CN-121981345-A - Inversion prediction method, device and equipment based on BIM and construction site state

CN121981345ACN 121981345 ACN121981345 ACN 121981345ACN-121981345-A

Abstract

The application provides an inversion prediction method, device and equipment based on BIM and a construction site state, and relates to the technical field of construction state prediction management. The method comprises the steps of constructing a BIM semantic graph and a construction site semantic graph, performing cross-modal matching fusion and association mapping to obtain a calculated component-level multi-modal fusion state vector and a deviation vector, then performing construction state condition judgment, dividing the construction state of a component into three conditions of consistency, single-side incomplete and double-side conflict, respectively executing forward correction, bidirectional complementation or deviation decomposition reconstruction inversion prediction aiming at different conditions, and integrating inversion prediction results to generate a control instruction to realize closed-loop regulation. The application can effectively solve the problem of unstable state identification caused by on-site shielding, model updating lag and virtual-real collision, and remarkably improves the robustness, precision and result interpretability of state monitoring in a complex construction environment through the differential inversion path driven by conditions and the self-adaptive trust optimization.

Inventors

XIAO CHENG
WU ZHIMING
LIAN YUXIN
LI XIUFANG
YANG RONGHUA
HUANG MINGQIANG
KANG RUNCHI

Assignees

厦门理工学院
厦门孪数信息科技有限公司

Dates

Publication Date: 20260505
Application Date: 20260409

Claims (10)

1. The inversion prediction method based on BIM and the construction site state is characterized by comprising the following steps: s1, acquiring a BIM semantic graph and a construction site semantic graph; s2, performing cross-modal matching fusion and association mapping on the component nodes in the BIM semantic graph and the construction site semantic graph to obtain a component-level multi-modal fusion state vector and a deviation vector describing the deviation degree of the actual construction site state and the BIM design state of the component; S3, judging construction state conditions based on the component-level multi-mode fusion state vector and the deviation vector, dividing the construction states of the components into different component construction state categories, and carrying out inversion prediction of the different component construction state categories to obtain an inversion prediction result; And S4, generating a control instruction based on the deviation between the inversion prediction result and a preset construction plan, and feeding back to a construction management system and a building information model system for early warning prompt and adjustment.
2. The inversion prediction method based on the BIM and the construction site state according to claim 1, wherein the BIM semantic graph and the construction site semantic graph are expressed in a form of four groups comprising a node set, an edge set, a node feature matrix and an edge attribute matrix; Each node in the node set corresponds to one construction member entity, the edge set represents a space adjacent relation or a construction logic dependency relation between members, the node characteristic matrix comprises multi-mode characteristic attributes of the members, and the edge attribute matrix is used for describing connection types between the members; Analyzing component entities and interrelationships thereof in the construction site image into structural data through a visual language model to form a construction site node set and a construction site node characteristic matrix, and acquiring a component subset consistent with a construction site view from the BIM by combining virtual view constraints to establish the BIM semantic node set and the BIM node characteristic matrix.
3. The inversion prediction method based on the BIM and the construction site state according to claim 2, wherein the cross-modal matching fusion and association mapping process specifically includes: Defining a matching indicating variable, and setting the value of the matching indicating variable of the nodes of the two types of semantic graphs to be 1 when the nodes in the BIM semantic graphs are mapped with the nodes in the construction site semantic graphs in a one-to-one correspondence manner; When the nodes in the two types of semantic graphs cannot establish a mapping relation, marking the current node as a single-side information incomplete node, and setting the value of a matching indication variable of the current node as 0; Then, the node characteristic vectors with the matching indication variable value of 1 are fused through a characteristic fusion function to obtain a component-level multi-mode fusion state vector, wherein the expression is as follows: ; ; Wherein, the Is a component A multi-modal fusion state vector of (2); Is a component The cross-modal collaborative gating vector is used for adaptively adjusting the contribution degree of two types of modal information in the characteristic dimension; representing an element-level product operator, i.e., a hadamard product; transforming operators for nonlinear features of BIM mode, i.e. pairs Performing nonlinear transformation; Is a component Is a bit of the BIM node feature vector, For BIM semantic graphs BIM node feature matrix of (a); transforming operators for non-linear features of field modes, i.e. pairs Performing nonlinear transformation; Is a component Is characterized by comprising a construction site node characteristic vector, For construction site semantic map Is a construction site node characteristic matrix; Is a component Is defined by a set of contiguous nodes; Is a component And an adjacent member Attention weight in between; is an activation function; A trainable weight matrix for the gating network; vector splicing operators; The deviation vector is formed by splicing node characteristic deviation and local topology characteristic deviation, wherein the node characteristic deviation is obtained by calculating the characteristic deviation of a component node, and then the local topology characteristic deviation is obtained by introducing a graph aggregation mechanism to extract the state difference of the adjacent edges of the component node, and the expression is as follows: ; ; ; Wherein, the Is a component Node characteristic deviations of (a); Is a component Is defined by a local topological feature bias; Is a component And (3) with The edge importance coefficient between; Is a weight matrix which can be learned; Is a component And (3) with A construction site edge feature vector between the two; Is a component And (3) with BIM edge feature vectors between; Is a component Is a deviation vector of (2); for a node with a matching indication variable value of 0, initializing the node feature vector at the missing side as a predefined occupation feature vector so as to perform feature fusion.
4. The inversion prediction method based on the BIM and the construction site state according to claim 3, wherein the construction state categories of the different components comprise a multi-mode consistent state, a single-side information incomplete state and a double-side information conflict state, wherein the single-side information incomplete state comprises a construction site information incomplete state and a BIM semantic information incomplete state; the construction state condition judging process comprises the following steps: The judging conditions of the multi-mode consistent state are as follows: And is also provided with Wherein, the method comprises the steps of, Is a component Is a deviation vector of (2); Is a component A match indicating variable of (2); Represents an L2 norm; a preset consistency threshold value; the judging conditions of the incomplete state of the construction site information are as follows: the missing sources of the component nodes belong to a construction site semantic graph; The judging conditions of the incomplete state of BIM semantic information are as follows: The missing source of the component nodes belongs to the BIM semantic graph; The judging conditions of the two-side information conflict state are as follows: And is also provided with 。
5. The inversion prediction method based on BIM and construction site states according to claim 4, wherein the process of performing inversion prediction of different construction state categories of components is specifically as follows: If the state type is a multi-mode consistent state, performing forward correction inversion, namely taking a multi-mode fusion state vector of a current component node as an initial state, fusing the states of the component and the states of adjacent components through a graph structural feature propagation mechanism in the deduction process of each time step to generate a predicted state, and adjusting the proportion between a model predicted value and an actual observed value through setting an observation fusion weight coefficient to obtain a corrected predicted state vector, thereby obtaining iterative predicted results of the component states of T time steps in the future, namely a state predicted sequence; If the state type is the incomplete state of the construction site information, performing forward deduction state completion based on BIM, namely taking the BIM design state of the component as an initial state inversion starting point, carrying out neighborhood propagation deduction on the state of the component which is not observed based on the adjacency relation of the construction site semantic graph, generating a candidate site state, introducing cross-modal projection loss, mapping the candidate site state deduced by BIM to a site observation space, combining the actual observation of the adjacent component as a consistency constraint, and obtaining a forward deduction prediction state of the construction site, namely a final inversion state of the component obtained by forward deduction by minimizing the projection loss; If the state type is the incomplete state of BIM semantic information, performing reverse push-back model correction based on construction site observation, namely taking the construction site observation state of a component as an initial state inversion starting point, reversely reasoning based on a graph structure and a construction rule, updating the component state, introducing design consistency constraint, mapping the candidate state of site deduction to a BIM modal space, matching and aligning with design information to obtain alignment loss, and obtaining a BIM semantic reverse push-back prediction state, namely a final inversion state of the component obtained by reverse push-back by minimizing the alignment loss; if the state type is a double-side information conflict state, performing deviation decomposition and reconstruction inversion, namely splitting the deviation vector into a geometric deviation component, a semantic deviation component and a topological deviation component by utilizing a projection matrix, calculating comprehensive anomaly degree based on a statistical distribution model constructed by each component to identify a dominant deviation type, constructing a multi-objective game optimization model perceived by modal uncertainty based on the dominant deviation type, dynamically balancing each weight coefficient, and solving to obtain a final reconstruction state vector of a component meeting global consistency.
6. The inversion prediction method based on BIM and job site states according to claim 5, wherein the expression of the corrected prediction state vector in the multi-modal consistency state is: ; ; ; Wherein, the Is a component A state vector at time t; Is a component A multi-modal fusion state vector of (2); Is a component Forward predicted state vector at the next moment; Maintaining a weight matrix for the state of the component itself; Is a component Is defined by a set of contiguous nodes; Is a component And (3) with The edge importance coefficient between; Is the first An adjacency propagation weight matrix corresponding to the class relation; Is a component At the moment of time State vectors of (2); the bias vector is used for improving the fitting capacity of the model; Is a component At the moment of time The corrected predicted state vector; For the adjacency propagation weight matrix, modeling the influence of the adjacency component on the current component state; the fusion weight coefficient is observed; at the moment of the construction site Opposite component Is defined by the node feature vector of (a); the expression of the state prediction sequence is as follows: ; ; Wherein, the Is a component In the future Corrected state vectors for each time step; the total number of the predicted time steps; Is a component In the future Corrected state vectors for each time step; To be in the future Time-step construction site opposite component Is defined by the node feature vector of (a); Is a component A state prediction sequence at T time steps in the future; Is a component In the future Corrected state vectors for each time step; Is the step length of the time step; The expression of the final inversion state of the component obtained by forward deduction under the incomplete state of the construction site information is as follows: initial state inversion starting point: ; Candidate state: ; cross-modal projection loss: ; Minimizing projection losses: ; Wherein, the Is a component Is used for inverting the initial state vector; Is the current moment; Is a component BIM node feature vectors of (a); 、 Respectively as components In the future 、 Predictive state vectors at each time step; Is a component In the future Predictive state vectors at each time step; is the projection loss; representing a cross-modal projection function of the BIM and the construction site; is a spatial relationship transfer function; Is a component Is used for constructing the node feature vector of the site; Is a component And (3) with Edge feature vectors between; is the L2 norm; Component derived for forward deduction Is a final inversion state of (a); Is a minimization operator; the expression of the final inversion state of the component obtained by reverse pushback under the incomplete state of BIM semantic information is as follows: initial state inversion starting point: ; Candidate state: ; Alignment loss: ; minimizing alignment loss: ; Wherein, the Is a component Is used for constructing the node feature vector of the site; loss of alignment; representing a cross-modal projection function of a construction site and a BIM; Is a component Is defined by the reference BIM node feature vector; Is a reference confidence weight; For designing topology-dependent functions in space for neighborhood-based building blocks Standard design attribute derivation component of (a) The due logic pose; Is a component BIM node feature vectors of (a); For members obtained by reverse pushing-back Is a final inversion state of (a); Representing the size of a set of contiguous nodes; The final reconstructed state vector of the component in the double-sided information conflict state has the following expression: and (3) deviation vector splitting: ; ; ; ; Constructing a statistical distribution model: ; calculating normalized anomaly degree: ; ; dominant deviation type determination: ; multi-objective game optimization model: ; ; Wherein, the Is a component Is a deviation vector of (2); Is a component Is used for the geometric deviation component of the (a); Is a component Node characteristic deviations of (a); Is a component Is a semantic deviation component of (1); Is a component Is a topology deviation component of (1); 、、 mapping parameters for subspaces; Is the first A class bias component; the type of the component of the deviation is indicated, Respectively representing geometric deviation, semantic deviation and topological deviation; Representing a normal distribution; Is the first A mean vector of the class bias components; Is the first Covariance matrix of class bias components; Is that An inverse matrix of (a); Is a component First, the Normalized anomaly degree of class deviation; Is a component Is the first of (2) A class bias component; Is a transposed matrix symbol; Is a component Is a comprehensive anomaly degree of (2); Is the first The weights of the class bias components; Is the dominant deviation type; is a maximization operator; Is a component Is used to reconstruct the final state vector; Reconstructing a state vector for a component to be optimally solved; 、 Respectively as components 、 Is a fusion state vector of (a); 、 the BIM mode and the construction site mode trust weights are respectively calculated; Dynamically adjusting the topological rigidity weight according to the topological deviation component; Is a scaling factor; information entropy of the visual model; Is a component BIM node feature vectors of (a); When (when) If it is determined that the construction execution deviates, the construction execution increases Lowering down When (1) If it is determined that the component quality or attribute is abnormal, the BIM and the site are both deviated in recognition, and the maintenance is performed 、、 When (1) If it is determined that the model expression or the connection relationship is abnormal, the reduction is made Increase in size Order-making Tending to 0.
7. The inversion prediction method based on BIM and a construction site state according to claim 6, wherein the process of generating the control command based on the deviation of the inversion prediction result from the preset construction plan is specifically as follows: for all the component nodes, selecting a corresponding inversion prediction result as a final state vector of the component nodes according to the construction state types of the component nodes to form a global construction state set, wherein the global construction state set is expressed as: ; Wherein, the Is a global construction state set; Is a component Is defined in the set of state vectors; Is a component Is a node of (a); Is a node set; And calculating a progress deviation index according to a state prediction sequence and a construction plan state of the component in a multi-mode consistent state to generate a dynamic construction progress adjustment strategy, wherein the dynamic construction progress adjustment strategy comprises the steps of adjusting construction sequence, optimizing resource allocation and correcting construction period estimation, and the expression of the progress deviation index is as follows: ; Wherein, the Is a component In the first place Progress deviation indexes of the time steps; 、 respectively represent components In the first place Predicted state vectors and construction plan state vectors corrected by the time steps; Representing norms of the vectors; For a component in a single-side information incomplete state, calculating an uncertainty evaluation index based on a final inversion state obtained by bidirectional inversion to trigger risk early warning prompt, wherein the expression of the uncertainty evaluation index is as follows: ; Wherein, the Is a component Is an uncertainty evaluation index of (1); Projection loss of the component in the incomplete state of the construction site information is achieved; Component derived for forward deduction Is a final inversion state of (a); the alignment loss of the component in the incomplete state of BIM semantic information is determined; For members obtained by reverse pushing-back Is a final inversion state of (a); For components in the two-sided information conflict state, based on dominant deviation type Generating a differential repair control strategy, wherein the differential repair control strategy is as follows: When (when) When the construction execution deviation is judged, the on-site construction adjustment is triggered, when When the component quality or attribute is determined to be abnormal, triggering quality inspection or material verification flow Judging that the model expression or the connection relation is abnormal, and triggering BIM update or connection relation reconstruction; and uniformly mapping the control strategies under the three states into a control instruction set.
8. The method of inversion prediction based on BIM and job site state according to claim 7, further comprising, after executing the control command, changing the job site state, recalculating the multi-modal fusion state vector, and performing a new round of iterative inversion prediction on the updated job site state.
9. An inversion prediction apparatus based on BIM and a construction site state, for implementing the inversion prediction method based on BIM and a construction site state according to any one of claims 1 to 8, comprising: The acquisition unit is used for acquiring the BIM semantic graph and the construction site semantic graph; The multi-mode fusion mapping unit is used for carrying out cross-mode matching fusion and association mapping on the component nodes in the BIM semantic graph and the construction site semantic graph to obtain component-level multi-mode fusion state vectors and deviation vectors for describing deviation degree of actual construction site states and BIM design states of the components; The inversion prediction unit is used for judging construction state conditions based on the component-level multi-mode fusion state vector and the deviation vector so as to divide the construction states of the components into different component construction state categories, and performing inversion prediction of the different component construction state categories to obtain an inversion prediction result; And the control instruction generation unit is used for generating a control instruction based on the deviation between the inversion prediction result and a preset construction plan, and feeding the control instruction back to the construction management system and the building information model system for early warning prompt and adjustment.
10. An inversion prediction apparatus based on BIM and a construction site state, comprising a processor and a memory, wherein the memory stores a computer program executable by the processor to implement an inversion prediction method based on BIM and a construction site state as claimed in any one of claims 1 to 8.

Description

Inversion prediction method, device and equipment based on BIM and construction site state Technical Field The invention relates to the technical field of construction state prediction management, in particular to an inversion prediction method, device and equipment based on BIM and construction site states. Background Along with the rapid development of the engineering construction field to the digitization and the intellectualization, the management of the construction process is gradually changed from the traditional experience drive to the data drive. Under the background, construction information expression modes based on a Building Information Model (BIM) are widely applied to engineering management, and structural description of construction objects is realized by organizing the geometric form, attribute information and constructional relations of building components, so that foundation support is provided for construction planning and process management. In the prior art, the multi-mode information fusion generally realizes the association between model information and site observation information through a component matching or state comparison peer-to-peer mode, so as to judge the construction state. The existing technical scheme mainly utilizes computer vision or a deep learning algorithm to analyze acquired data such as images, point clouds and the like, or uses BIM as a core to update the states of components through manual input and on-site sensors. However, these processes are often based on the assumption that the information is relatively complete or consistent, and have significant limitations in a practical complex construction environment. Firstly, a state identification method based on construction site perception has high dependence on the observation integrity, and partial components cannot be stably observed due to shielding, visual angle dead angles, temporary components and interference of operators in the construction site, so that an identification result is incomplete. Secondly, the BIM-based information management method lacks automatic and real-time state reasoning capability in updating, and when actual deviation or model lag occurs in a construction site, the system is difficult to adaptively correct. In the aspect of virtual-real information association, the conventional scheme generally adopts unified matching logic, and when the field observation is limited or BIM information lags to cause conflict between the two, the matching process is very easy to generate instability, and the state judgment result deviation is larger. In addition, the current multi-mode fusion method mostly adopts unified processing logic, and lacks the capability of carrying out conditional division and differential inversion according to the completeness, conflict degree or consistency degree of virtual and real information, so that the stability and the interpretability of a fusion result are insufficient under the condition of information deficiency or inconsistency, and component-level state reasoning and closed-loop control decision under a complex construction scene cannot be effectively supported. In view of this, the present application has been made. Disclosure of Invention The invention aims to provide an inversion prediction method, device and equipment based on BIM and construction site states, which are used for solving the problems of limited observation, lag in updating a building information model, failure in matching virtual and real information and insufficient robustness and interpretability in complex scenes in the conventional construction state prediction management technology in a multi-mode fusion process. In order to solve the technical problems, the invention is realized by the following technical scheme: An inversion prediction method based on BIM and construction site state comprises the following steps: s1, acquiring a BIM semantic graph and a construction site semantic graph; s2, performing cross-modal matching fusion and association mapping on the component nodes in the BIM semantic graph and the construction site semantic graph to obtain a component-level multi-modal fusion state vector and a deviation vector describing the deviation degree of the actual construction site state and the BIM design state of the component; S3, judging construction state conditions based on the component-level multi-mode fusion state vector and the deviation vector, dividing the construction states of the components into different component construction state categories, and carrying out inversion prediction of the different component construction state categories to obtain an inversion prediction result; And S4, generating a control instruction based on the deviation between the inversion prediction result and a preset construction plan, and feeding back to a construction management system and a building information model system for early warning prompt and adjustment. Preferably, the BI