated. The feature with the largest information gain ratio is selected as the splitting feature of the current node. In the numerical distribution of this feature, the splitting point that makes the child node purest is found. The value of this splitting point is set as the condition threshold of the current node after update. The above process is recursively executed until the preset decision tree depth limit is reached, so as to finally obtain the condition threshold of each internal node in the updated dynamic decision tree model.

[0077] S6. After the component installation is completed, when the deviation between the actual installation state and the BIM benchmark model exceeds the historical average installation deviation value and the risk environment marker is triggered during the installation process, the time series prediction model and the dynamic decision tree model are incrementally updated.

[0078] In the construction environment, changes in seasons, weather, and engineering locations can alter environmental patterns and component characteristics, gradually reducing the accuracy of model predictions and decisions. However, the actual deviation data after component installation contains the latest environmental-deformation-installation correlation features. If such effective data can be integrated into the model, continuous model optimization can be achieved.

[0079] Therefore, it is necessary to set clear update trigger conditions, and to perform incremental updates on the two core models when the conditions are met. In one specific embodiment, the method for obtaining the deviation data is as follows: after the component to be verified is installed and fixed by the fixing device, the laser scanning device is started to accurately scan the installation control points and obtain the actual three-dimensional coordinates of each control point as three-dimensional point cloud data.

[0080] The reference coordinates of each installation control point are extracted from the BIM reference model. The actual coordinates are then registered with the reference coordinates in three-dimensional point cloud to ensure that the registration accuracy meets the error requirements and generate the registered installation status.

[0081] Calculate the three-dimensional Euclidean distance between the actual coordinates of each installation control point and the BIM reference coordinates in the installation state after registration, and take the maximum value among all distances as the installation deviation value of the component.

[0082] Query the system log of the component installation process to determine whether the risk environment marker is triggered. If the risk environment marker is triggered during the component installation process and the installation deviation value is greater than the average installation deviation value of the component in the historical risk-free installation records, it indicates that the current installation deviation is very likely caused by a specific environmental risk factor that has been identified. The deviation data contains a strong correlation between environmental risk and component deformation.

[0083] It should be noted that in the case of initial construction or lack of historical data, the historical average installation deviation value adopts the allowable installation deviation value specified in the design specifications, and is gradually replaced with the actual statistical value as construction progresses.

[0084] If 50 components are installed consecutively without triggering an incremental update, a full model verification and update will be automatically initiated.

[0085] Collect temperature and humidity data of the current construction environment, material property parameters of the component, and BIM design dimension data, and use the installation deviation value as the actual deformation to form deviation data.

[0086] Furthermore, the specific method for incrementally updating the time series prediction model and the dynamic decision tree model is as follows: set the preset batch of deviation data as K, where K is a positive integer. By analyzing the relationship curve between model performance and training sample size in historical construction data, select the sample size corresponding to the inflection point where the performance improvement significantly tends to level off as the basis for determination, which is usually 10-30.

[0087] When the number of collected deviation data reaches K, these K deviation data are added to the original training sample set to form an updated training sample set.

[0088] The parameters of the historical time series prediction model are loaded as initial parameters. Multiple rounds of training iterations are performed on the updated training sample set. After each round of training, the model accuracy is verified through the validation set. Training is stopped when the accuracy no longer improves, and the updated time series prediction model is obtained.

[0089] Extract newly added records of geometric mismatch severity, risk frequency, risk duration, component weight, horizontal distance difference, number of missing components, and corresponding actual installation sequence from historical installation records to form an updated feature sample set.

[0090] Based on the updated feature sample set, the information gain ratio of each feature parameter is calculated, and the feature with the largest information gain ratio is selected as the splitting feature of the decision tree node.

[0091] The split point location of the split feature is determined by the binary split method to ensure that the sample purity of the two child nodes after splitting is maximized. At the same time, the upper limit of the decision tree depth is set to N layers to avoid model overfitting.

[0092] The depth parameter N can be determined using cross-validation: the training samples are divided into multiple parts, and the model performance is trained and validated at different depths, such as 3-10 layers. The depth that achieves the optimal balance between prediction accuracy and model complexity on the validation set is selected as the value of N. During system initialization, if there is no historical data for validation, N=6 can be temporarily set as an empirical value, and dynamically optimized based on new data in subsequent incremental updates.

[0093] Based on the new splitting features and splitting point locations, the node structure of the dynamic decision tree model is reconstructed to obtain the updated dynamic decision tree model. If the performance metrics of the updated model on the validation set decrease by more than 10%, the historical model version is retained, and a model update anomaly alert is issued.

[0094] The updated two models replace the original models and are used for subsequent component deformation prediction and installation sequence adjustment.

[0095] Example 2

[0096] Please see Figure 3 As shown, the present invention provides an intelligent matching system based on artificial intelligence, including: an environmental risk prediction module, a model environmental compensation module, a geometric matching decision module, and a model incremental update module. The connection relationship between the modules is as follows: the environmental risk prediction module and the model environmental compensation module are connected, the model environmental compensation module and the geometric matching decision module are connected, and the geometric matching decision module and the model incremental update module are connected.

[0097] The environmental risk prediction module is used to collect temperature, humidity and component location data in the construction environment in real time, and trigger risk environment markers based on the dynamic change characteristics of the data. The temperature and humidity data, component material properties and BIM design dimensions corresponding to the risk environment markers are input into the pre-trained time series prediction model, and the predicted deformation of the components is output.

[0098] The model environment compensation module is used to correct the interface geometric parameters of the components to be matched in the BIM model based on the deformation prediction, and generate a virtual component model after environment compensation.

[0099] The geometric matching decision module is used to perform geometric matching analysis based on the virtual component model using a 3D point cloud registration algorithm. When the hole alignment error or interface gap value between the actual component and the virtual model exceeds the corresponding tolerance value, it is judged as geometric mismatch. Based on the geometric mismatch judgment result, risk environment markers and construction logic rules, the installation sequence of the components to be installed is adjusted through a dynamic decision tree model.

[0100] The incremental update module is used to incrementally update the time-series prediction model and the dynamic decision tree model when the deviation between the actual installation state and the BIM benchmark model exceeds the historical average installation deviation value and the risk environment marker is triggered during the installation process after the component installation is completed.

[0101] In summary, this invention identifies and marks risky environments by real-time collection of temperature, humidity, and component location data in the construction environment; it inputs the risky environment data, component material properties, and BIM design dimensions into a time-series prediction model to predict component deformation; it corrects the geometric parameters of the BIM model interface based on the predicted deformation to generate a virtual component model after environmental compensation; it uses a 3D point cloud registration algorithm to perform geometric matching analysis to determine the matching status between the actual component and the virtual model; and when geometric mismatch occurs, it intelligently adjusts the component installation sequence through a dynamic decision tree model, combining risky environment markings and construction logic rules.

[0102] After installation, when the actual installation deviation exceeds the historical average and a risky environment is triggered, the prediction model and decision tree model are incrementally updated to achieve system self-learning optimization. This method effectively solves the problem of component deformation and geometric mismatch caused by environmental factors, improving the accuracy and efficiency of building construction.

[0103] The above embodiments can be implemented, in whole or in part, by software, hardware, firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented, in whole or in part, in the form of a computer program product.

[0104] Those skilled in the art will recognize that the algorithmic steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this application.

[0105] In addition, the functional modules in the various embodiments of this application can be integrated into one processing module, or each module can exist physically separately, or two or more modules can be integrated into one module.

[0106] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

[0107] Finally, the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. An intelligent matching method based on artificial intelligence, characterized in that, include: S1. Real-time collection of temperature, humidity and component location data in the construction environment, and triggering of risk environment markers based on the dynamic change characteristics of the data; S2. Input the temperature and humidity data, component material properties and BIM design dimensions corresponding to the risk environment markers into the pre-trained time series prediction model, and output the deformation prediction of the components. S3. Correct the interface geometric parameters of the components to be matched in the BIM model based on the deformation prediction, and generate a virtual component model after environmental compensation. S4. Based on the virtual component model, a geometric matching analysis is performed using a 3D point cloud registration algorithm. When the hole alignment error or interface gap value between the actual component and the virtual model exceeds the corresponding tolerance value, it is determined to be a geometric mismatch. The specific content of S4 is as follows: acquiring the three-dimensional point cloud data of the actual component to be matched through on-site laser scanning equipment; performing three-dimensional point cloud registration with the virtual component model to generate the registered point cloud alignment state; In the alignment state of the registered point cloud, the coordinate differences between the center coordinates of the actual component interface and the center coordinates of the virtual component model interface in the X, Y, and Z axes are calculated, and the three-dimensional Euclidean distance is calculated based on the coordinate differences as the hole alignment error value. Extract the point cloud of the interface surface of the actual component and the point cloud of the interface surface of the virtual component model, and calculate the maximum Euclidean distance between corresponding points in the interface area of ​​the two sets of point clouds as the interface gap value. When the hole alignment error value is greater than the hole allowable tolerance value, or the interface gap value is greater than the interface allowable tolerance value, it is judged as geometric mismatch; S5. Based on the geometric mismatch determination results, risk environment markings, and construction logic rules, the installation sequence of the components to be installed is adjusted through a dynamic decision tree model. The specific content of S5 is as follows: calculate the first difference between the hole alignment error value and the hole allowable tolerance value. If the first difference is greater than zero, the first difference is used as the hole mismatch amount; otherwise, the hole mismatch amount is zero. Calculate the second difference between the interface gap value and the allowable tolerance value of the interface. If the second difference is greater than zero, the second difference is used as the interface mismatch amount; otherwise, the interface mismatch amount is zero. Take the maximum value of the two mismatch amounts as the geometric mismatch severity. The number of times a risk environment marker is triggered within a preset time window is used as the risk frequency value. The maximum value of the single trigger duration of the risk environment marker is taken as the risk duration value; Obtain the weight of each component to be installed, the absolute difference between the horizontal distance from the current position of the hoisting equipment to the installation point and the preset working radius, and the number of adjacent components that must be installed first but have not yet been installed; use the above geometric mismatch severity, risk frequency value, risk duration value, weight value, absolute difference, and number of components not yet installed as input feature vectors, and output the priority score of each component through a dynamic decision tree model; rearrange the component order according to the priority score from largest to smallest to generate the adjusted installation order; S6. After the component installation is completed, when the deviation between the actual installation state and the BIM benchmark model exceeds the historical average installation deviation value and the risk environment marker is triggered during the installation process, the time series prediction model and the dynamic decision tree model are incrementally updated.

2. The intelligent matching method based on artificial intelligence according to claim 1, wherein, The specific method for triggering the risk environment marker is as follows: Based on the time sequence of the temperature, humidity and component location data, in a preset sampling period, when the value of the temperature and humidity data at a certain sampling moment shows a local extreme value relative to the values ​​at the adjacent sampling moments before and after it, or when the direction of movement of the component's horizontal projection position in two adjacent sampling periods is reversed, a risk environment marker is triggered.

3. The intelligent matching method based on artificial intelligence according to claim 1, wherein, The training method for the time series prediction model is as follows: Pre-collect historical construction temperature and humidity time series data, corresponding component material properties, BIM design dimensions and corresponding actual deformation as training sample set; The time-series prediction model is trained using a training sample set. The inputs are historical temperature and humidity data during construction, component material properties, and BIM design dimensions. The actual deformation is used as the label, and the output is the displacement component of the component in three-dimensional space as the deformation prediction.

4. The intelligent matching method based on artificial intelligence according to claim 1, wherein, The method for obtaining the virtual component model is as follows: Extract the original coordinates of the interface center and the original diameter of the hole of the component to be matched from the BIM reference model; Extract the deformation prediction quantity, which includes the displacement components of the interface center in the X-axis, Y-axis and Z-axis directions in the three-dimensional spatial coordinate system; Project the displacement components of the hole boundary points onto the direction of the hole's central axis and calculate the change in the hole's diameter. Adjust the X-axis value of the interface center coordinate of the component to be matched to the algebraic sum of the original X-axis value of the interface center coordinate and the X-axis displacement component; adjust the Y-axis and Z-axis values ​​in the same way. The hole diameter is adjusted to be the algebraic sum of the original hole diameter and the change in hole diameter; The BIM model after the above adjustments is used as the virtual component model after environmental compensation.

5. The intelligent matching method based on artificial intelligence according to claim 1, wherein, The dynamic decision tree model also includes: Extract the geometric mismatch severity, risk frequency, risk duration, component weight, difference between the hoisting equipment's operating radius and the horizontal distance between the installation point, the number of adjacent components not installed, and the corresponding actual installation sequence from recent historical installation records, and update the condition thresholds of each node in the dynamic decision tree model.

6. The intelligent matching method based on artificial intelligence according to claim 1, wherein, When the actual installation status deviates from the BIM baseline model beyond the historical average installation deviation value and risk environment markers are triggered during installation, including: After the component to be verified is installed, the three-dimensional point cloud data of the component, which is fixed to the building structure, is obtained by on-site laser scanning equipment; Register the 3D point cloud data with the BIM benchmark model to generate the registered installation status. Extract the preset set of installation control point coordinates from the BIM reference model, calculate the three-dimensional Euclidean distance between the actual coordinates of each installation control point and the BIM reference coordinates in the registered installation state, and take the maximum value as the installation deviation value. When a risk environment marker is triggered during the installation of a component, and the installation deviation value is greater than the average installation deviation value of the component in the historical risk-free installation records, the temperature and humidity data of the current construction environment, the material property parameters of the component, and the BIM design dimension data are collected. The installation deviation value is then used as the actual deformation amount to form deviation data.

7. The intelligent matching method based on artificial intelligence according to claim 6, wherein, The specific method for incrementally updating the time series prediction model and the dynamic decision tree model is as follows: When the number of biased data reaches the preset batch, the biased data is added to the training sample set to form an updated training sample set. Using the parameters of the historical time series prediction model as initial parameters, multiple rounds of training iterations are performed on the updated training sample set to obtain the updated time series prediction model; The newly added installation record data is extracted and collected, including geometric mismatch severity, risk frequency value, risk duration value, component weight value, horizontal distance difference, number of missing components and corresponding actual installation sequence records, to form an updated feature sample set; Based on the updated feature sample set, the splitting features and splitting point positions of each node in the dynamic decision tree model are recalculated using the information gain ratio, and the depth of the decision tree is limited to no more than N layers to obtain the updated dynamic decision tree model.

8. An intelligent matching system based on artificial intelligence, configured to perform the steps of an intelligent matching method based on artificial intelligence according to any one of claims 1-7, characterized in that, Includes the following modules: The environmental risk prediction module is used to collect temperature, humidity and component location data in the construction environment in real time, and trigger risk environment markers based on the dynamic change characteristics of the data. The temperature and humidity data, component material properties and BIM design dimensions corresponding to the risk environment markers are input into the pre-trained time series prediction model, and the predicted deformation of the component is output. The model environment compensation module is used to correct the interface geometric parameters of the components to be matched in the BIM model based on the deformation prediction, and generate a virtual component model after environment compensation. The geometric matching decision module is used to perform geometric matching analysis based on the virtual component model using a 3D point cloud registration algorithm. When the hole alignment error or interface gap value between the actual component and the virtual model exceeds the corresponding tolerance value, it is judged as geometric mismatch. Based on the geometric mismatch judgment result, risk environment markers and construction logic rules, the installation sequence of the components to be installed is adjusted through a dynamic decision tree model. The incremental update module is used to incrementally update the time-series prediction model and the dynamic decision tree model when the deviation between the actual installation state and the BIM benchmark model exceeds the historical average installation deviation value and the risk environment marker is triggered during the installation process after the component installation is completed.

Images