Electromechanical construction method and system based on BIM technology

Abstract

The invention discloses an electromechanical construction method and system based on a BIM technology, and relates to the field of electromechanical construction, and the method comprises the steps: collecting a three-dimensional point cloud of a construction site; a construction site buult BIM model is constructed; constructing an electromechanical BIM model of the construction site; obtaining a logic dependency relationship between electromechanical tasks in the electromechanical BIM model by utilizing a judgment matrix method DSM, and extracting attribute parameters of task nodes; constructing a directed acyclic graph (DAG) for task collaborative optimization according to the logic dependency relationship between the electromechanical tasks and the time sequence and resource parameters of the nodes; obtaining task nodes on the critical path according to a directed acyclic graph (DAG) of task collaborative optimization; establishing a multi-objective optimization model according to task nodes on the critical path, and solving through a multi-objective particle swarm optimization algorithm MOPSO by taking the shortest construction period and the highest resource utilization rate as optimization objectives and taking process logic and resource occupation as constraint conditions to obtain a resource optimization configuration set; and the collaboration of electromechanical construction tasks is improved.

Description

Technical Field

[0001] The present application relates to the field of electromechanical construction, and in particular to an electromechanical construction method and system based on BIM technology. Background Art

[0002] With the increasing complexity and scale of construction projects, electromechanical installation engineering has become an important part of modern construction. Electromechanical engineering involves multiple professional systems such as HVAC, water supply and drainage, electrical, and fire protection. The project volume is large, the types of work are diverse, and the coordination is difficult, which puts higher requirements on construction organization and management. The traditional electromechanical construction model has many problems, such as insufficient professional coordination, unreasonable resource allocation, and lagging progress control, which makes it difficult to ensure the quality and efficiency of the project, affecting the overall performance of the construction project.

[0003] In order to adapt to the development needs of construction projects in the new era, it is urgent to innovate the management model and technical means of mechanical and electrical construction. In recent years, the development of emerging technologies such as building information modeling (BIM), virtual construction, and intelligent optimization has provided new ideas and methods for solving many problems in mechanical and electrical construction. Introducing BIM technology into mechanical and electrical engineering and realizing the visual design, refined construction, and intelligent management and control of mechanical and electrical systems by building an information-based and digital construction management platform has become a research hotspot and development trend in the industry. This has brought new opportunities and challenges to optimizing the organization of mechanical and electrical construction and improving the level of engineering management.

[0004] Mechanical and electrical installation engineering is an important professional division of construction engineering, involving multiple systems and types of work, such as HVAC, water supply and drainage networks, power and weak current systems, fire protection facilities, etc. These systems are intertwined and tightly coupled, which places high demands on the coordination and precision of construction. However, the traditional mechanical and electrical construction management model generally has the problem of poor task coordination. Summary of the invention

[0005] In response to the problem of poor coordination of electromechanical construction tasks in the prior art, the present application provides an electromechanical construction method and system based on BIM technology, which improves the coordination of electromechanical construction tasks through three-dimensional reconstruction of the construction site, collaborative analysis of electromechanical tasks and multi-objective optimization scheduling.

[0006] The purpose of this application is achieved through the following technical solutions.

[0007] The present application provides a mechanical and electrical construction method based on BIM technology, comprising: collecting a three-dimensional point cloud of a construction site;

[0008] The 3D point cloud is registered and integrated with the BIM model in the design phase to construct a built BIM model of the construction site;

[0009] According to the built BIM model of the construction site, the electromechanical BIM model of the construction site is constructed by integrating the electromechanical BIM model of the design stage; the judgment matrix method DSM is used to obtain the logical dependencies between electromechanical tasks in the electromechanical BIM model, and the attribute parameters of the task nodes are extracted; according to the logical dependencies between electromechanical tasks, the timing and resource parameters of the nodes, the directed acyclic graph DAG of task collaborative optimization is constructed; according to the directed acyclic graph DAG of task collaborative optimization, the task nodes on the critical path are obtained; wherein the critical path represents the path corresponding to the process with zero free time difference; according to the task nodes on the critical path, a multi-objective optimization model is established, with the shortest construction period and the highest resource utilization as the optimization goals, and the process logic and resource occupancy as the constraints, and the multi-objective particle swarm optimization algorithm MOPSO is used to solve it to obtain the resource optimization configuration set; according to the resource optimization configuration set, the electromechanical construction scheduling plan is generated.

[0010] Furthermore, a three-dimensional point cloud model of the construction site is constructed, including: scanning the construction site with a three-dimensional laser scanner to obtain a point cloud data set of the construction site; denoising the point cloud data set; spatially registering the denoised point cloud data set with an ICP registration algorithm; and based on the spatially registered point cloud data set.

[0011] Furthermore, a builtBIM model of the construction site is constructed, including: importing the three-dimensional point cloud of the construction site and the BIM model of the design stage into the BIM software platform respectively; in the BIM software platform, according to the geometric features of the three-dimensional point cloud, using the regional growing algorithm to extract the point cloud data of the building components including floor slabs, walls and columns; according to the extracted point cloud data of the building components and the geometric elements of the corresponding components in the BIM model of the design stage, the point cloud data are aligned, and the optimal transformation matrix from the point cloud data to the geometric elements is calculated by the least squares method to obtain the three-dimensional point cloud after local alignment and the BIM model of the design stage; according to the three-dimensional point cloud after local alignment and the BIM model of the design stage, the ICP alignment algorithm is used to minimize the global error through iterative calculation.

[0012] Preferably, the geometric elements of the 3D point cloud P after local registration and the BIM model Q in the design stage are converted into point set forms, which are represented as P = {p1, p2, ..., p n} and Q={q1,q2,......,q m}, where n and m are the number of points in the two point sets respectively; in the point sets P and Q, feature points with significant geometric features are extracted, such as pipeline connection points, equipment corner points, etc. Harris corner point detection, SIFT feature description and other algorithms are used to obtain a stable and representative feature point set P f ={pf1,pf2,......,pf k} and Qf ={qf1,qf2,......,qf k}, where k is the number of feature points.

[0013] Initialize the transformation matrix T as the unit matrix, set the iteration number threshold and error threshold; enter the iteration process: for the feature point set P f Each feature point pf in i , in the feature point set Q f Find the closest point qf i , construct corresponding point pairs (pf i ,qf i );According to the corresponding point pairs of feature points, the initial transformation matrix T is calculated by the least squares method f , so that the transformed feature point set P f To the feature point set Q f The distance error is the smallest; using the initial transformation matrix T f , transform the entire point set P to obtain a new point set P': P'={p1',p2',......,p n '}, where p i '=T f ×p i ; For each point p in the point set P' i ', find the nearest point q in the point set Q i , construct corresponding point pairs (p i ',q i ) According to all corresponding point pairs, the weightedICP algorithm is used to introduce the weight coefficient w, calculate the optimal transformation matrix T, and minimize the weighted error function: Among them, T is the 3×3 rigid body transformation matrix, w i is the weight coefficient of the ith point pair, which is set according to factors such as the geometric characteristics of the point pair, measurement accuracy, and outlier possibility. The point set P' is transformed using the calculated transformation matrix T to obtain a new point set P": P" = {p1", p2", ..., p n ”},p i ”=Tp i '; Calculate the weighted error between the transformed point set P" and the point set Q: If the number of iterations reaches the threshold or the error e is less than the set threshold, the iteration is stopped and the current transformation matrix T is output; otherwise, P" is assigned to P and the iteration is returned to continue. The optimal transformation matrix T obtained by iteration is used to transform the entire 3D point cloud after local registration to obtain a 3D point cloud that is globally registered with the BIM model in the design phase; the 3D point cloud after global registration is attached to the BIM model in the design phase, and the component geometry information in the BIM model is updated and corrected, especially the position and size information of electromechanical components such as pipelines and equipment, to obtain the built BIM model of the construction site. The built BIM model is checked for quality and integrity, errors, omissions, and collisions are identified, the attribute information of pipelines and equipment is extracted, and a complete built BIM model of electromechanical specialty is generated.

[0014] Furthermore, the electromechanical BIM model of the construction site is constructed, including: importing the built BIM model and the electromechanical BIM model in the design stage into the BIM software platform, and using a three-dimensional coordinate conversion method to unify the coordinates of the built BIM model and the electromechanical BIM model in the design stage; traversing the electromechanical BIM model in the design stage to obtain geometric parameters and non-geometric parameters; wherein the geometric parameters include the size, position and direction of the components; the non-geometric parameters include the type, material, specification and connection relationship of the components; according to the built BIM model, using a three-dimensional space search algorithm, taking the geometric parameters of the electromechanical components as query conditions, to obtain point cloud clusters whose spatial position coincidence with the electromechanical components is higher than a threshold; using the non-geometric parameters of the electromechanical components as semantic labels, matching the semantic labels with the obtained point cloud clusters, and determining the installation position and connection relationship of each electromechanical component at the construction site by calculating the semantic similarity and coordinate deviation; according to the matching results, updating the electromechanical BIM model in the design stage to obtain the electromechanical BIM model of the construction site.

[0015] Furthermore, a directed acyclic graph (DAG) for task collaborative optimization is constructed, including: extracting the construction tasks corresponding to each component according to the electromechanical BIM model of the construction site, obtaining the dependencies between the tasks according to the construction process and procedures, and establishing a dependency matrix between the electromechanical tasks; wherein, in the dependency matrix, when task i must start after task j is completed, the value of the position (i, j) in the dependency matrix is ​​set to 1, otherwise it is set to 0; extracting the attribute parameters and geometric parameters of each task node according to the electromechanical BIM model of the construction site; wherein the attribute parameters include the work content, workload and timing parameters of each task; the timing parameters include the duration, earliest start time and latest completion time of the task; the geometric parameters include the spatial position coordinates of the components corresponding to the task; taking the dependency matrix as the adjacency matrix and adopting the critical path method (CPM) to construct a directed weighted graph between the task nodes; establishing a directed acyclic graph (DAG) for task collaborative optimization according to the directed weighted graph and the attribute parameters and geometric parameters of each task node.

[0016] Furthermore, a directed acyclic graph DAG for task collaborative optimization is established, including: based on a directed weighted graph, using the critical path method CPM, calculating the earliest start time of each task node by forward recursion, calculating the latest start time of each task node by reverse recursion, and calculating the free time difference of the task node based on the earliest start time and the latest start time; obtaining critical tasks and non-critical tasks based on the earliest start time, the latest start time and the free time difference of the task node; wherein, a critical task indicates that the free time difference of the corresponding task node is 0, and a non-critical task indicates that the free time difference of the corresponding task node is greater than 0; extracting all edges between critical task nodes, and calculating the dependency strength of each edge; wherein, the dependency strength is measured by the tightness of the time constraints between the critical task nodes, and the degree of time constraints is represented by the ratio of the task duration to the difference between the earliest start time and the latest start time.

[0017] Furthermore, according to the directed acyclic graph DAG of task collaborative optimization, the task nodes on the critical path are obtained, including: according to the directed acyclic graph DAG of task collaborative optimization, the earliest start time, the latest start time and the free time difference of all nodes are extracted; all paths in the directed acyclic graph DAG of task collaborative optimization are traversed, and according to the free time difference of the nodes, the path with zero free time difference is obtained as the critical path; all nodes on the critical path are extracted to obtain a set of critical task nodes; and the total construction period of the critical path is calculated as the benchmark construction period for electromechanical construction optimization.

[0018] Furthermore, the multi-objective particle swarm optimization algorithm MOPSO is used to solve the problem and obtain the optimal resource configuration set, including: establishing the objective function and constraint conditions of multi-objective optimization according to the key task node set: Objective function: Among them, f1 represents the total duration of the project, that is, the sum of the duration of all key task nodes, and the goal is to minimize the total duration; f2 represents the project resource utilization, that is, the ratio of the actual amount of resources allocated to all key task nodes to the total amount of resources in the project, and the goal is to maximize the resource utilization; d i represents the duration of critical task node i, that is, the time required to complete task i; x i is the execution decision variable of key task node i. i =1 means executing task i, when x i =0 means that task i is not executed; n represents the total number of key task nodes in the project; r ik represents the demand of key task node i for resource type k, that is, the number of k-th resource required to complete task i; x ik is the allocation decision variable between key task node i and resource type k. ik =1 indicates that task i is assigned the kth resource. ik = 0 means that task i is not assigned the kth resource; R k It represents the total amount of the k-th resource in the project, that is, the available quantity of the k-th resource in the entire project; m represents the total number of resource types involved in the project, that is, there are m different types of resources in the project.

[0019] Constraints: Process logic constraints: Resource usage constraints: Start time constraint: End time constraint: Variable restrictions: in, Represents the priority relationship constraint between task nodes, that is, if task node i is the predecessor task of task node j, then task j can be executed only when task i is executed; Represents the resource occupancy constraint, that is, the total amount of resources occupied by all task nodes that allocate the k-th resource cannot exceed the total amount of the k-th resource R k ; Represents the start time constraint of the task node, that is, the actual start time s of task node i i No earlier than its earliest start time ES i , where ES i Calculated by CPM method; represents the end time constraint of the task node, that is, the actual end time e of task node i i No later than the latest end time LS i , where LS i Calculated by CPM method; Represents the task node execution decision variable x iIt is a 0-1 variable, that is, it can only take the value 0 or 1; Represents the task node and resource allocation decision variable x ik It is a 0-1 variable, that is, it can only be 0 or 1. Initialize the parameters of the MOPSO algorithm, including the number of particles N, the maximum number of iterations T max , learning factors c1 and c2, inertia weight w; randomly generate the initial particle swarm Pop(0), and use the position vector of each particle as a candidate solution for resource optimization configuration; calculate the fitness of each particle according to the particle swarm, and update the particle's historical optimal position and global optimal position; according to the particle's current position, historical optimal position and global optimal position, use the speed update formula and position update formula to update the speed v of each particle i and position i Update; calculate the fitness of the updated particles, and update pbest and gbest according to the fitness; determine whether the maximum number of iterations T is reached max If yes, the particle corresponding to pbest is taken as the optimal solution of MOPSO and the resource optimization configuration set is output; otherwise, let t = t + 1 and return to continue iterative update.

[0020] Further, the fitness of each particle is calculated, and the historical optimal position and the global optimal position of the particle are updated, including: calculating the fitness itness of each particle, including: itness = w1(t)f1'+w2(t)f2'; Among them, itness represents the comprehensive fitness value of the particle, which is used to evaluate the quality of the particle. The larger the itness value, the better the quality of the particle. w1(t) and w2(t) represent the adaptive weight coefficients of the objective functions f1 and f2 at the tth iteration, respectively, which are used to balance the optimization weights of the two objective functions, satisfying w1(t)+w2(t)=1, and dynamically adjusted during the iteration process. f1 represents the value of the objective function 1 of the particle corresponding to the solution, that is, the total duration of the project. The calculation formula of f1 is: Among them, d i represents the duration of task node i, x i is the execution decision variable of task node i; f2 represents the value of objective function 2 of the particle corresponding solution, that is, the project resource utilization rate. The calculation formula of f2 is: Among them, r ik represents the demand of task node i for resource k, x ik is the allocation decision variable between task node i and resource k, R k represents the total amount of resource k; f 1max and f 1min They represent the maximum and minimum f1 values ​​of all particles in the current iteration, which are used to normalize f1; 2max and f2min They represent the maximum and minimum values ​​of f2 of all particles in the current iteration, which are used to normalize f2; f1' represents the value of the normalized objective function f1 of the particle corresponding to the solution. The calculation formula of f1' is: Through this formula, f1 is mapped to the interval [0, 1]. The larger the f1' is, the smaller f1 is, that is, the shorter the total project duration is; f2' represents the value of the normalized objective function f2 of the particle corresponding solution. The calculation formula of f2' is: Through this formula, f2 is mapped to the interval [0, 1]. The larger f2' is, the larger f2 is, that is, the higher the project resource utilization rate is.

[0021] Furthermore, the adaptive weight coefficient is calculated by the following formula: when t>0, w1(t)=w1(t-1)+Δw1(t); w2(t)=w2(t-1)+Δw2(t); where Δw1(t) and Δw2(t) are the changes in the weight coefficient at the tth iteration, and are adaptively adjusted according to the convergence of the particle swarm:

[0022] Among them, λ1 and λ2 are weight adjustment coefficients, which are used to control the amplitude of weight change. Their value range is (0, 1], and they are usually smaller values, such as 0.1 or 0.2; f1'(gbest) and f2'(gbest) are the normalized objective function values ​​corresponding to the current global optimal solution gbest, reflecting the quality of the known optimal solution; f1'(t) and f2'(t) are the normalized objective function means of all particles in the t-th iteration, reflecting the current overall optimization level of the particle swarm; f1'(0) and f2'(0) are the normalized objective function means of the initial particle swarm, reflecting the initial state of the optimization problem; t is the current iteration number, T max is the maximum number of iterations, It reflects the progress of the optimization process; k1 and k2 are convergence factors, which are used to adjust the relationship between weight change and iteration progress, and their value range is [0.5, 2].

[0023] Compared with the prior art, the advantages of this application are:

[0024] By performing 3D laser scanning on the construction site to obtain comprehensive and accurate point cloud data, and then using the ICP algorithm to align with the BIM model, an as-built model that reflects the actual progress of the construction can be efficiently constructed. The as-built model integrates the information model from the design stage and the real scene collection of the construction site. The full life cycle application of BIM can provide a reliable data foundation for the subsequent collaborative optimization of electromechanical tasks and improve the accuracy of optimization.

[0025] Using the DSM analysis method, the electromechanical tasks and their attribute parameters are extracted from the BIM model, and the logical dependency matrix between tasks is constructed. Through the network graph theory, the intricate topological structure and timing constraints between electromechanical tasks are revealed, laying a foundation for collaborative optimization. On this basis, the close relationship between key tasks and the spatial position of non-key tasks are further considered, and the DAG graph is optimized to maximize the parallelism between electromechanical tasks, reasonably adjust the task layout, and improve the electromechanical construction organization from the two dimensions of time and space.

[0026] In response to the common multi-objective optimization needs in electromechanical construction, this application constructs a mathematical model with construction cycle and resource utilization as optimization targets. The MOPSO algorithm is innovatively introduced for solving the problem. By updating the particle position and velocity, the Pareto frontier is searched under the process logic and resource occupancy constraints to obtain a resource allocation solution that takes into account both progress and cost. Compared with traditional heuristic algorithms, MOPSO has stronger global optimization capabilities and convergence speed, and can efficiently handle large-scale, strongly constrained electromechanical construction optimization problems.

[0027] Based on the critical path method, this application proposes a strategy to focus on analyzing key task nodes. By extracting the task nodes on the critical path and identifying the tasks that have the most critical impact on the construction period, it helps managers to quantify and analyze the progress risks of mechanical and electrical construction and carry out progress control in a focused manner. At the same time, for non-critical tasks, by optimizing their spatial layout and minimizing the spatial interference between tasks, the plane organization of mechanical and electrical construction can be improved and the efficiency of on-site operations can be improved.

[0028] In view of the problem that the objective weights are difficult to quantify in multi-objective optimization, this application designs an adaptive weight adjustment strategy. According to the convergence of the particle swarm during the MOPSO iteration process, the weights of the construction period and resource utilization objectives are dynamically adjusted to balance the optimization direction of the two objectives and avoid single-objective-dominated optimization. At the same time, the number of iterations factor is introduced to gradually reduce the amplitude of weight changes as the search proceeds, which can not only expand the early search range, but also fine-tune the guidance algorithm convergence in the later stage, improving the optimization efficiency and solution quality. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] Figure 1 is an exemplary flow chart of a mechanical and electrical construction method based on BIM technology according to the present application;

[0030] Figure 2 is an exemplary flow chart of a built BIM model of a civil construction site according to the present application;

[0031] Figure 3 is an exemplary flow chart for constructing a mechanical and electrical BIM model of a construction site according to the present application;

[0032] Figure 4 It is an exemplary flow chart of a directed acyclic graph DAG for collaborative optimization of construction tasks according to the present application. DETAILED DESCRIPTION

[0033] The method and system provided in the embodiments of the present application are described in detail below with reference to the accompanying drawings.

[0034] Figure 1 This is an exemplary flow chart of an electromechanical construction method based on BIM technology according to the present application, including: using Leica ScanStation P50 3D laser scanner to scan and collect electromechanical construction site of a large medical complex project. The scanner has the following technical indicators: scanning distance up to 1000 meters, ranging accuracy up to 1.2mm+10ppm, scanning speed up to 1,000,000 points / second, field of view of 360° (horizontal) × 290° (vertical), built-in HDR camera resolution of 20 million pixels. According to the spatial scale and complexity of the electromechanical construction site, this application has set up 15 scanning stations on site, covering different areas, floors, and angles to ensure complete collection of spatial information of the building electromechanical system. The scanning parameters of each station are set as follows: the resolution is 1 / 4, that is, one point is sampled every 4 meters; the point cloud quality is 4×, that is, each point is scanned 4 times to take the average value to reduce the impact of noise; the scanning field of view range is 360° (horizontal) × 290° (vertical) full-range scanning. At the same time, four target balls were placed at each site, and their spatial coordinates were obtained using a total station as control points for alignment.

[0035] This application uses a statistical filtering algorithm to denoise the original point cloud data. Taking the point cloud data of the first station as an example, the data set contains 105 million three-dimensional points, and there are a small number of discrete points or noise. The parameters of the denoising process are set as follows: the neighborhood search radius is 5cm, that is, with each point as the center, the neighboring points within a radius of 5cm are extracted to form a local point set; the mean of the local point set is used as the threshold to remove noise points that deviate from the mean by more than 1 standard deviation. After statistical filtering, the number of point cloud data at the first station was reduced from 105 million to 101 million, and about 4% of the noise points were removed. The point cloud surface is smoother, and abnormal points introduced by occlusion, movement and other factors are eliminated. The point cloud data of 15 stations were batch processed, and the average denoising rate was about 3.5%, and 15 denoised point cloud data sets were obtained, with a total data volume of about 38GB.

[0036] The ICP registration algorithm is used to perform spatial registration on the denoised point cloud dataset, including: using the nearest neighbor search algorithm to find the nearest corresponding point in the target point cloud for each point in the reference point cloud. Considering the complexity of the electromechanical construction environment, point cloud data may have uneven density, local missing points, etc. In order to improve the robustness of the matching, two constraints, distance and normal vector, are introduced. For each point in the reference point cloud, the nearest points within a radius of r are searched in the target point cloud, and the Euclidean distance d and the normal vector angle θ between them are calculated. When d is less than the distance threshold d t And θ is less than the angle threshold θ t When , the two points are considered to be matched successfully. By setting the threshold value (d in this application) reasonably t =5cm,θ t =30°), the wrong matching point pairs can be eliminated to improve the registration accuracy. According to the selected corresponding point pairs, the ICP algorithm iteratively optimizes the rigid body transformation between the point clouds, making the position of the registered point cloud more accurate.

[0037] This application uses the least squares method to solve the optimal rotation matrix R and translation vector t, so that the distance error between the point set P in the reference point cloud and the corresponding point set Q in the target point cloud is minimized after transformation. The objective function is: Among them, p i and q i is the corresponding point pair. For this least squares problem, the singular value decomposition (SVD) algorithm is used to solve it. Decentralize the point sets P and Q and construct the matrix Perform SVD decomposition on W and obtain the rotation matrix W = U∑V T . Then calculate the translation vector t=q-Rp based on R, where and are the centroids of P and Q respectively. By optimizing R and t, the coordinates of the reference point cloud are updated to gradually approach the target point cloud. For the multi-site scanning data of the electromechanical construction site, it is necessary to stitch them step by step. Taking the point cloud of the first site as the benchmark, ICP alignment is performed with the point clouds of other sites in turn. Through cascading, the point cloud data of all sites are unified to the coordinate system of the first site to achieve global consistency. In the process of stitching, the rich geometric features of the overlapping area are used to ensure the accuracy and robustness of the alignment. Finally, the point cloud data of multiple sites are seamlessly stitched together to obtain a complete three-dimensional point cloud model of the construction site.

[0038] In this embodiment, first, the coordinates of the four target balls in each point cloud data set are manually marked as reference points for initial registration. Then, the convergence conditions of the ICP algorithm are set: the maximum number of iterations is 100 times, and the root mean square error (RMSE) is less than 2mm. In each round of iteration, the ICP algorithm automatically searches for corresponding points in different data sets, uses singular value decomposition (SVD) to calculate the optimal rotation matrix and translation vector, and updates the point cloud coordinates until convergence. During the registration process, the point clouds of 15 stations are registered in sequence to form a unified point cloud coordinate system. After the point cloud data is registered, the overlapping areas of different stations are seamlessly spliced, eliminating the splicing gaps. After ICP optimization, the final registration RMSE is 1.8mm, which meets the convergence conditions and achieves high-precision global registration of point cloud data. First, the coordinates of the four target balls in each point cloud data set are manually marked as reference points for initial registration. Then, the convergence conditions of the ICP algorithm are set: the maximum number of iterations is 100 times, and the root mean square error (RMSE) is less than 2mm. In each round of iteration, the ICP algorithm automatically searches for corresponding points in different data sets, uses singular value decomposition (SVD) to calculate the optimal rotation matrix and translation vector, and updates the point cloud coordinates until convergence. During the registration process, the point clouds of 15 stations are registered in sequence to form a unified point cloud coordinate system. After the registration of the point cloud data, the overlapping areas of different stations are seamlessly spliced, eliminating the splicing gap. After ICP optimization, the final registration RMSE is 1.8mm, which meets the convergence conditions and achieves high-precision global registration of point cloud data. The point cloud data set after spatial registration is used as the collected three-dimensional point cloud.

[0039] Figure 2 This is an exemplary flow chart of building a builtBIM model of a construction site according to the present application, which aligns and fuses the three-dimensional point cloud with the BIM model in the design stage to build a builtBIM model of the construction site, including: converting the three-dimensional point cloud P after local alignment and the BIM model Q in the design stage into a unified point set form to facilitate subsequent data processing. The three-dimensional point cloud P is directly sampled to obtain the point set P = {p1, p2, ..., p n}, where p i is the three-dimensional point coordinate in the point cloud. The geometric elements in the BIM model Q, such as pipelines, equipment, etc., are converted into a point set Q = {q1, q2, ..., q m}, where q j is the sampling point coordinate of the BIM model surface. The point sets P and Q contain n and m 3D points respectively.

[0040] In the point sets P and Q, feature points with significant geometric features are extracted as the basis for registration. The Harris corner detection algorithm is used to automatically identify stable feature points such as pipeline connection points and equipment corners by calculating the curvature changes of the local surface of the point cloud. For the extracted feature points, the SIFT (Scale Invariant Feature Transform) algorithm is further used to calculate their feature descriptors to obtain the high-dimensional feature vectors of the feature points. Finally, two feature point sets P are obtained. f ={p f1 ,p f2 ,......,p fk} and Q f = {q f1 ,q f2 ,......,q fk}, each feature point contains its 3D coordinates and SIFT feature vector, k is the number of feature points. Using the extracted feature points, calculate the initial transformation matrix T between the point cloud P and the BIM model Q f , for the feature point set P f Each feature point p in fi , in the feature point set Q f Find the nearest neighbor feature point q fi , construct the corresponding relationship of feature points (p fi ,q fi ). The nearest neighbor search is based on two criteria: Euclidean distance and SIFT feature similarity. Calculate p fi With Q f The spatial distance between all feature points in the grid is used to select the feature point with the smallest distance as the candidate matching point. The calculation formula of Euclidean distance is: Among them, p fi,x ,p fi,y ,p fi,z and p fj,x ,p fj,y ,p fj,z are the three-dimensional coordinates of the two feature points respectively. fi The SIFT feature vector of the candidate matching point is used to calculate the cosine similarity between them. The SIFT feature vector is 128-dimensional and represents the gradient distribution of the local neighborhood of the feature point. The formula for calculating cosine similarity is: Among them, p fi,sift and q fj,sift are the SIFT feature vectors of two feature points respectively, and ||*|| represents the L2 norm of the vector. Considering the Euclidean distance and SIFT similarity comprehensively, the feature point q with the smallest distance and the largest similarity is selected. fj As p fi To improve the reliability of matching, the distance threshold d is set. t and similarity threshold s tWhen the Euclidean distance d(p fi ,q fj )>d t Or cosine similarity s(p fi ,q fj )<s t When , the match is considered invalid and no corresponding relationship is established. The selection of threshold needs to be adjusted according to the characteristics of point cloud and BIM model. This application sets d t =0.5m,s t =0.8.

[0041] According to the established feature point correspondence, the least squares method is used to calculate the optimal rigid body transformation matrix T f , so that the distance error between the transformed feature point set Pf and the feature point set Qf is minimized. Construct the error function: For each feature point corresponding to (p fi ,q fi ), calculate the coordinate residual between them: r i =q fi -T f p fi ; Among them, T f is the rigid body transformation matrix to be solved, including the rotation matrix R (3×3) and the translation vector t (3×1). The residual sum of squares of all feature point pairs is used as the error function: Linearization: Linearize the rotation matrix R into a rotation vector r (3×1), and use the Rodrigues formula to convert the rotation vector and the rotation matrix. Introduce increments Δr and Δt, and perform a first-order Taylor expansion on the error function: Where, [Δr] × is the antisymmetric matrix corresponding to Δr. Solve the increment: Derivative the linearized error function with respect to Δr and Δt to obtain the canonical equation of the least squares problem: Where J is the Jacobi matrix of the error function for Δr and Δt, and f is the error vector. Solve the regular equation to obtain the least squares solution of the increments Δr and Δt. Update the transformation: Use the obtained increments to update the rotation vector r and translation vector t: r: = r + Δr; t: = t + Δt; Convert the updated rotation vector r into the rotation matrix R to obtain the new rigid body transformation matrix T f Iterative optimization: Repeatedly update T f , until the norm of the increments Δr and Δt is less than the set threshold (set to 1e-6 in this application), or the number of iterations reaches the upper limit (set to 100 in this application). Output result: The final converged rigid body transformation matrix Tf is used as the initial transformation matrix between the point cloud P and the BIM model Q for the subsequent iterative optimization process.

[0042] Based on the initial registration, the weightedICP algorithm is used to iteratively optimize the point cloud P' and the BIM model Q to solve the precise rigid body transformation matrix T. The optimized transformation matrix is ​​used for the final fusion and update of the point cloud and the BIM model to generate the builtBIM model. Specifically, for each point p in the transformed point cloud P' i ', find the point q with the closest Euclidean distance in the BIM model Q i , construct point correspondence (p i ',q i ). The nearest neighbor search uses efficient spatial index structures such as KD-tree to reduce the search complexity. The weight coefficient w is introduced i , indicating a point pair (p i ',q i ) in the calculation of registration error. i The value of takes into account the following factors: Distance error: The smaller the Euclidean distance between the point pairs, the greater the weight. Introducing the distance threshold d max , when the distance is greater than d max When , the weight is 0. Normal vector deviation: Calculate p i 'and q i The normal vector at and When the angle between the two normal vectors is less than the set threshold a max When , the weight is larger; otherwise, the weight is smaller. Point density: For areas with low local point density in P' and Q, reduce the weight value to reduce the impact of sparse points on the registration results. i The calculation formula is:

[0043] Where density(*) represents the local point density, which can be estimated by calculating the average distance of the k nearest neighbors of the point.

[0044] Error function construction: Using weighted point pairs, construct the least squares error function: Where T is the rigid body transformation matrix to be optimized, which includes the rotation matrix R (3×3) and the translation vector t (3×1). The goal is to find the optimal T to minimize the weighted distance error. Nonlinear optimization solution: The Levenberg-Marquardt (LM) algorithm is used to perform nonlinear optimization on the error function to solve the optimal transformation matrix T. The LM algorithm is a trust region method that updates T in an iterative manner and adaptively switches between the Gauss-Newton method and the gradient descent method, taking into account both convergence speed and stability. The error function is linearized with respect to the rotation R and translation t in T to obtain the incremental equation: Where J is the Jacobi matrix of the error function for the rotation vector r and the translation vector t, W is the weight matrix, and f is the error vector. Introducing the damping factor λ, the LM increment equation is constructed: Initialize λ to a small positive value (such as 0.001), solve the incremental equation to get [Δr; Δt]. Use the incremental update T to calculate the new weighted distance error If e new <e, accept the new T and reduce λ (such as dividing by 10); otherwise, increase λ (such as multiplying by 10) and solve the incremental equation again. Repeat the above process until e new The change in is less than the set threshold, or the number of iterations reaches the upper limit (such as 20 times). Iterative update: Use the optimized transformation matrix T to transform the point cloud P' to obtain a new point cloud P". Calculate the weighted distance error e between P" and Q. If e is less than the set threshold (such as 0.01m) or the number of iterations reaches the upper limit (such as 50 times), the iteration is terminated and the optimal transformation matrix T is output; otherwise, P" is assigned to P', and the nearest neighbor search and weight calculation are repeated to continue the next round of iteration.

[0045] Point cloud BIM fusion: The optimal rigid body transformation matrix T obtained by iterative optimization is used to transform the complete original point cloud P to obtain the globally registered point cloud P aligned . aligned Attach it to the BIM model Q, and update and correct the component information of the BIM model by comparing the geometric deviations of the two. Geometric deviation analysis: For P aligned Every point p in i , calculate the directed distance d to the nearest plane in the BIM model Q i .d i The sign of indicates whether the point is inside or outside the plane. i The distribution histogram of P aligned The overall geometric deviation from Q. Component matching: Focus on electromechanical components such as pipelines and equipment, and match P alignedMatch the point cloud blocks in the BIM model with the corresponding components in the BIM model. A rule-based matching method is used to make full use of the prior knowledge of the mechanical and electrical profession to improve the accuracy of matching. Component update: For successfully matched components, the point cloud blocks are used to update their geometric parameters such as position, size, angle, etc. in the BIM model. Through least squares fitting, principal component analysis and other methods, the precise pose information of the components is extracted from the point cloud blocks and applied to the correction of the BIM model. For linear components such as pipes, their center lines are reconstructed, and parameters such as pipe diameter and elbow position are adjusted; for complex components such as equipment, parametric modeling methods are used to optimize their geometric shapes according to point cloud blocks. Missing component completion: For components that exist in the BIM model but are missing in the point cloud, their information in the BIM model is retained and marked as "unverified". For components that exist in the point cloud but are missing in the BIM model, the corresponding components are supplemented in the BIM model according to the geometric features and topological relationships of the point cloud blocks and marked as "newly added". Connection relationship reconstruction: According to the updated component position and geometric parameters, its topological connection relationship in the BIM model is recalculated. For pipeline components, the connectivity of their endpoints is detected and the pipeline network is rebuilt. For equipment components, the spatial relationship between them and adjacent components such as pipelines and supports is analyzed, and the position and direction of the connection ports are updated.

[0046] Figure 3 This is an exemplary flow chart for constructing an electromechanical BIM model of a construction site according to the present application. According to the built BIM model of the construction site, the electromechanical BIM model of the construction site is constructed by integrating the electromechanical BIM model of the design stage; by integrating the built BIM model obtained by on-site scanning with the design model, the precise positioning and attribute matching of the electromechanical components are achieved, and an electromechanical BIM model reflecting the actual construction status is generated, including: importing the built BIM model and the electromechanical BIM model of the design stage into the same BIM software platform, such as Autodesk Revit, Bentley AECOsim, etc. The coordinate systems of the two models are unified into the same world coordinate system by using a three-dimensional coordinate conversion method. Common reference points in the built BIM model and the design model, such as building corners, column grid intersections, etc., are identified, and their three-dimensional coordinates are extracted. Through the three-dimensional coordinate conversion, the coordinate transformation matrix T between the built BIM model and the design model is calculated. T contains a rotation matrix R (3×3) and a translation vector t (3×1). Using the transformation matrix T, the coordinates of all components in the design model are converted to the coordinate system of the built BIM model to unify the coordinates of the two models.

[0047] Design model parameter extraction: Traverse the electromechanical BIM model in the design phase and extract the geometric parameters and non-geometric parameters of each electromechanical component. Geometric parameter extraction: Through the API interface of the BIM software platform, obtain the geometric information such as the size (length, width, height), position (starting point, end point coordinates), direction (rotation angle) of each electromechanical component. Non-geometric parameter extraction: From the attribute information of the component, read its type (such as pipeline, equipment), material (such as steel, plastic), specification (such as pipe diameter, rated power), connection relationship (such as pipeline connection port, equipment interface) and other non-geometric parameters. Organize the extracted geometric parameters and non-geometric parameters according to the component ID and store them in a structured database, such as SQL Server, Oracle, etc., to facilitate subsequent query and matching.

[0048] Point cloud cluster acquisition: Based on the built BIM model, a three-dimensional spatial search algorithm is used to obtain point cloud clusters with high spatial position overlap with each electromechanical component. Spatial index construction: Spatial index the point cloud data in the built BIM model, such as octree, KD tree, etc., to accelerate the spatial search process. Spatial search: Use the geometric parameters (size, position, direction) of each electromechanical component as the query condition, search in the spatial index, and obtain the point cloud blocks that intersect or contain the component boundary box. Point cloud cluster extraction: Cluster and segment the searched point cloud blocks to extract point cloud clusters with high spatial position overlap with the electromechanical component. Density-based clustering algorithms, such as DBSCAN and OPTICS, are used to adaptively divide point cloud clusters according to point cloud density and distance thresholds. Noise filtering: De-noise the extracted point cloud clusters, remove outliers and small-scale point cloud fragments, and improve the point cloud quality. Statistical filtering, radius filtering and other methods are used to remove noise points according to local features of the point cloud.

[0049] Semantic matching and positioning: The non-geometric parameters of electromechanical components are used as semantic labels, and the semantic labels are matched with the obtained point cloud clusters to determine the installation position and connection relationship of each electromechanical component at the construction site. Semantic similarity calculation: For each point cloud cluster, its geometric features (such as size, direction) and texture features (such as color, reflection intensity) are extracted and compared with the non-geometric parameters of the electromechanical components in the design model to calculate the semantic similarity. A rule-based matching method is used to set the similarity threshold, and the components with the highest semantic similarity are selected as candidate matching objects. Coordinate deviation calculation: For candidate matching objects, the deviation distance between their coordinates in the design model and the center coordinates of the point cloud cluster is calculated. Optimization algorithms such as the least squares method are used to estimate the precise installation position and angle of the component to minimize the coordinate deviation. Matching optimization: The semantic similarity and coordinate deviation are combined to determine the optimal matching result through a weighted scoring mechanism. Components with a matching degree lower than the threshold are marked as "unmatched" and left for manual processing. Connection relationship reconstruction: According to the matching results, the spatial position relationship between adjacent components is analyzed to reconstruct their connection relationship at the construction site. For pipeline components, detect the connectivity of their endpoints and generate a pipeline topology network; for equipment components, identify their connection ports with pipelines and supports and hangers, and update connection properties.

[0050] According to the matching results, the electromechanical BIM model in the design phase is updated to generate the as-built electromechanical BIM model of the construction site. Geometry update: For successfully matched components, the geometric parameters (size, position, direction) in the design model are updated to the actual values ​​extracted from the point cloud cluster, and the spatial position and shape of the components are corrected. Attribute update: For successfully matched components, the non-geometric parameters (such as material and specification) in the design model are synchronized with the semantic labels of the point cloud cluster to ensure the consistency of attribute information. Missing component completion: For components that exist in the design model but are missing in the built BIM model, their information in the design model is retained and marked as "unconstructed". For components that exist in the built BIM model but are missing in the design model, the corresponding components are supplemented in the design model according to the semantic labels and geometric features of the point cloud cluster, and marked as "newly added on site". Model output: The updated electromechanical BIM model is output in a common data format (such as IFC, XML) for use in the construction management and operation and maintenance stages. At the same time, a model update report is generated to record the matching results, correction content, problem components and other information during the model integration process to facilitate quality traceability and audit.

[0051] Figure 4It is an exemplary flow chart of the directed acyclic graph DAG for collaborative optimization of construction tasks according to the present application, and the judgment matrix method DSM is used to obtain the logical dependencies between electromechanical tasks in the electromechanical BIM model, and the attribute parameters of the task nodes are extracted; according to the logical dependencies between electromechanical tasks, the timing and resource parameters of the nodes, the directed acyclic graph DAG for collaborative optimization of tasks is constructed, including: task extraction and dependency matrix establishment: traverse the electromechanical BIM model of the construction site, and extract the construction tasks corresponding to each component. A rule-based method is adopted to match the predefined task template according to the type, material, specification and other attributes of the component, and automatically generate a construction task list. For example, for pipeline components, tasks such as "pipeline prefabrication", "pipeline installation", and "pipeline connection" are generated; for equipment components, tasks such as "equipment foundation construction", "equipment in place", and "equipment commissioning" are generated.

[0052] According to the construction technology and process, the dependencies between tasks are analyzed. Dependencies include: Logical dependency: the order of tasks, such as "pipeline installation" must be carried out after "pipeline prefabrication" is completed. Resource dependency: the competition between tasks for shared resources, such as "equipment in place" and "pipeline installation" may need to use the same lifting equipment. Spatial dependency: the spatial position constraint relationship between tasks, such as "pipeline installation" must be carried out after "equipment in place" is completed to avoid spatial conflicts. A rule-based reasoning method is used to automatically infer the dependencies between tasks according to predefined dependency templates. Dependency matrix establishment: Based on the identified dependencies, a dependency matrix A between tasks is established. The number of rows and columns of matrix A is the number of tasks n, and the element A(i, j) represents the dependency of task i on task j. When task i must start after task j is completed, the value of A(i, j) is set to 1, otherwise it is set to 0. The dependency matrix A is an asymmetric matrix that reflects the directed dependencies between tasks.

[0053] Task attribute and geometric parameter extraction: For each construction task, extract its attribute parameters from the electromechanical BIM model, including: Work content: the specific construction content of the task, such as "pipe welding" and "equipment installation". Workload: the engineering quantity of the task, such as pipeline length, number of equipment, etc. Timing parameters: the duration, earliest start time and latest completion time of the task. The duration is estimated based on the workload and quota efficiency of the task; the earliest start time is calculated based on the logical dependency of the task; the latest completion time is reversed based on the duration of the task and the earliest start time of the subsequent tasks. Regular expressions, keyword matching and other text parsing technologies are used to automatically extract the attribute parameters of the task from the attribute information of the BIM component. Geometric parameter extraction: For each construction task, extract the geometric parameters of the corresponding component from the electromechanical BIM model, including: Spatial position coordinates: coordinate information such as the starting point, end point, and center point of the component. Dimensional parameters: dimensional information such as the length, diameter, and volume of the component. The API interface of the BIM software is used to read the geometric information of the component and convert it into a common data format, such as JSON, CSV, etc.

[0054] Directed weighted graph construction: Graph node generation: Data structure: Use structures or classes to represent graph nodes, including node ID, task work content, workload, earliest start time, latest start time, earliest completion time, latest completion time and other attributes. Data storage: Store node data in arrays or vectors in memory for quick access and search. Data reading: Read the attribute information of each task from the generated task attribute parameter file (such as JSON, CSV format) and convert it into the data structure of graph nodes. Graph edge generation: Data structure: Use structures or classes to represent graph edges, including attributes such as starting node ID, end node ID, edge weight and other attributes. Adjacency matrix conversion: According to the dependency matrix A, traverse each element A(i, j) of the matrix. If A(i, j) = 1, add a directed edge from node i to node j in graph G. Adjacency list generation: Convert the adjacency matrix into the data structure of the adjacency list to facilitate subsequent graph traversal and algorithm implementation. The adjacency list can be stored using an array or vector, and each node corresponds to a linked list to store the information of the edges connected to it. Edge weight calculation: Data acquisition: For each edge (i, j) in the graph G, obtain the relevant timing parameters from the attribute information of node i and node j, including the earliest completion time of the predecessor task i, the latest start time of the successor task j, etc. Time interval setting: Set the time interval between task i and task j according to factors such as the spatial position relationship of the tasks and resource occupancy. A rule-based method can be used to automatically calculate the time interval value according to the predefined time interval template. Edge weight calculation: According to the formula of the CPM algorithm, calculate the weight value of edge (i, j): weight value = earliest start time of the successor task j - earliest completion time of the predecessor task i - time interval; if the calculated weight value is a negative number, it is set to 0, indicating that there is no time margin between the predecessor task i and the successor task j. Data storage: The calculated edge weight value is stored in the edge data structure, and together with the start and end point information of the edge, it forms a complete edge data.

[0055] Critical path identification: Forward traversal: Starting from the starting node, traverse all nodes of the directed graph G in the order of topological sorting. For each node, calculate its earliest start time and earliest completion time: the earliest start time of the node = Max (the earliest completion time of the immediate predecessor task) + time interval; the earliest completion time of the node = the earliest start time of the node + the duration of the node; Reverse traversal: Starting from the end node, traverse all nodes of the directed graph G in the order of reverse topological sorting. For each node, calculate its latest start time and latest completion time: the latest completion time of the node = Min (the latest start time of the immediate successor task) - time interval; the latest start time of the node = the latest completion time of the node - the duration of the node; Critical path identification: For each node, calculate its time margin: the time margin of the node = the latest start time of the node - the earliest start time of the node; if the time margin of the node is 0, the node is a node on the critical path. Data output: Output all nodes on the critical path as a list of critical tasks and arrange them in the order of topological sorting. At the same time, calculate the total duration of the critical path as the shortest completion time of the entire project. The time complexity of the CPM algorithm is O(n+m), where n is the number of nodes and m is the number of edges. For dense graphs, the storage space complexity of the adjacency matrix is ​​O(n+m). 2 ), the space complexity of the adjacency list is O(n+m). Therefore, when the task scale is large, it is recommended to use the adjacency list data structure and adopt an optimized topological sorting algorithm, such as the Kahn algorithm or the DFS algorithm, to reduce the time and space complexity.

[0056] According to the directed weighted graph, as well as the attribute parameters and geometric parameters of each task node, a directed acyclic graph DAG for task collaborative optimization is established. Including: critical path and free time difference calculation: forward recursive calculation: starting from the starting node, recursively calculate the earliest start time of each task node in the order of topological sorting. For node i, its earliest start time ES i The calculation formula is: ES i =max{ES j +d j}, where j is the immediate predecessor task of i; where d j represents the duration of task j, ES j Represents the earliest start time of the immediate predecessor task j. Reverse recursive calculation: Starting from the end node, recursively calculate the latest start time of each task node in the order of reverse topological sorting. For node i, its latest start time LS i The calculation formula is: LS i = min{LS j -d j}, where j is the task immediately following i; j represents the duration of task j, LS jIndicates the latest start time of the next task j. Free time difference calculation: For each task node i, calculate its free time difference FT i :FT i =LS i -ES i ; If FT i =0, then task i is a critical task; if FT i >0, then task i is a non-critical task.

[0057] Critical task and non-critical task extraction: Critical task extraction: traverse all task nodes, extract nodes with free time difference of 0, and form a critical task set. Non-critical task extraction: traverse all task nodes, extract nodes with free time difference greater than 0, and form a non-critical task set. Critical task dependency strength calculation: Edge extraction: extract all edges between critical task nodes to form a critical task edge set. Dependency strength calculation: For each critical task edge (i, j), calculate its dependency strength DS(i, j): Among them, d j represents the duration of task j, LS j represents the latest start time of task j, ES i represents the earliest start time of task i, R ij represents the number of common resources required by tasks i and j, R j Represents the total amount of resources required for task j, α and β are the weight coefficients of the time constraint tightness and resource constraint tightness respectively. Data storage: Store the calculated dependency strength value in the data structure of the critical task edge. Non-critical task spatial distance calculation: Geometric parameter extraction: Extract the spatial position coordinates corresponding to the non-critical task node. Spatial distance calculation: For each pair of non-critical task nodes (i, j), calculate its spatial distance SD(i, j): Among them, (x i ,y i ,z i ) and (x j ,y j ,z j ) represent the spatial position coordinates corresponding to task i and task j respectively. Data storage: The calculated spatial distance value is stored in the data structure of non-critical task node pairs.

[0058] Task adjustment and graph update: Deletion of key tasks: Dependency strength threshold setting: Set the dependency strength threshold DT according to the actual project situation and optimization goals. The threshold can be determined by empirical values, statistical analysis, etc. Key task edge screening: Traverse the key task edge set, and for each edge (i, j), determine whether its dependency strength DS(i, j) is lower than the threshold DT. If DS(i, j) < DT, add edge (i, j) to the set of edges to be deleted. If DS(i, j) ≥ DT, retain edge (i, j). Deletion of key task nodes: Delete the corresponding key task nodes according to the set of edges to be deleted. For each edge to be deleted (i, j), if the out-degree of node i is 0 and node i is not the starting node, delete node i; if the in-degree of node j is 0 and node j is not the ending node, delete node j. Update the key task node set and key task edge set.

[0059] Non-critical task adjustment: Spatial distance threshold setting: Set the spatial distance threshold ST according to the actual project situation and optimization goals. The threshold can be determined by empirical values, statistical analysis and other methods. Non-critical task node screening: Traverse the set of non-critical task node pairs, and for each pair of nodes (i, j), determine whether its spatial distance SD(i, j) is greater than the threshold ST. If SD(i, j)>ST, add node i and node j to the set of nodes to be adjusted. If SD(i, j)≤ST, retain node i and node j. Non-critical task node adjustment: For each node i to be adjusted, adjust it according to its time and space attributes. Time adjustment: At the earliest start time ES of node i i and the latest start time LS i Choose a suitable start time, such as (ES i +LS i ) / 2, update the start time attribute of node i. Spatial adjustment: According to the spatial position coordinates (x i ,y i ,z i ), select a suitable new position (x i ',y i ',z i '), update the spatial location attribute of node i. Update the set of non-critical task nodes.

[0060] Graph update: Node update: Update the node set of the directed weighted graph according to the results of the deletion of critical task nodes and the adjustment of non-critical task nodes. Delete the deleted critical task nodes and update the node numbers. Update the attribute information of the adjusted non-critical task nodes, including start time, spatial location, etc. Edge update: Update the edge set of the directed weighted graph according to the results of the deletion of critical task nodes and the adjustment of non-critical task nodes. Delete the edges connected to the deleted nodes. Update the weight values ​​of the edges connected to the adjusted nodes, such as time intervals, dependency strength, etc. Graph reconstruction: Reconstruct the data structure of the directed weighted graph, such as adjacency lists, adjacency matrices, etc., based on the updated node set and edge set.

[0061] Directed acyclic graph DAG generation: Topological sorting: Initialization: Create a queue Q to store nodes with an in-degree of 0. Create an array indegree to record the in-degree of each node. In-degree calculation: Traverse all nodes of the updated directed weighted graph, calculate the in-degree of each node, and add nodes with an in-degree of 0 to the queue Q. Topological sorting: When the queue Q is not empty, repeat the following steps: Take a node i from the queue Q and add it to the result array of topological sorting. For each adjacent node j of node i, reduce the in-degree of the edge (i, j) by 1. If the in-degree of node j becomes 0, add node j to the queue Q. If all nodes are included in the result array of topological sorting, it means that the graph is a directed acyclic graph DAG; otherwise, there is a loop in the graph and it is not a DAG. DAG construction: Node generation: Generate a node set of DAG based on the result array of topological sorting. Each node contains attribute information of the task, such as task name, duration, start time, end time, spatial location, etc. Edge generation: Generate the edge set of the DAG based on the updated edge set of the directed weighted graph. Each edge contains information such as the starting node, the end node, and the weight of the edge. DAG storage: Use appropriate data structures to store DAG, such as adjacency lists, adjacency matrices, etc., to facilitate subsequent traversal and analysis.

[0062] According to the directed acyclic graph DAG of task collaborative optimization, the task nodes on the critical path are obtained; the critical path represents the path corresponding to the process with zero free time difference; including: node time parameter extraction: forward traversal: starting from the starting node of DAG, forward traverse all nodes of DAG in the order of topological sorting. For node i, calculate its earliest start time ES i : If node i is the starting node, then ES i = 0. If node i is not the starting node, then ES i =max{ES j +d j}, j is the immediate predecessor task node of i}, where d jRepresents the duration of node j. Reverse traversal: Starting from the end node of DAG, traverse all nodes of DAG in reverse order according to the reverse topological sorting. For node i, calculate its latest start time LS i :If node i is a terminal node, then LS i =ES i If node i is not a terminal node, then LS i = min{LS j -d j}, j is the next task node of i}, where dj represents the duration of node j. Free time difference calculation: For each node i, calculate its free time difference FT i :FT i =LS i -ES i ; Data storage: The earliest start time, latest start time and free time difference of each node are stored in the node's attributes.

[0063] Critical path extraction: Path traversal: Use the depth-first search (DFS) or breadth-first search (BFS) algorithm to traverse all paths in the DAG. DFS traversal: Starting from the start node of the DAG, recursively traverse the successor nodes of each node until the end node is reached. During the traversal, record the sequence of nodes passed through to generate a complete path. Repeat this process until all possible paths are traversed. BFS traversal: Start from the start node of the DAG and add it to the queue. Then repeat the following steps until the queue is empty: Take a node from the queue and add it to the current path. If the current node is the end node, add the current path to the path set. If the current node is not the end node, add all its successor nodes to the queue and copy the current path to each successor node. Critical path screening: For each traversed path, determine whether it is a critical path. Extract the free time difference values ​​of all nodes on the path. If the free time difference of all nodes on the path is 0, the path is a critical path and is added to the critical path set. If there is a node on the path whose free time difference is greater than 0, the path is not a critical path and will be discarded. Critical task node extraction: Node extraction: Traverse each critical path in the critical path set and extract all nodes on the path. Deduplication: Deduplication is performed on the extracted nodes to obtain a unique set of critical task nodes. Data storage: The critical task node set is stored in memory for subsequent analysis. Benchmark duration calculation: Critical path duration calculation: For each critical path, calculate its total duration. Extract the duration values ​​of all nodes on the critical path. Add the duration values ​​to obtain the total duration of the critical path. Benchmark duration determination: Among the total durations of all critical paths, select the maximum value as the benchmark duration for electromechanical construction optimization. If the DAG is large in size, it may result in excessive memory usage. Appropriate data structures, such as sparse matrices, hash tables, etc., need to be used to compress storage space.

[0064] According to the task nodes on the critical path, a multi-objective optimization model is established, with the shortest construction period and the highest resource utilization as the optimization goals, and the process logic and resource occupancy as the constraints. The multi-objective particle swarm optimization algorithm MOPSO is used to solve the problem and obtain the resource optimization configuration set; including: multi-objective optimization model construction, clarifying the optimization goals: minimizing the total construction period f1 and maximizing the resource utilization f2. Define decision variables: task execution variable x i and resource allocation variable x ik . Establish the objective function: total construction period f1: Resource utilization f2: Establish constraints: Task precedence constraints: Resource usage constraints: Time window constraint: Start time constraint: End time constraint: Decision variable value constraints: Integrate the objective function and constraints to form a complete multi-objective 0-1 integer programming model. Initialize the MOPSO algorithm parameters and set the task execution variable x i and resource allocation variable x ik Combined into the position vector of the particle. Number of particles N: generally 20 to 100. Maximum number of iterations T max :Determined according to the problem scale and solution time requirements. Learning factors c1 and c2: Generally 2. Inertia weight w: Generally 0.4 to 0.9, or a linear decay strategy. Randomly generate position vectors of N particles to form the initial particle swarm Pop(0). Calculate the fitness of each particle, initialize the individual optimal position pbest and the global optimal position gbest.

[0065] Particle fitness calculation and update, fitness function construction: The fitness function needs to comprehensively consider the two core goals of the total construction period and resource utilization of electromechanical installation. The total construction period reflects the project progress and is a key time goal; resource utilization reflects the efficiency of the use of resources such as manpower and equipment, which is related to the project cost. Combining the two, a weighted sum fitness function is constructed: itness=w1(t)f1'+w2(t)f2'; where, represents the normalized total construction period, represents the normalized resource utilization rate, w1(t) and w2(t) are the dynamic weight coefficients of the two. This fitness function can quantitatively evaluate the pros and cons of each particle (i.e., the construction schedule). Objective function normalization: Since the dimensions of the two objectives, total construction period (such as 50 days) and resource utilization rate (such as 85%), are not uniform and the numerical differences are large, they need to be normalized to the dimensionless interval of [0, 1] so that they can participate equally in the fitness value calculation: Normalization of total construction period: Resource utilization normalization: where f 1max 、f 1min 、f 2max and f 2min is the extreme value of the objective function in the current group search state. The normalized f1' and f2' weaken the dimension effect and facilitate the weighted calculation of the fitness function.

[0066] Adaptive weight adjustment: During the algorithm iteration process, the importance of the two goals, total construction period and resource utilization, may change dynamically. In the early stage, more emphasis may be placed on shortening the construction period, while in the later stage, more emphasis may be placed on improving resource utilization. Therefore, it is necessary to adaptively adjust the weights w1(t) and w2(t) in the fitness function to balance the convergence and diversity of the algorithm: Weight initialization: w1(0) = w2(0) = 0.5, that is, giving the two goals equal importance. Dynamic weight update:

[0067] w1(t)=w1(t-1)+Δw1(t); w2(t)=w2(t-1)+Δw2(t);

[0068]

[0069] Where gbest is the current optimal solution, f1'(gbest) and f2'(gbest) are their normalized target values, t is the current iteration number, T max is the maximum number of iterations, k1 and k2 are factors that control the convergence speed, and λ1 and λ2 are the weight change steps. Through the adaptive formula, the weight of the corresponding target is increased when approaching the optimal solution, and the weight is reduced when deviating from the optimal solution, guiding the search direction and improving the convergence efficiency. Based on the dynamic fitness function obtained above, the fitness value of each particle (i.e., the electromechanical installation solution) is re-evaluated in each round of iteration, and high-quality particles are selected for subsequent position and speed updates, so that the population gradually evolves towards the global optimal solution with high fitness.

[0070] In the optimization of the electromechanical installation project schedule, particle position update and historical optimal position and global optimal position selection, particle speed and position update: Each particle represents a feasible electromechanical installation schedule. According to the particle fitness evaluation obtained in the current iteration, the speed update formula and position update formula are used to update the particles to make them evolve towards a better solution: i (t+1)=wv i (t)+c1r1(pbest i (t)-x i (t))+c2r2(gbest i (t)-x i (t)); x i (t+1)=x i (t)+v i (t+1)xi(t+1)=xi(t)+vi(t+1); where v i is the particle velocity, x i is the particle position, w is the inertia weight, c1 and c2 are learning factors, r1 and r2 are random numbers in [0, 1], pbest i is the particle's historical optimal position, and gbest is the global optimal position. Through the velocity formula, the particle is affected in three aspects: inertia, self-cognition, and social cognition, and moves to pbest with a certain probability. i Learn from pbest, update its search direction and step size, and then update to a new feasible solution through the position formula, so that the particles are continuously optimized in the solution space.

[0071] Particle historical optimal position update: Dominance relationship judgment: When comparing the current position of particle i i and the historical best position pbest i When we first determine their dominant relationship in the objective function space, the objective function here includes two aspects: total construction period f1 and resource utilization rate f2. i The total construction period and resource utilization rate are better than pbest i , that is, f1(x i )=f1(pbest i ) and f2(x i )=f2(pbest i ), then x i Dominate pbest i If pbest i The total construction period and resource utilization rate are better than x i , that is, f1(x i )>f1(pbest i ) and f2(x i )<f2(pbest i ), then it is called pbest i Dominate x i If x i If xi is better in total construction period but worse in resource utilization, or xi is better in resource utilization but worse in total construction period, that is, both have advantages and disadvantages, then x is called i With pbest i By judging the dominance relationship, the advantages and disadvantages of the solution can be defined in multi-objective optimization, thereby guiding pbest i Updates.

[0072] Comparison of crowding distances under non-dominance conditions: When x i With pbest i When they do not dominate each other, it is impossible to judge which one is better based on the dominance relationship alone. At this time, the crowding distance metric needs to be introduced to compare the distribution of the two in the solution space. The crowding distance measures the degree of openness near a solution. The larger the distance, the sparser the solutions in the area where the solution is located, and the more beneficial it is to maintain the diversity of the population. The calculation steps are as follows: sort the individuals in the population in ascending order on each target. For each target, assign an infinite distance value to the sorted boundary individuals (the first and the last). For each target, assign a distance value to the sorted middle individual, that is, the sum of the differences between the normalized target values ​​of its two adjacent individuals on the target. Sum the distance values ​​of each individual on each target to get its final crowding distance. When x i With pbest iWhen they do not dominate each other, compare the crowding distance CD(x i ) and CD(pbest i ), the one with a larger distance is better and is used as the new pbest i .

[0073] Update strategy: Comprehensive dominance relationship and crowding distance comparison results, pbest i The update strategy is as follows: If x i Dominate pbest i , then pbest i Update directly to x i This means that on all targets x i Both are better than pbest i , is an obviously better solution and should become the new pbest i If pbest i Dominate x i , then pbest i This means that pbest i Both are better than x i , there is no need to update. i With pbest i If they do not dominate each other, then compare their crowding distances CD(x i ) and CD(pbest i ), the one with the larger distance becomes the new pbest i This reflects the balance between the target and the distribution of solutions, and the solution in the sparser area is selected as pbest. i , to promote population diversity. Update process: pbest i The update of is an iterative process, which requires executing the above-mentioned dominance relationship judgment, crowding distance comparison and update strategy on all particles in the population until the pbest of all particles is i This process is carried out in each iteration of the algorithm, through pbest i The continuous updating of enables the particles to gradually approach the real Pareto frontier in the framework of multi-objective optimization and find the optimal trade-off point between the total construction period and resource utilization.

[0074] The local search of the particle's historical optimal position, introducing a local search strategy into the particle swarm algorithm, can make up for the premature convergence problem of the algorithm to a certain extent and improve the particle's historical optimal position pbest i The local search trigger: For each particle i, according to a preset local search probability pl s To decide whether to use pbest iPerform local search. Specifically, in each iteration, a random number rand between [0, 1] is generated for particle i. If rand < pls, local search is triggered; otherwise, no search is performed. By triggering in this probabilistic way, a balance can be achieved between the frequency of local search and the computational cost. The value of pl s can be appropriately set according to factors such as the problem scale and computational resources.

[0075] Neighborhood solution generation: Once the pbest of particle i i is selected to perform local search, then using pbest i as the initial solution, a neighborhood solution set N(pbest i ) is generated through the tabu search algorithm. Tabu search is a meta-heuristic algorithm that, by defining components such as the solution space, neighborhood structure, and tabu list, guides the search towards a better solution while avoiding cyclic search. In the problem of optimizing the progress of electromechanical installation projects, some typical local operations can be used to generate neighborhood solutions. For example: Insertion: Randomly select an operation, remove it from its current position, and insert it into another feasible position in the scheduling plan. Swap: Randomly select two operations and swap their execution order in the scheduling plan. Inversion: Randomly select a subsequence of operations and reverse its execution order. Movement: Randomly select an operation and move it to another feasible position in the scheduling plan. Through the above operations, based on pbest i , a series of neighborhood solutions with similar structures but local changes are generated to form the neighborhood solution set N(pbest i ). At the same time, in order to avoid the search falling into a cycle, a tabu list needs to be set to record the solutions visited recently and avoid repeated search within a certain number of iterations.

[0076] Neighborhood solution evaluation: For each candidate solution in N(pbest i ), its fitness under multi-objective optimization needs to be evaluated to decide whether to update the solution of pbest i . First, perform a fast non-dominated sorting on the solutions in N(pbest i ), and divide them into different levels of non-dominated fronts F i according to the dominance relationship between the solutions. The set composed of non-dominated solutions is the candidate set ND i , and the optimal solution is selected from it to update pbest i . Optimal neighborhood solution selection: In order to select the optimal solution from the candidate set ND i , the crowding distance of each solution needs to be calculated to compare its distribution in the solution space. The larger the crowding distance, the lower the solution density in the area where the solution is located, and the greater the contribution to the population diversity. Select the solution with the largest crowding distance in ND i as pbest iThe updated value of pbest is completed by local search i improvements.

[0077] Iteration process: The local search strategy is embedded in the iterative process of the particle swarm algorithm. For each particle in the population, in each iteration, according to the probability pl s Trigger local search, generate neighborhood solution set through taboo search, and select the one with the largest crowding distance from the non-dominated solution to update pbest until the local search of all particles is completed. i With the continuous improvement of the particle swarm algorithm, the overall search ability of the particle swarm algorithm has been enhanced, and multiple potential local optimization areas have been formed in the solution space, which has accelerated the convergence speed of the algorithm. Introducing the local search strategy into the mechanical and electrical installation project schedule optimization problem can give full play to its advantages in local optimization and optimize the historical optimal position of the particle pbest. i Deep mining and improvement are carried out. Through the measurement of non-dominated sorting and crowding distance, the superiority of the selected neighborhood optimal solution in the sense of multiple objectives is guaranteed, taking into account both the convergence of the understanding in the objective space and the diversity of the understanding in the decision space. This strategy makes up for the defect of the particle swarm algorithm that is prone to premature convergence, so that the algorithm has a strong local optimization ability while maintaining the global search ability, which is more conducive to solving complex multi-objective scheduling optimization problems.

[0078] External archive set update and global optimal position selection: In each round of iteration, the pbest of all particles is i The external archive set EP is formed to save the current non-dominated solution. The update of EP and the selection of gbest are as follows: i The external archive set EP is formed. The solutions in EP are quickly non-dominated sorted and divided into non-dominated frontiers F of different levels according to the dominance relationship of the solutions. i . Calculate the crowding distance of the solution in each frontier. According to the frontier level and crowding distance, select the N best non-dominated solutions from EP as the new EP, where N is a fixed value or related to the number of iterations. Keep the solutions with high level and large crowding distance to balance convergence and diversity. Randomly select a solution from the updated EP as the global optimal position gbest for the next round of iteration. The reason for random selection is to reduce the selection pressure of gbest and prevent premature convergence.

[0079] The multi-objective particle swarm optimization algorithm (MOPSO) is closely integrated with the specific construction task scheduling, and the optimal resource allocation scheme is found through iterative search, and the corresponding construction schedule is generated accordingly. During the algorithm iteration process, it is necessary to determine whether the preset maximum number of iterations T has been reached. max . T maxThe value of T needs to be appropriately set according to the scale and complexity of the problem, ensuring that the algorithm has enough search time to obtain high-quality solutions while avoiding unnecessary waste of calculations. Usually, T can be determined through experience or experiments. max After each round of iteration, the current number of iterations t and T max Compare: If t<T max , indicating that the maximum number of iterations has not been reached, then set t = t + 1, return to the main loop of the algorithm, and continue to execute the steps of particle position update, fitness evaluation, historical optimal position update, etc., and perform the next round of search. max x, indicating that the maximum number of iterations has been reached, the algorithm will terminate the search process and enter the stage of generating the optimal solution.

[0080] When the algorithm reaches the maximum number of iterations T max , it is necessary to select the global optimal solution gbest from the current population as the final output of MOPSO. gbest represents the optimal resource allocation solution in the sense of multi-objective optimization, reflecting the balance between construction period and resource utilization. Specifically, the particles corresponding to gbest contain the execution order and resource allocation of each construction task. By sorting tasks and optimizing resources, the efficiency of resource utilization is maximized while meeting the construction period constraints. By decoding and converting the particle information corresponding to gbest, an optimized resource configuration set can be obtained, which clarifies the key elements such as the start and end time of each task, the type and quantity of required resources, etc.

[0081] The global optimal solution gbest obtained from the multi-objective particle swarm algorithm (MOPSO) can be further decoded and converted into a detailed construction scheduling plan to guide the actual project execution. This process involves key elements such as task sorting, resource allocation, and time schedule. It is necessary to map the abstract information contained in gbest to specific construction activities and resource allocation. Decoding of the optimal solution: The optimal solution gbest obtained by the MOPSO algorithm search is essentially a high-dimensional particle, each dimension of which represents a decision variable of the problem, such as task execution order, resource allocation plan, etc. In order to convert gbest into an executable construction plan, it is first necessary to decode it and extract key scheduling information. The specific decoding process is as follows: Task execution order: According to the dimension information of gbest, determine the relative execution order of each task in the scheduling plan. By sorting or mapping the particle dimensions, a task sequence is obtained, which represents the execution relationship of the tasks in sequence. Resource allocation plan: From the dimension information of gbest, extract the type and quantity of resources allocated to each task. Resources usually include multiple aspects such as manpower, equipment, and materials, which need to be reasonably matched and allocated according to task characteristics and resource constraints. Task duration: Estimate the duration of each task based on the workload and the number of resources allocated. You can use empirical formulas or historical data to make inferences, or you can consider the impact of factors such as resource efficiency and learning effects.

[0082] Task dependency: Extract the logical dependencies between tasks, such as structural dependency (such as foundation first and then main body), resource dependency (tasks with shared resources cannot be performed at the same time), etc. These dependencies have an important impact on the execution order and time arrangement of tasks. Through the decoding process, gbest can be converted into an optimized resource configuration set, which clarifies the execution order, required resources, duration and dependencies of each task, providing a basis for generating a detailed construction schedule. Generation of construction schedule: After obtaining the optimized resource configuration set, it is necessary to further generate a detailed construction schedule to determine the specific execution time and resource arrangement of each task. This process can be achieved through classic project management tools such as critical path method (CPM) and Gantt chart. The specific generation steps are as follows: According to the execution order and dependency of tasks, the forward calculation of CPM is used to determine the earliest start time (EST) and earliest end time (EFT) of each task. According to the latest start time (LST) and latest end time (LFT) of the task, the task progress is optimized and adjusted to compress the construction period as much as possible, while avoiding resource conflicts and logical paradoxes. Draw the EST, EFT, duration, required resources and other information of each task on the Gantt chart according to the timeline to intuitively display the time arrangement and parallel relationship of the tasks. Mark important nodes such as critical paths and milestones in the Gantt chart to highlight the key tasks and time constraints of the project and improve the pertinence of progress control. Make necessary beautification and annotations to the Gantt chart, such as adding task descriptions, resource information, responsible persons, etc., to improve the readability and practicality of the scheduling plan. Output a complete mechanical and electrical construction scheduling plan, including Gantt charts, task lists, resource allocation tables, etc., as an important basis for project teams and site management. Generate an optimized mechanical and electrical construction scheduling plan based on the optimal solution gbest obtained by the MOPSO algorithm. The plan comprehensively considers multiple factors such as task execution order, resource allocation, time constraints, etc., while meeting the construction period requirements, maximizing resource utilization efficiency and achieving multi-objective balanced optimization.

Claims

1. A mechanical and electrical construction method based on BIM technology, characterized in that: include: Collect 3D point clouds of construction sites; The 3D point cloud is registered and integrated with the BIM model in the design phase to construct a built BIM model of the construction site; According to the built BIM model of the construction site, the mechanical and electrical BIM model of the construction site is constructed by integrating the mechanical and electrical BIM model of the design stage; The judgment matrix method DSM is used to obtain the logical dependencies between electromechanical tasks in the electromechanical BIM model and extract the attribute parameters of the task nodes; according to the logical dependencies between electromechanical tasks, the timing and resource parameters of the nodes, a directed acyclic graph DAG for task collaborative optimization is constructed; According to the directed acyclic graph DAG of task collaborative optimization, the task nodes on the critical path are obtained; wherein the critical path represents the path corresponding to the process with zero free time difference; According to the task nodes on the critical path, the objective function of multi-objective optimization is established, with the shortest construction period and the highest resource utilization as the optimization goals, and the process logic and resource occupancy as the constraints. The multi-objective particle swarm optimization algorithm MOPSO is used to solve the problem and obtain the optimal resource configuration set. Generate mechanical and electrical construction schedule based on resource optimization configuration set; Among them, the multi-objective particle swarm optimization algorithm MOPSO is used to solve the resource optimization configuration set, including: According to the set of key task nodes, the objective function and constraints of multi-objective optimization are established: The objective function expression of multi-objective optimization is as follows: Constraints: Process logic constraints: Resource usage constraints: Start time constraint: End time constraint: Variable restriction constraints: Among them, f1 represents the total duration of the project; f2 represents the utilization rate of project resources; d i represents the duration of critical task node i; x i is the execution decision variable of key task node i; n represents the total number of key task nodes in the project; r ik represents the demand of key task node i for resource type k; x ik is the allocation decision variable between key task node i and resource type k; R k represents the total amount of the kth resource in the project; m represents the total number of resource types involved in the project; According to the particle swarm, calculate the fitness of each particle, and update the particle's historical optimal position and global optimal position; According to the current position, historical optimal position and global optimal position of the particle, the speed update formula and position update formula are used to update the speed vi and position positioni of each particle; Determine whether the maximum number of iterations T has been reached max If yes, the particle corresponding to gbest is taken as the optimal solution of MOPSO and the resource optimization configuration set is output; otherwise, let t = t + 1 and continue the iteration.

2. The electromechanical construction method based on BIM technology according to claim 1, characterized in that: Calculate the fitness of each particle and update the particle's historical optimal position and global optimal position, including: Calculate the fitness of each particle, including: itness=w1(t)f1'+w2(t)f2' Where w1(t) and w2(t) represent the adaptive weight coefficients of objective functions f1 and f2 at the tth iteration respectively; f 1max and f 1min They represent the maximum and minimum f1 values ​​of all particles in the current iteration respectively; f 2max and f 2min They represent the maximum and minimum values ​​of f2 of all particles in the current iteration respectively; they represent the value of the normalized objective function f1 of the particle corresponding to the solution; and f2' represents the value of the normalized objective function f2 of the particle corresponding to the solution.

3. The electromechanical construction method based on BIM technology according to claim 2 is characterized in that: The adaptive weight coefficient is calculated by the following formula: When t>0, w1(t)=w1(t-1)+Δw1(t); w2(t)=w2(t-1)+Δw2(t); Among them, Δw1(t) and Δw2(t) are the changes of weight coefficients at the tth iteration, which are adaptively adjusted according to the convergence of the particle swarm: Among them, λ1 and λ2 are weight adjustment coefficients; f1'(gbest) and f2'(gbest) are the normalized objective function values ​​corresponding to the current global optimal solution gbest; f1'(t) and f2'(t) are the normalized objective function means of all particles in the tth iteration; f1'(0) and f2'(0) are the normalized objective function means of the initial particle swarm; t is the current iteration number, T max is the maximum number of iterations; k1 and k2 are convergence factors.

4. The electromechanical construction method based on BIM technology according to claim 2 is characterized in that: Construct a builtBIM model of the construction site, including: Import the 3D point cloud of the construction site and the BIM model of the design phase into the BIM software platform respectively; In the BIM software platform, based on the geometric features of the 3D point cloud, the region growing algorithm is used to extract the point cloud data of the building components including floor slabs, walls and columns; The point cloud data of the extracted building components are registered with the geometric elements of the corresponding components in the BIM model of the design stage. The optimal transformation matrix from the point cloud data to the geometric elements is calculated by the least squares method to obtain the three-dimensional point cloud and the BIM model of the design stage after local registration. According to the 3D point cloud after local registration and the BIM model in the design stage, the ICP registration algorithm is adopted to minimize the global error through iterative calculation to obtain the builtBIM model of the construction site.

5. The electromechanical construction method based on BIM technology according to claim 2 is characterized in that: Build the mechanical and electrical BIM model of the construction site, including: Import the builtBIM model of the construction site and the electromechanical BIM model of the design phase into the BIM software platform, and use the three-dimensional coordinate conversion method to unify the coordinates of the builtBIM model and the electromechanical BIM model of the design phase; Traverse the electromechanical BIM model in the design phase to obtain geometric parameters and non-geometric parameters; geometric parameters include the size, position and direction of components; non-geometric parameters include the type, material, specification and connection relationship of components; According to the builtBIM model of the construction site, a three-dimensional space search algorithm is used, and the geometric parameters of the electromechanical components are used as query conditions to obtain point cloud clusters whose spatial position coincidence with the electromechanical components is higher than the threshold. The non-geometric parameters of electromechanical components are used as semantic labels, and the semantic labels are matched with the acquired point cloud clusters. The installation position and connection relationship of each electromechanical component at the construction site are determined by calculating the semantic similarity and coordinate deviation. According to the matching results, the electromechanical BIM model in the design phase is updated to obtain the electromechanical BIM model of the construction site.

6. The electromechanical construction method based on BIM technology according to claim 4 is characterized in that: Construct a directed acyclic graph (DAG) for task collaborative optimization, including: According to the electromechanical BIM model of the construction site, the construction tasks corresponding to each component are extracted. According to the construction process and flow, the dependencies between the tasks are obtained, and the dependency matrix between the electromechanical tasks is established. In the dependency matrix, when task i must start after task j is completed, the value of the position (i, j) in the dependency matrix is ​​set to 1, otherwise it is set to 0. According to the electromechanical BIM model of the construction site, the attribute parameters and geometric parameters of each task node are extracted; the attribute parameters include the work content, workload and timing parameters of each task; the timing parameters include the duration, earliest start time and latest completion time of the task; the geometric parameters include the spatial position coordinates of the components corresponding to the task; The dependency matrix is ​​used as the adjacency matrix and the critical path method (CPM) is used to construct a directed weighted graph between task nodes. According to the directed weighted graph, as well as the attribute parameters and geometric parameters of each task node, a directed acyclic graph DAG for task collaborative optimization is established.

7. The electromechanical construction method based on BIM technology according to claim 5 is characterized in that: Based on the directed weighted graph, as well as the attribute parameters and geometric parameters of each task node, a directed acyclic graph DAG for task collaborative optimization is established, including: According to the directed weighted graph, the critical path method (CPM) is used to calculate the earliest start time of each task node through forward recursion, and the latest start time of each task node is calculated by reverse recursion. The free time difference of the task node is calculated based on the earliest start time and the latest start time; According to the earliest start time, latest start time and free time difference of the task node, obtain the critical task and non-critical task; wherein, the critical task means that the free time difference of the corresponding task node is 0, and the non-critical task means that the free time difference of the corresponding task node is greater than 0; Extract all edges between key task nodes and calculate the dependency strength of each edge; the dependency strength is measured by the tightness of the time constraints between key task nodes, and the time constraint degree is represented by the ratio of the task duration to the difference between the earliest start time and the latest start time; Extracting geometric parameters of non-critical task nodes and calculating spatial distances between non-critical task nodes; wherein the spatial distance is represented by the Euclidean distance between spatial position coordinates corresponding to the task nodes; Delete the tasks whose critical task dependency strength is lower than the threshold, and adjust the tasks whose spatial distance is greater than the threshold among the non-critical tasks to obtain an updated directed weighted graph; According to the updated directed weighted graph, through topological sorting, the directed acyclic graph DAG of task collaborative optimization is obtained.

8. The electromechanical construction method based on BIM technology according to claim 5 is characterized in that: According to the directed acyclic graph DAG of task collaborative optimization, the task nodes on the critical path are obtained, including: According to the directed acyclic graph DAG of task collaborative optimization, the earliest start time, the latest start time and the free time difference of all nodes are extracted; Traverse all paths in the directed acyclic graph DAG of task collaborative optimization, and obtain the path with zero free time difference according to the free time difference of the nodes as the key path; Extract all nodes on the critical path and obtain a set of critical task nodes; Calculate the total duration of the critical path as the benchmark duration for mechanical and electrical construction optimization.

9. The electromechanical construction method based on BIM technology according to claim 1, characterized in that: Build a 3D point cloud model of the construction site, including: Use a 3D laser scanner to scan the construction site and obtain a point cloud data set of the construction site; De-noising the point cloud dataset; The ICP registration algorithm is used to perform spatial registration on the denoised point cloud dataset; The point cloud dataset after spatial registration is used as the collected three-dimensional point cloud.

10. A mechanical and electrical construction system based on BIM technology, characterized in that: include: At least one processing unit; used to execute instructions to implement the electromechanical construction method based on BIM technology as described in any one of claims 1 to 9.

Images