Historic building group temporary support system collaborative optimization method and system

Abstract

The application provides a historical building group temporary support system collaborative optimization method and system, relates to the building protection technical field, and comprises the following steps: obtaining point cloud data through three-dimensional laser scanning, performing component identification and force transmission sequence analysis, adopting a spectral clustering algorithm to perform support partitioning, constructing a group action strength matrix based on geometric similarity and structure coupling degree, and optimizing support parameters by using a genetic algorithm. The application realizes the overall collaborative arrangement of the historical building group support system, improves support efficiency and safety, and reduces support cost.

Description

Technical Field

[0001] This invention relates to the field of building preservation technology, and in particular to a collaborative optimization method and system for temporary support systems of historical building complexes. Background Technology

[0002] Historic building complexes, as invaluable cultural heritage, play a vital role in urban renewal and historical preservation. During the protection and restoration of historic building complexes, the design and implementation of temporary support systems are crucial for ensuring building safety and preventing collapse. Traditional temporary support systems for historic building complexes are typically designed using a combination of manual experience and simple calculations, including various forms such as timber supports and steel pipe scaffolding. With the development of technologies such as 3D laser scanning and digital modeling, optimized design of temporary support systems for historic building complexes has become possible. Existing support technologies for historic building complexes are primarily based on individual building support methods. These methods involve structural stress analysis and stability assessment to design appropriate support schemes, ensuring structural safety during the restoration or reinforcement of historic buildings.

[0003] However, existing temporary support technologies for historical building complexes have shortcomings. Traditional support methods for historical building complexes lack consideration for the overall interconnectedness of the complex, often simplifying the complex into independent individual buildings for support design. This ignores the spatial and mechanical relationships between components within the complex, leading to blind spots in the support system design and failing to achieve the desired synergistic support for the entire complex. Existing support methods lack sufficient precision in component identification and stress analysis for historical building complexes, making it difficult to accurately capture the unique complex structural features and damage conditions of historical buildings. This results in deviations between support design and actual needs, potentially wasting support resources and creating safety hazards. Existing support design methods lack systematic optimization tools, often relying on engineers' experience and judgment. This makes it difficult to find the optimal support scheme under multi-objective constraints and adapt to the support needs of large-scale, structurally complex historical building complexes. Especially when dealing with scenarios involving the coordinated action of multiple components, the overall performance of the support system is difficult to guarantee. Summary of the Invention

[0004] This invention provides a collaborative optimization method and system for temporary support systems of historical building complexes, which can solve the problems in the prior art.

[0005] A first aspect of this invention provides a collaborative optimization method for a temporary support system of a historical building complex, comprising:

[0006] Three-dimensional laser scanning was used to obtain multi-site scanning results of the historical building complex, and point cloud data was determined by point cloud registration and stitching.

[0007] Density distribution sampling and component identification are performed on point cloud data. The component types and spatial locations of historical building complexes are analyzed through force transmission sequence analysis to generate a component relationship matrix.

[0008] Based on the component relationship matrix, the spectral clustering algorithm is used to divide the historical building complex into support zones. By calculating the spatial similarity and stress relationship between components, a support correlation matrix is ​​constructed to divide the support groups.

[0009] Based on the geometric similarity index and structural coupling index of the support group, the gray relational analysis method is used to obtain the group action intensity matrix;

[0010] A spatial grid for support layout is constructed based on the group action intensity matrix. An optimization objective function is constructed based on the group action intensity value and action gradient. A genetic algorithm with a secondary mutation mechanism is used to optimize the support parameter values ​​of the support facilities and determine the support layout scheme.

[0011] In one optional embodiment, density distribution sampling and component identification are performed on the point cloud data. The component types and spatial locations of historical building complexes are analyzed through force transmission sequence analysis to generate a component relationship matrix, including:

[0012] A density distribution map is calculated based on point cloud data. The sampling radius of each region is determined according to the density distribution state in the density distribution map. The point cloud data is resampled using the sampling radius to obtain sampled point cloud. Local structural feature values ​​and overall contour feature values ​​of components are extracted hierarchically from the sampled point cloud and then weighted and superimposed to obtain component feature data.

[0013] The component feature data is input into the recognizer to calculate the component type probability value of each sampling point. The component type of the sampling point is determined according to the component type probability value. Spatial clustering is performed on sampling points with the same component type, and sampling points with a spatial distance lower than a preset distance threshold are divided into the same component individual. Boundary fitting is performed on each component individual to obtain the spatial position and shape information of the component.

[0014] Calculate the shortest spatial distance and projected overlap area between adjacent components, generate the distance matrix and overlap matrix between components, determine the support or connection relationship between components, and generate a component connection relationship table;

[0015] Establish a force transmission sequence based on the component connection relationship table, determine the force transmission path between components, and identify the force convergence node components; calculate the interaction strength between components based on the connection relationship table and the force transmission sequence, and generate a component relationship matrix.

[0016] In one optional embodiment, the process includes calculating the shortest spatial distance and projected overlap area between adjacent components, generating a distance matrix and overlap matrix between components, determining the support or connection relationship between components, and generating a component connection relationship table, including:

[0017] Calculate the shortest spatial distance and projected overlap area between adjacent components, generate the distance matrix and overlap matrix between components, and construct the initial component connection graph by treating each component as a node in the graph network.

[0018] Local neighborhood features of each node are extracted, including component type, relative positional relationship, and overlap. Based on the attention mechanism, node features are transmitted and converged within the neighborhood to obtain the fused environment node features.

[0019] The system integrates environmental node features into the policy network, generates connection probability distributions based on spatial relationships between components, calculates reward values ​​based on building static constraints and foundation construction rules, and optimizes connection judgment strategies through gradient updates.

[0020] Based on the optimized policy network, the connection probability value is calculated for each pair of adjacent components in the initial component connection graph. The edge weights are updated according to the connection probability value. The connectivity analysis and stress verification are performed on the updated component connection graph to generate the final component connection relationship table.

[0021] In one optional embodiment, based on the component relationship matrix, a spectral clustering algorithm is used to partition the historical building complex into support zones. By calculating the spatial similarity and stress relationship between components, a support correlation matrix is ​​constructed, and the support groups are divided into:

[0022] The component relationship matrix of the historical building complex is decomposed into eigenvalue array and transformation basis array. Based on the eigenvalue array, the transformation basis vector corresponding to the largest eigenvalue is selected according to a preset number of preferred values ​​to construct a feature mapping space. The component mapping vector is obtained by mapping the components in the component relationship matrix to the feature mapping space.

[0023] Based on the component mapping vector, the component spatial similarity matrix is ​​obtained by Gaussian kernel function operation, and the component force action matrix is ​​calculated based on the force transmission path in the component relationship matrix.

[0024] Based on the component spatial similarity matrix and the component force matrix, a support correlation matrix is ​​generated through weighted fusion. The action intensity matrix is ​​calculated based on the support correlation matrix. A Laplace operator matrix is ​​constructed using the action intensity matrix and the support correlation matrix. A discrete spectrum sequence is obtained by solving the generalized characteristic equation of the Laplace operator matrix. The number of groups is determined based on the discrete spectrum sequence. A mapping basis set of the Laplace operator matrix is ​​selected based on the number of groups. K-means clustering is used to group the mapping basis set to obtain the support unit partitioning result and determine the support groups.

[0025] In an optional embodiment, based on the geometric similarity index and structural coupling index of the support group, the grey relational analysis method is used to obtain the group action intensity matrix, including:

[0026] The spatial centroid coordinates and projected area of ​​the support group are calculated to obtain the centroid distance ratio, boundary overlap, and morphological similarity of the group, which constitute the geometric similarity index. The number of component connections, force transmission paths, and distribution of node components of the support group are statistically analyzed to obtain the connection density index, force transmission coefficient, and proportion of node components, which constitute the structural coupling index. The geometric similarity index and the structural coupling index are normalized according to the index type and value range to form a grey relational factor sequence.

[0027] The maximum values ​​of the indicators in the grey relational factor sequence are used to construct a reference sequence, and the remaining indicator values ​​are used to construct a comparison sequence. The absolute difference sequence between the reference sequence and the comparison sequence is calculated. The absolute difference sequence is partitioned by a two-level resolution coefficient to obtain the correlation interval. Within the correlation interval, the corresponding adaptive weight coefficient is set according to the degree of difference, and the local correlation degree and the overall correlation degree are calculated respectively. The corresponding weighted result is used as the grey relational coefficient between groups.

[0028] For each indicator in the grey relational factor sequence, the initial weight interval is determined by the upper and lower limits of the interval. The maximum deviation method is used to iteratively optimize the initial weight interval to obtain the comprehensive weight of each indicator. The group effect intensity is obtained by weighted fusion of the grey relational coefficient and the corresponding comprehensive weight. The group effect intensity is then used to form a group effect intensity matrix.

[0029] In one optional embodiment, a support layout spatial grid is constructed based on the group action intensity matrix, an optimization objective function is constructed based on the group action intensity values ​​and action gradients, and a genetic algorithm with a secondary mutation mechanism is used to optimize the support parameter values ​​of the support facility, including:

[0030] The group action intensity matrix is ​​transformed into a mesh partitioning weight matrix, a support layout spatial mesh is constructed, the group action intensity value of the mesh node is extracted, and the rate of change between adjacent nodes is calculated to obtain the group action gradient.

[0031] The spacing and dimensions of the support facilities at each grid are collected. The support cost and support response are calculated based on the group action intensity value and action gradient, and the support optimization target is constructed.

[0032] The maximum variation of support parameters between adjacent grids is set based on the difference in the intensity of the action in the group, and the continuous variation range of support parameters is set based on the intensity distribution in the local area to determine the parameter constraints.

[0033] The iterative operation based on parameter constraints includes: constructing chromosome codes for the spacing and dimensions of support facilities at each grid; clustering the chromosome codes based on the group action strength value; storing the chromosome codes within the same cluster in an associated gene pool; selecting chromosome codes within the same associated gene pool for crossover operations to obtain crossover codes; performing mutation operations on the crossover codes; calculating the mutation coefficient based on the group action strength value; adjusting the support parameter values ​​in the chromosome codes according to the mutation coefficient to obtain mutated codes; calculating the parameter similarity between the mutated codes and the chromosome codes in the associated gene pool; triggering a secondary mutation when the parameter similarity is greater than a preset similarity threshold; and repeating the iteration until the support optimization objective is met to determine the final support parameter values.

[0034] A second aspect of the present invention provides a collaborative optimization system for temporary support systems of historical building complexes, comprising:

[0035] The first unit is used to obtain multi-site scanning results of historical building complexes using three-dimensional laser scanning, and to determine point cloud data through point cloud registration and stitching processing.

[0036] The second unit is used to sample the density distribution of point cloud data and identify components. By analyzing the force transmission sequence, it analyzes the component types and spatial locations of historical building complexes and generates a component relationship matrix.

[0037] The third unit is used to partition the support of historical building complexes based on the component relationship matrix and using the spectral clustering algorithm. By calculating the spatial similarity and stress relationship between components, a support correlation matrix is ​​constructed to divide the support groups.

[0038] The fourth unit is used to obtain the group action intensity matrix based on the geometric similarity index and structural coupling index of the support group using the grey relational analysis method.

[0039] The fifth unit is used to construct a support layout spatial grid based on the group action intensity matrix, construct an optimization objective function based on the group action intensity value and action gradient, and use a genetic algorithm with a secondary mutation mechanism to optimize the support parameter values ​​of the support facilities and determine the support layout scheme.

[0040] A third aspect of the present invention provides an electronic device, comprising:

[0041] processor;

[0042] Memory used to store processor-executable instructions;

[0043] The processor is configured to invoke instructions stored in the memory to execute the aforementioned method.

[0044] A fourth aspect of the present invention provides a computer-readable storage medium having stored thereon computer program instructions that, when executed by a processor, implement the aforementioned method.

[0045] In this embodiment of the invention, point cloud data of a building complex is acquired using three-dimensional laser scanning technology. Combined with force transmission sequence analysis and component relationship matrix, accurate identification of the interaction relationships between components in a complex historical building complex is achieved, avoiding the subjectivity and uncertainty in structural relationship judgment in traditional methods. A spectral clustering algorithm is used for support zoning, and a group action intensity matrix is ​​established based on geometric similarity and structural coupling indices. This transforms the support design from traditional single-unit support to collaborative support, fully considering the interaction and integrity between components within the building complex, thus improving the scientific rigor and relevance of the support scheme. A genetic algorithm with a secondary mutation mechanism is used to optimize support parameters, establishing a support layout spatial grid and optimization objective function. This achieves a reasonable layout and parameter optimization of support facilities, not only improving the reliability and safety of temporary support for historical building complexes but also reducing the amount of support materials used, resulting in significant economic and social benefits. Attached Figure Description

[0046] Figure 1 This is a flowchart illustrating the collaborative optimization method for the temporary support system of historical building complexes according to an embodiment of the present invention.

[0047] Figure 2 Optimize the logic flowchart for support parameters. Detailed Implementation

[0048] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0049] The technical solution of the present invention will be described in detail below with reference to specific embodiments. These specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described again in some embodiments.

[0050] Figure 1 This is a flowchart illustrating the collaborative optimization method for temporary support systems of historical building complexes according to an embodiment of the present invention. Figure 1 As shown, the method includes:

[0051] Three-dimensional laser scanning was used to obtain multi-site scanning results of the historical building complex, and point cloud data was determined by point cloud registration and stitching.

[0052] Density distribution sampling and component identification are performed on point cloud data. The component types and spatial locations of historical building complexes are analyzed through force transmission sequence analysis to generate a component relationship matrix.

[0053] Based on the component relationship matrix, the spectral clustering algorithm is used to divide the historical building complex into support zones. By calculating the spatial similarity and stress relationship between components, a support correlation matrix is ​​constructed to divide the support groups.

[0054] Based on the geometric similarity index and structural coupling index of the support group, the gray relational analysis method is used to obtain the group action intensity matrix;

[0055] A spatial grid for support layout is constructed based on the group action intensity matrix. An optimization objective function is constructed based on the group action intensity value and action gradient. A genetic algorithm with a secondary mutation mechanism is used to optimize the support parameter values ​​of the support facilities and determine the support layout scheme.

[0056] In one optional implementation, density distribution sampling and component identification are performed on the point cloud data. The component types and spatial locations of historical building complexes are analyzed through force transmission sequence analysis to generate a component relationship matrix, including:

[0057] A density distribution map is calculated based on point cloud data. The sampling radius of each region is determined according to the density distribution state in the density distribution map. The point cloud data is resampled using the sampling radius to obtain sampled point cloud. Local structural feature values ​​and overall contour feature values ​​of components are extracted hierarchically from the sampled point cloud and then weighted and superimposed to obtain component feature data.

[0058] The component feature data is input into the recognizer to calculate the component type probability value of each sampling point. The component type of the sampling point is determined according to the component type probability value. Spatial clustering is performed on sampling points with the same component type, and sampling points with a spatial distance lower than a preset distance threshold are divided into the same component individual. Boundary fitting is performed on each component individual to obtain the spatial position and shape information of the component.

[0059] Calculate the shortest spatial distance and projected overlap area between adjacent components, generate the distance matrix and overlap matrix between components, determine the support or connection relationship between components, and generate a component connection relationship table;

[0060] Establish a force transmission sequence based on the component connection relationship table, determine the force transmission path between components, and identify the force convergence node components; calculate the interaction strength between components based on the connection relationship table and the force transmission sequence, and generate a component relationship matrix.

[0061] In one specific implementation, based on the acquired point cloud data of a historical building complex, a density distribution map of the point cloud is calculated. The point cloud space is divided into grids, and the number of points contained in each grid is counted to generate a density heatmap. For high-density areas such as main components like columns and beams, a smaller sampling radius (e.g., 0.05 meters) is set; for medium-density areas such as walls and roofs, a medium sampling radius (e.g., 0.1 meters) is set; and for low-density areas such as decorative components, a larger sampling radius (e.g., 0.2 meters) is set. For example, in the point cloud of an ancient building, the point cloud density in the main hall's column area is 500 points / square meter, using a sampling radius of 0.05 meters; the density in the roof tile area is 200 points / square meter, using a sampling radius of 0.1 meters.

[0062] Resampling is performed using a defined sampling radius. For each region, a sphere is created with the sampling radius as its radius. A center point is randomly selected within the sphere as a sampling point, and the feature information of all original point cloud data within the sphere is recorded. In this way, the entire point cloud is transformed into a representative set of sampling points. The number of points in the sampled point cloud is reduced from tens of millions to hundreds of thousands, while retaining key structural information.

[0063] Component features are extracted from the sampling point cloud. Local structural feature values ​​are obtained by calculating the geometric properties of points in the neighborhood of the sampling point, including: normal distribution dispersion, principal curvature value, and surface roughness. Overall contour feature values ​​include: height distribution characteristics, cross-sectional shape characteristics, and longitudinal extension characteristics. These two types of features are weighted and superimposed in a 7:3 ratio to obtain comprehensive component feature data. For example, for a column component sampling point, its local normal dispersion is 0.15, principal curvature value is 0.03, and surface roughness is 0.08; the overall height distribution characteristic is 0.78, and the cross-sectional circularity feature value is 0.92. The comprehensive feature value is the weighted sum of these indicators.

[0064] The component feature data is input into a pre-trained random forest recognizer to calculate the probability value of each sampling point belonging to each component type. The recognizer contains 100 decision trees, each with a depth of 15 layers. For example, if the calculated component type probability for a sampling point is: column (0.85), beam (0.10), bracket (0.03), other (0.02), then the point is determined to belong to a column component. After determining the component type for all sampling points, the DBSCAN algorithm is used to perform spatial clustering on sampling points of the same type. The clustering parameter eps is set to 10% of the average size of the component type, and minPts is set to 5. For example, for column components, the clustering distance threshold is set to 0.1 meters, and column type sampling points with a distance less than 0.1 meters and a number greater than 5 are classified as the same component.

[0065] Boundary fitting is performed on each individual component to obtain its spatial position and shape information. A cylinder fitting algorithm is used for column components, a cuboid fitting algorithm for beam components, and a surface fitting algorithm for roof components. The fitting process employs the RANSAC method, with 1000 iterations and a tolerance threshold of 0.02 meters. For example, the fitting result for a column component is: cylinder, radius 0.25 meters, height 3.6 meters, center point coordinates (12.5, 8.3, 1.8), and verticality 0.98.

[0066] Calculate the spatial relationship between adjacent components. The shortest distance between two components is obtained by calculating the minimum Euclidean distance between the boundary point sets of the two components; the projected overlap area is obtained by calculating the intersection area of ​​the projections of the components on the horizontal and vertical planes. If the distance between two components is less than 0.05 meters and there is projected overlap, a connection relationship is determined. For example, if the shortest distance between a column and a beam is 0.01 meters and the vertical projected overlap area is 0.2 square meters, it is determined to be a support connection relationship and recorded in the component connection relationship table.

[0067] A force transmission sequence is established based on the component connection relationship table. Starting from the ground foundation components, the force transmission path is determined by tracing upwards according to the support relationships. If a component is supported by multiple components, the force distribution ratio is calculated based on the support area ratio; if a component supports multiple components, that component is a force transmission node. For example, in an ancient building, the foundation supports columns, the columns support beams, the beams support rafters, and the rafters support roof tiles, forming a complete force transmission sequence. Components at force convergence points are identified, such as column heads, beam-frame junctions, and bracket sets.

[0068] The strength of the interaction relationships between structural members is calculated to generate a member relationship matrix. The relationship strength comprehensively considers three factors: the proportion of support area, the positional relationship in the force transmission sequence, and the importance weight of the members. For example, the support relationship strength between a column and a beam is 0.85, and the support relationship strength between a beam and a purlin is 0.72. The member relationship matrix is ​​represented as an n×n matrix, where n is the total number of members, and the matrix element values ​​represent the relationship strength between members, ranging from 0 to 1, where 0 indicates no relationship and 1 indicates the strongest relationship. This matrix can be used for the structural stability analysis of historical buildings, the formulation of restoration plans, and the assessment of cultural value.

[0069] In one optional implementation, the shortest spatial distance and projected overlap area between adjacent components are calculated to generate a distance matrix and overlap matrix between components. The support or connection relationship between components is determined, and a component connection relationship table is generated, including:

[0070] Calculate the shortest spatial distance and projected overlap area between adjacent components, generate the distance matrix and overlap matrix between components, and construct the initial component connection graph by treating each component as a node in the graph network.

[0071] Local neighborhood features of each node are extracted, including component type, relative positional relationship, and overlap. Based on the attention mechanism, node features are transmitted and converged within the neighborhood to obtain the fused environment node features.

[0072] The system integrates environmental node features into the policy network, generates connection probability distributions based on spatial relationships between components, calculates reward values ​​based on building static constraints and foundation construction rules, and optimizes connection judgment strategies through gradient updates.

[0073] Based on the optimized policy network, the connection probability value is calculated for each pair of adjacent components in the initial component connection graph. The edge weights are updated according to the connection probability value. The connectivity analysis and stress verification are performed on the updated component connection graph to generate the final component connection relationship table.

[0074] In one specific implementation, when constructing the initial component connection diagram, the shortest spatial distance and projected overlap area between all adjacent components are calculated. For two components A and B, the shortest spatial distance is calculated using an iterative nearest-point algorithm. This algorithm first uniformly samples a point cloud set on the surface of each component, sampling 500-1000 points on each component surface. For each point in component A, the nearest point in component B is searched, and the Euclidean distance between the two points is calculated. The minimum distance value among all point pairs is selected as the shortest spatial distance between the two components. In practical applications, when the calculated shortest spatial distance is less than a preset threshold (e.g., 20 cm), the two components are considered spatially adjacent. The projected overlap area is obtained by projecting the components onto horizontal and vertical planes and calculating the intersection area of ​​their projected polygons. The projection uses orthogonal projection, projecting the three-dimensional geometry of the components onto the XY and XZ planes. A polygon clipping algorithm is used to calculate the intersection area of ​​the two projected polygons, and this is divided by the projected area of ​​the smaller component to obtain the normalized overlap. For example, if the projected areas of two components are 2 square meters and 3 square meters respectively, and their intersection area is 1 square meter, then the overlap is 1 / 2 = 0.5.

[0075] Based on the calculated spatial distance and overlap matrices, an initial component connection graph is constructed. Each component is treated as a node in the graph. When the spatial distance between two components is less than a threshold (e.g., 20 cm) or the overlap is greater than a threshold (e.g., 0.3), a connection edge is established between the corresponding nodes. The initial weight of the edge is set to 1.0, representing the initial estimate of the connection probability. For example, when a beam and a column are detected to be 15 cm apart with an overlap of 0.4, an edge with a weight of 1.0 is established between these two component nodes.

[0076] When extracting local neighborhood features for each node, for node i, first obtain the set Ni of all nodes in its first-order neighborhood. Node features include component type (column, beam, wall, floor slab, etc., encoded as one-hot vectors), component geometry (length, width, height, normalized), and component material properties (density, strength, etc., normalized). For each pair of nodes i and j in the neighborhood, extract relative positional relationship features, including the relative coordinate difference between the center points of the two components, the relative direction vector, the shortest spatial distance, and the overlap. For example, the node features of a wooden column can be represented as: type [1, 0, 0, 0], size [0.3, 0.3, 3.0], material [0.6, 0.5], and the relative positional relationship with the adjacent wooden beam is: relative coordinate difference [0.1, 0.2, 1.5], relative direction vector [0, 0, 1], spatial distance 0.15, and overlap 0.4.

[0077] When transmitting and converging node features within a neighborhood based on the attention mechanism, for each node i, the attention weight between it and each node j in the neighborhood is calculated. The attention weight calculation is based on the similarity between node features and edge features, using the dot product similarity method. The features of node i are weighted and summed with the features of each node in the neighborhood to obtain the fused environment node features. The weight coefficients are dynamically adjusted based on the spatial distance and overlap between two nodes; neighboring nodes with closer distance and higher overlap receive higher weights. For example, when the central node is a wooden pillar and there are 3 nodes in the neighborhood (wooden beam, stone wall, and wooden plank), if the wooden beam is closest to the wooden pillar and has the highest overlap, then the wooden beam node is given a higher weight when fusing features.

[0078] When inputting the fused environment node features into the policy network, a multilayer perceptron structure is employed, containing three fully connected layers with 128 and 64 hidden neurons, respectively, and using ReLU activation. The network input is the fused environment feature vector of the node (dimension 256), and the output is a binary classification probability value representing the probability of establishing a connection between two components. A value network is also designed to evaluate the value of the current state, reducing the variance of the policy gradient. During training, for each pair of adjacent components, the probability distribution output by the policy network is sampled to determine whether to establish a connection. For example, if the policy network outputs a connection probability of 0.85 between a wooden pillar and a wooden beam, there is an 85% probability that a connection will be established between these two components.

[0079] When calculating the reward value based on building static constraints and foundation construction rules, the structural stability, material efficiency, and construction feasibility of the support system are considered. Structural stability is assessed using a simplified mechanical model to check whether key support points meet the load-bearing requirements. For example, when the support system needs to withstand a 10 kN lateral load, the support capacity of each support point is checked to ensure it exceeds the design load. Material efficiency is assessed by calculating the total number and volume of connecting components, encouraging the use of the fewest components possible while meeting support requirements. Construction feasibility considers connection complexity and operational space; for example, connections requiring large equipment installations are avoided in confined spaces. Taking all these factors into account, the reward value for the current connection scheme is calculated. The reward value ranges from -1 to +1, with higher values ​​indicating a better scheme. For example, a scheme that meets load-bearing requirements, uses less material, and has moderate construction difficulty might have a reward value of 0.8.

[0080] When optimizing the connection decision policy through gradient updates, the loss function is calculated using the policy gradient method based on the calculated reward value and the output probability of the policy network. The loss function consists of two parts: policy gradient loss and value function loss. The Adam optimizer is used for parameter updates, with a learning rate of 0.001. During training, multiple samples are generated for different historical building support scenarios, each containing 50-200 components. The training iterations are 5000 times, with model performance evaluated every 100 iterations. The model is considered converged when the average reward value changes by less than 1% over five consecutive evaluations. The optimized policy network can output reasonable connection probabilities for given spatial relationships between components.

[0081] Based on the optimized strategy network, a connection probability value is calculated for each pair of adjacent components in the initial component connection diagram. A threshold (e.g., 0.7) is used to determine whether to establish a connection. Connectivity analysis is performed on the updated component connection diagram to ensure effective support paths exist between critical components. Stress verification checks the stress state of each connection point to ensure it does not exceed material strength limits. Finally, a component connection relationship table is generated, containing information such as the ID, connection type, connection location, and connection strength of each pair of connected components. For example, components with IDs A001 and B002 have a connection type of "mortise and tenon joint," a connection location of "(x=1.2, y=0.8, z=2.5)," and a connection strength of "medium."

[0082] In existing technologies, the design of temporary support systems for historical building complexes mainly relies on manual experience and lacks systematic optimization methods. Traditional methods typically use pre-set support templates, which cannot fully adapt to the special structural characteristics of different historical buildings, leading to material waste and structural safety hazards. While existing computer-aided design methods can provide 3D modeling support, they lack intelligent connection relationship inference capabilities and cannot automatically generate optimal support schemes. The method in this embodiment overcomes the limitations of traditional methods by introducing deep reinforcement learning technology, which automatically learns component connection strategies based on the spatial relationships and mechanical properties between components. Compared with existing technologies, this method can adaptively generate connection relationships according to the structural characteristics of the building, reduce the amount of support material used, improve the stability of the support structure, and shorten the design time, significantly improving the design efficiency and quality of temporary support systems for historical building complexes.

[0083] In one optional implementation, based on the component relationship matrix, a spectral clustering algorithm is used to partition the historical building complex into support zones. By calculating the spatial similarity and stress relationship between components, a support correlation matrix is ​​constructed, and the support groups are divided into:

[0084] The component relationship matrix of the historical building complex is decomposed into eigenvalue array and transformation basis array. Based on the eigenvalue array, the transformation basis vector corresponding to the largest eigenvalue is selected according to a preset number of preferred values ​​to construct a feature mapping space. The component mapping vector is obtained by mapping the components in the component relationship matrix to the feature mapping space.

[0085] Based on the component mapping vector, the component spatial similarity matrix is ​​obtained by Gaussian kernel function operation, and the component force action matrix is ​​calculated based on the force transmission path in the component relationship matrix.

[0086] Based on the component spatial similarity matrix and the component force matrix, a support correlation matrix is ​​generated through weighted fusion. The action intensity matrix is ​​calculated based on the support correlation matrix. A Laplace operator matrix is ​​constructed using the action intensity matrix and the support correlation matrix. A discrete spectrum sequence is obtained by solving the generalized characteristic equation of the Laplace operator matrix. The number of groups is determined based on the discrete spectrum sequence. A mapping basis set of the Laplace operator matrix is ​​selected based on the number of groups. K-means clustering is used to group the mapping basis set to obtain the support unit partitioning result and determine the support groups.

[0087] In one specific implementation, when performing eigenvalue decomposition on the component relationship matrix of a historical building complex, the component relationship matrix is ​​an N×N square matrix, where N represents the total number of components. Each element in the matrix represents the strength of the relationship between two components, including comprehensive indicators such as spatial distance, material similarity, and force transmission relationship. Taking a historical building complex containing 120 components as an example, the constructed component relationship matrix is ​​120×120. The singular value decomposition method is used to perform eigenvalue decomposition on the component relationship matrix, resulting in an eigenvalue array and a transformation basis array. The eigenvalue array is a vector of length N, arranged in descending order, representing the importance of different feature dimensions. The transformation basis array is an N×N matrix, where each column represents a transformation basis vector, indicating the transformation direction from the original space to the feature space. In practice, when the relation matrix composed of 120 components is subjected to eigenvalue decomposition, 120 eigenvalues ​​are obtained. The first 10 eigenvalues ​​are: 12.5, 9.8, 7.6, 6.2, 5.1, 4.3, 3.5, 2.9, 2.4, and 2.0. These values ​​reflect the variance of the data distribution in different dimensions.

[0088] Based on the eigenvalue array, a feature mapping space is constructed by selecting the transformation basis vector corresponding to the largest eigenvalue according to a preset number of preferred eigenvalues. The preset number of preferred eigenvalues ​​can be determined by the energy retention rate, and usually the top k eigenvalues ​​whose cumulative energy accounts for more than 90% of the total energy are selected. In the example above, the transformation basis vectors corresponding to the top 5 eigenvalues ​​are selected to construct the feature mapping space because the cumulative energy of these 5 eigenvalues ​​accounts for 91.3% of the total energy. In this way, the original 120-dimensional component space is mapped to a 5-dimensional feature space. The component mapping vector is obtained by mapping the components in the component relation matrix to the feature mapping space. For each component, the 5-dimensional representation vector of the component in the feature mapping space is obtained by performing an inner product operation with the selected 5 transformation basis vectors. For example, the component with component ID A015 has the representation vector [0.52, -0.31, 0.28, 0.15, -0.08] in the feature mapping space.

[0089] Based on the component mapping vectors, a component spatial similarity matrix is ​​obtained through Gaussian kernel function operation. The Gaussian kernel function adopts the form of radial basis function, and the kernel width parameter is set to the average value of the Euclidean distance between the component mapping vectors. For any two components i and j, the Euclidean distance between their mapping vectors is calculated, and the similarity value is calculated by substituting it into the Gaussian kernel function. This value ranges from 0 to 1, and the larger the value, the more similar the two components are in the feature space. For example, the mapping vectors of components A015 and B027 are [0.52, -0.31, 0.28, 0.15, -0.08] and [0.48, -0.29, 0.25, 0.18, -0.10], respectively, and the calculated similarity value is 0.92, indicating that the two components are very similar in the feature space.

[0090] The component force action matrix is ​​calculated based on the force transfer path in the component relationship matrix. The force transfer path is determined by analyzing the contact relationship between components and the load transfer direction. Using a force flow analysis method, starting from the load application point, the direction and magnitude of force transfer are traced step by step along the component connection relationship. For each pair of components i and j, if a direct force transfer relationship exists, the corresponding element value in the component force action matrix is ​​set as the force transfer intensity; if no direct force transfer relationship exists, it is set to 0. The force transfer intensity is comprehensively evaluated based on the connection type, contact area, and load magnitude, and ranges between 0 and 1. For example, when a wooden beam A is directly supported on a stone column B, the force transfer intensity between them may be 0.85; while when the two components are only connected by secondary supports, the force transfer intensity may be only 0.32.

[0091] Based on the spatial similarity matrix and the force-feeding matrix of the components, a support correlation matrix is ​​generated through weighted fusion. The weighted fusion uses a linear combination method, with weight coefficients adjusted according to support requirements. In the temporary support design of historical building complexes, force transmission relationships are usually more important than spatial similarity; therefore, the weight coefficients can be set to 0.7 for the force-feeding matrix and 0.3 for the spatial similarity matrix. For each pair of components i and j, the element value in the support correlation matrix is ​​equal to 0.7 multiplied by the corresponding element value in the force-feeding matrix plus 0.3 multiplied by the corresponding element value in the spatial similarity matrix. For example, if the force intensity of two components is 0.85 and the spatial similarity is 0.92, then the calculated support correlation is 0.7 × 0.85 + 0.3 × 0.92 = 0.871.

[0092] The action intensity matrix is ​​calculated based on the support correlation matrix. The action intensity matrix represents the importance or centrality of each component within the entire building complex. The action intensity value of each component is obtained by summing the elements in each row of the support correlation matrix. A higher action intensity value indicates a stronger support correlation between the component and other components, and thus a greater importance within the support system. In practical applications, the action intensity values ​​are normalized to a range between 0 and 1. For example, a calculated action intensity value of 0.78 for component C103 indicates that this component has high importance within the entire support system.

[0093] A Laplace operator matrix is ​​constructed using the action intensity matrix and the support correlation matrix. The Laplace operator matrix is ​​an important tool in graph theory for representing graph structures and is used in spectral clustering analysis. The construction process involves two steps: first, calculating the degree matrix, which is a diagonal matrix composed of the diagonal elements of the action intensity matrix; then, subtracting the support correlation matrix from the degree matrix to obtain the Laplace operator matrix. The Laplace operator matrix is ​​characterized by positive diagonal elements, negative or zero off-diagonal elements, and a zero sum of elements in each row. This matrix structure effectively captures the support relationship structure between components.

[0094] The discrete spectral sequence is obtained by solving the generalized characteristic equation of the Laplacian operator matrix. The generalized characteristic equation is solved using either the power iteration method or the QR decomposition method. The calculated eigenvalues ​​constitute the discrete spectral sequence, sorted in ascending order. The size of the eigenvalue intervals in the discrete spectral sequence reflects the clustering structure of the data. The maximum interval is found in the spectral sequence; the number of eigenvalues ​​preceding this interval indicates the appropriate number of groups. For example, the first 10 calculated eigenvalues ​​are: 0.02, 0.05, 0.09, 0.15, 0.42, 0.61, 0.79, 0.88, 0.95, 1.02. The interval between the 4th and 5th eigenvalues ​​is the largest (0.42 - 0.15 = 0.27), therefore the number of groups is determined to be 4.

[0095] The number of groups is determined based on the discrete spectral sequence, and a mapping basis set for the Laplacian operator matrix is ​​selected according to the number of groups. The eigenvectors corresponding to the k smallest eigenvalues ​​of the Laplacian operator matrix (where k is the determined number of groups) are selected to form the mapping basis set. The dimension of each eigenvector is the number of components N, and the dimension of the mapping basis set is N×k. In the example above, the eigenvectors corresponding to the first four eigenvalues ​​are selected to form a 120×4 mapping basis set matrix.

[0096] K-means clustering is used to group the mapped basis set, resulting in the support unit partitioning and determining the support groups. Each row of the mapped basis set is treated as a sample point, and K-means clustering is performed on these sample points, with k cluster centers. The clustering process uses Euclidean distance as a metric and iteratively optimizes until convergence. The clustering result is the component grouping result, with each group corresponding to one support unit. In practical applications, for a historical building complex containing 120 components, four support units may be obtained, with 32, 28, 35, and 25 components respectively. The components within each support unit have high similarity in spatial distribution and stress transmission, making it suitable to use the same or similar support schemes.

[0097] The above methods can automatically divide the components in the historical building complex into reasonable support unit groups, providing a scientific basis for the design and implementation of temporary support systems, improving support efficiency, reducing support costs, and ensuring the structural safety of the historical building complex during the repair and protection process.

[0098] In one optional implementation, based on the geometric similarity index and structural coupling index of the support group, the grey relational analysis method is used to obtain the group action intensity matrix, including:

[0099] The spatial centroid coordinates and projected area of ​​the support group are calculated to obtain the centroid distance ratio, boundary overlap, and morphological similarity of the group, which constitute the geometric similarity index. The number of component connections, force transmission paths, and distribution of node components of the support group are statistically analyzed to obtain the connection density index, force transmission coefficient, and proportion of node components, which constitute the structural coupling index. The geometric similarity index and the structural coupling index are normalized according to the index type and value range to form a grey relational factor sequence.

[0100] The maximum values ​​of the indicators in the grey relational factor sequence are used to construct a reference sequence, and the remaining indicator values ​​are used to construct a comparison sequence. The absolute difference sequence between the reference sequence and the comparison sequence is calculated. The absolute difference sequence is partitioned by a two-level resolution coefficient to obtain the correlation interval. Within the correlation interval, the corresponding adaptive weight coefficient is set according to the degree of difference, and the local correlation degree and the overall correlation degree are calculated respectively. The corresponding weighted result is used as the grey relational coefficient between groups.

[0101] For each indicator in the grey relational factor sequence, the initial weight interval is determined by the upper and lower limits of the interval. The maximum deviation method is used to iteratively optimize the initial weight interval to obtain the comprehensive weight of each indicator. The group effect intensity is obtained by weighted fusion of the grey relational coefficient and the corresponding comprehensive weight. The group effect intensity is then used to form a group effect intensity matrix.

[0102] In one specific implementation, when calculating the spatial centroid coordinates of a support group, the arithmetic mean of the coordinates of all components within each group is taken. Taking an ancient architectural complex as an example, it is divided into four support groups. Group A contains 32 components. By calculating the average of the three-dimensional coordinates of these 32 components, the spatial centroid coordinates of Group A are obtained as (15.6, 23.2, 8.5) meters. Similarly, the spatial centroid coordinates of Groups B, C, and D are calculated to be (18.9, 25.7, 8.2) meters, (12.3, 22.8, 7.9) meters, and (16.5, 19.6, 8.7) meters, respectively. The centroid distance ratio of groups is obtained by calculating the Euclidean distance between the centroids of any two groups and dividing it by the maximum size of the entire architectural complex. For example, the centroid distance between Group A and Group B is 4.1 meters, and the maximum size of the architectural complex is 42 meters. Therefore, the centroid distance ratio of Group A and Group B is 0.098. The projected area is calculated by projecting all components within a group onto the horizontal and vertical planes and then calculating their convex hull areas. Group A has a projected area of ​​320 square meters on the horizontal plane and 156 square meters on the vertical plane. Boundary overlap is calculated as the ratio of the intersection area of ​​the projected convex hulls of two groups to the area of ​​the smaller convex hull. The intersection area of ​​the projected convex hulls of groups A and B on the horizontal plane is 85 square meters, and the projected area of ​​group B is 285 square meters. Therefore, the boundary overlap of group AB on the horizontal plane is 0.298. Morphological similarity is calculated by comparing the aspect ratio, orientation, and compactness of the projected shapes of two groups. Group A has an aspect ratio of 1.8, an orientation of 76 degrees, and a compactness of 0.72; group B has an aspect ratio of 1.6, an orientation of 82 degrees, and a compactness of 0.68. The calculated morphological similarity of group AB is 0.85.

[0103] When calculating the number of component connections in a support group, the number of connections between components within the group and between components in other groups are calculated. Group A has 127 component connections, 23 connections with Group B, 18 connections with Group C, and 9 connections with Group D. The connection density index is the ratio of the number of connections between groups to the total number of components in both groups. Group A has 32 components, Group B has 28 components, and there are 23 connections between the two groups. Therefore, the connection density index for group AB is 23 / (32+28) = 0.383. The force transmission path is the number of main force transmission channels between the two groups. Through mechanical analysis, 5 main force transmission paths are identified between groups A and B. The total number of components in the groups is 60, and the force transmission coefficient for group AB is calculated to be 5 / 60 = 0.083. Node components refer to the key components connecting different groups; the proportion of node components is the ratio of the number of node components to the total number of components in the group. There are 7 node components between groups A and B, so the node component ratio is 7 / 60 = 0.117.

[0104] When normalizing the geometric similarity index and the structural coupling index according to their index type and value range, the range standardization method is used. For positive indices (the larger the value, the better), the normalization formula is: current value minus minimum value, then divided by the difference between the maximum and minimum values. For negative indices (the smaller the value, the better), the normalization formula is: maximum value minus current value, then divided by the difference between the maximum and minimum values. Taking the group centroid distance ratio as an example, this is a negative index. The maximum distance ratio between all group pairs is 0.215, and the minimum is 0.098. The AB group distance ratio is 0.098, and after normalization, it is (0.215-0.098) / (0.215-0.098)=1. After normalization, the value range of all indices is unified to between 0 and 1, forming a grey relational factor sequence.

[0105] When constructing a reference sequence by maximizing the values ​​of indicators in the grey relational factor sequence, for each indicator type, the maximum value of that indicator across all group pairs is selected. For example, among the normalized indicators, the maximum value of the group centroid distance ratio is 1 (group AB), the maximum value of the boundary overlap is 0.95 (group BC), the maximum value of the morphological similarity is 0.92 (group CD), the maximum value of the connection density index is 0.89 (group AC), the maximum value of the force transmission coefficient is 0.78 (group BD), and the maximum value of the node component proportion is 0.83 (group AD). These maximum values ​​constitute the reference sequence [1, 0.95, 0.92, 0.89, 0.78, 0.83]. The indicator values ​​of the remaining group pairs constitute a comparison sequence, such as the comparison sequence for group AC being [0.62, 0.75, 0.81, 0.89, 0.56, 0.71]. Calculate the absolute difference sequence between the reference sequence and the comparison sequence to obtain the absolute difference sequence of group AC as [0.38, 0.20, 0.11, 0, 0.22, 0.12].

[0106] The absolute difference sequence was partitioned using two-level resolution coefficients to obtain correlation intervals. The two-level resolution coefficients were set to 0.1 and 0.5, respectively, dividing the absolute difference sequence into three intervals: a strongly correlated interval [0, 0.1], a moderately correlated interval (0.1, 0.5], and a weakly correlated interval (0.5, 1). For the absolute difference sequence of group AC [0.38, 0.20, 0.11, 0, 0.22, 0.12], the interval division results were as follows: the connection density index (0) belonged to the strongly correlated interval; morphological similarity (0.11), node component ratio (0.12), boundary overlap (0.20), and force transmission coefficient (0.22) belonged to the moderately correlated interval; and the group centroid distance ratio (0.38) belonged to the moderately correlated interval. Within each correlation interval, an adaptive weight coefficient is set according to the degree of difference. The weight coefficient for a strong correlation interval is 0.6; within a moderately correlated interval, the smaller the difference, the greater the weight (calculated as 0.5 minus the difference value); and the weight coefficient for a weakly correlated interval is 0.1. Based on this, the local correlation degree of each indicator in the AC group is calculated as [0.12, 0.30, 0.39, 0.6, 0.28, 0.38]. The overall correlation degree is the weighted average of the local correlation degrees, and the calculated overall correlation degree of the AC group is 0.345. The overall correlation degree of all group pairs is used as the grey correlation coefficient between groups, forming a matrix of group number × group number.

[0107] For each indicator in the grey relational factor sequence, the initial weight interval is determined by upper and lower limit constraints. Based on expert experience and historical data analysis, the initial weight intervals for each indicator are set as follows: group centroid distance ratio [0.10, 0.25], boundary overlap [0.15, 0.30], morphological similarity [0.10, 0.20], connection density index [0.15, 0.30], force transmission coefficient [0.20, 0.40], and node component proportion [0.10, 0.25]. The maximum deviation method is used to iteratively optimize the initial weight intervals to obtain the comprehensive weight of each indicator. The maximum deviation method determines the optimal weight by maximizing the information gain between indicators. After 50 iterations, the comprehensive weights of each indicator are obtained as follows: group centroid distance ratio 0.18, boundary overlap 0.22, morphological similarity 0.15, connection density index 0.17, force transmission coefficient 0.32, node component proportion 0.16, and the sum of all weights is 1. The group effect strength is obtained by weighted fusion of the grey relational coefficient and the corresponding comprehensive weight. For example, the grey relational coefficient of group AC is 0.345, and the effect strength of group AC is 0.312 after weighted fusion with the weights of each indicator. The effect strengths of all group pairs are combined into a group effect strength matrix. This matrix is ​​a symmetric matrix with diagonal elements of 1, representing the effect strength of the group itself.

[0108] Traditional methods struggle to accurately grasp the synergistic relationships between different support units, leading to unstable support effects, unreasonable resource allocation, and an inability to meet the support needs of large-scale or complex historical building complexes. While existing computer-aided support design technologies incorporate numerical simulation and optimization algorithms, they primarily focus on the stress analysis of individual support units, lacking the ability to optimize the system from an overall synergistic perspective. This embodiment proposes a multi-dimensional quantitative evaluation method for the relationships between support groups based on grey relational analysis theory. It comprehensively captures the spatial relationships and mechanical connections between groups through two major index systems: geometric similarity and structural coupling degree. This embodiment establishes a comprehensive evaluation system including six key indicators such as the group centroid distance ratio and boundary overlap, making the quantification of support relationships more comprehensive and accurate. It introduces a two-level resolution coefficient and an adaptive weight calculation method to enhance the ability of grey relational analysis to distinguish different degrees of correlation. The maximum deviation method is used to dynamically optimize the index weights, improving the objectivity and adaptability of the evaluation results.

[0109] In one optional implementation, a support layout spatial grid is constructed based on the group action intensity matrix, an optimization objective function is constructed based on the group action intensity values ​​and action gradients, and a genetic algorithm with a secondary mutation mechanism is used to optimize the support parameters of the support facility, including:

[0110] The group action intensity matrix is ​​transformed into a mesh partitioning weight matrix, a support layout spatial mesh is constructed, the group action intensity value of the mesh node is extracted, and the rate of change between adjacent nodes is calculated to obtain the group action gradient.

[0111] The spacing and dimensions of the support facilities at each grid are collected. The support cost and support response are calculated based on the group action intensity value and action gradient, and the support optimization target is constructed.

[0112] The maximum variation of support parameters between adjacent grids is set based on the difference in the intensity of the action in the group, and the continuous variation range of support parameters is set based on the intensity distribution in the local area to determine the parameter constraints.

[0113] The iterative operation based on parameter constraints includes: constructing chromosome codes for the spacing and dimensions of support facilities at each grid; clustering the chromosome codes based on the group action strength value; storing the chromosome codes within the same cluster in an associated gene pool; selecting chromosome codes within the same associated gene pool for crossover operations to obtain crossover codes; performing mutation operations on the crossover codes; calculating the mutation coefficient based on the group action strength value; adjusting the support parameter values ​​in the chromosome codes according to the mutation coefficient to obtain mutated codes; calculating the parameter similarity between the mutated codes and the chromosome codes in the associated gene pool; triggering a secondary mutation when the parameter similarity is greater than a preset similarity threshold; and repeating the iteration until the support optimization objective is met to determine the final support parameter values.

[0114] In one specific implementation, when converting the group action intensity matrix into a grid partitioning weight matrix, the spatial extent of the historical building complex is divided into grids. The grid size is determined based on the scale and structural complexity of the building complex, typically ranging from 0.5 to 2 meters. For a medium-sized historical building complex with an area of ​​approximately 5,000 square meters, a 1-meter × 1-meter grid can be used, generating approximately 5,000 grid cells. Each grid cell corresponds to a region in the planar space of the building complex, and vertically, it can be divided into 3-5 levels based on the building height. The values ​​in the group action intensity matrix are mapped to each grid cell, and the mapping weight is determined based on the spatial relationship between the grid location and each support group. For example, when a grid is located at the boundary of two support groups, the group action intensity value of that grid is a weighted average of the action intensities of the two groups, with the weight inversely proportional to the distance from the grid to the center of the group. A grid cell (25, 32) is located at the boundary between support groups A and B, 5 meters from the center of group A and 7 meters from the center of group B. The action intensity of group AB is 0.83. Therefore, the group action intensity value of this grid cell is (7 / (5+7))×0.83=0.48. After constructing the support layout spatial grid, the group action intensity values ​​of the grid nodes are extracted, and the rate of change between adjacent nodes is calculated to obtain the group action gradient. The action gradient is the difference in action intensity between adjacent grid nodes in the horizontal and vertical directions divided by the grid spacing. For example, if the action intensity of grid (25, 32) is 0.48, the action intensity of the adjacent grid (26, 32) is 0.52, and the grid spacing is 1 meter, then the action gradient in the X direction is (0.52-0.48) / 1=0.04.

[0115] When collecting the layout spacing and specifications of the support facilities at each grid, the layout spacing refers to the spacing between structural components such as support columns and support beams, and the specifications include parameters such as the cross-sectional dimensions and material thickness of the support components. These initial values ​​can be set based on industry standards and expert experience; for example, the initial spacing of wooden support columns can be set to 0.8 meters, with a cross-sectional dimension of 10 cm × 10 cm. Support costs and support responses are calculated based on the group action intensity value and action gradient, respectively, to construct a support optimization objective. Support costs are related to the amount of support materials used and the difficulty of construction. The calculation formula is the unit price of support materials multiplied by the amount used, plus the construction cost. For example, if the unit price of wooden support materials is 2000 yuan / cubic meter, and a certain grid area requires 10 columns, each with a volume of 0.01 cubic meters, and the construction cost is 500 yuan, then the support cost for this area is 2000 × 10 × 0.01 + 500 = 700 yuan. Support response represents the load-bearing capacity of a support system for structural loads and is related to the specifications, dimensions, spacing, and material strength of the support components. Areas with higher load intensity require a higher support response, while areas with larger load gradients need to consider a gradual transition in support response. The goal of support optimization is to minimize the total support cost while meeting support response requirements.

[0116] When setting the maximum variation of support parameters between adjacent grids based on the difference in group action intensity, the continuity of the support structure and construction feasibility are considered. When the difference in group action intensity between adjacent grids is less than 0.1, the maximum variation in the spacing between support columns is 0.1 meters, and the maximum variation in the cross-sectional dimensions of support components is 1 centimeter. When the difference is between 0.1 and 0.3, the maximum variations are 0.2 meters and 2 centimeters, respectively. When the difference is greater than 0.3, the maximum variations are 0.3 meters and 3 centimeters, respectively. The continuous variation range of support parameters is set based on the distribution of action intensity within a local area to ensure the coordination of the overall support system. Within a 5×5 meter local area, the continuous variation range of the spacing between support columns does not exceed 0.5 meters, and the continuous variation range of the cross-sectional dimensions of support components does not exceed 5 centimeters. These parameter constraints ensure both the structural stability of the support system and the feasibility of construction.

[0117] When performing iterative operations based on parameter constraints, the spacing and dimensions of the support facilities at each grid are constructed as chromosome codes. Chromosome codes use real-number encoding, with each gene representing a support parameter, such as the spacing between support columns, the width of the column cross-section, the height of the column cross-section, and the spacing between support beams. For 5000 grid cells and 3 support parameters, the chromosome length is 15000. The chromosome codes are clustered based on the group action intensity values, grouping grid regions with similar group action intensity values ​​into the same cluster. The K-means clustering algorithm is used, and the number of clusters is determined based on the distribution characteristics of the group action intensity, typically 3-7. For a historical building complex, the action intensity is divided into 5 levels: 0-0.2, 0.2-0.4, 0.4-0.6, 0.6-0.8, and 0.8-1.0, corresponding to 5 clusters. Chromosome codes within the same cluster are stored in an associated gene pool, forming 5 associated gene pools. Chromosome codes within the same associated gene pool are selected for cross-validation to obtain cross-coded codes. The crossover operation uses an arithmetic crossover method, randomly selecting two chromosome codes from the same gene pool as the parents. The gene values ​​in the generated offspring chromosomes are a weighted average of the corresponding gene values ​​of the parents. The weight can be set to 0.5, meaning the offspring gene values ​​are the average of the parent gene values. For example, if a gene value of parent chromosome 1 is 0.8 meters (column spacing), and the corresponding gene value of parent chromosome 2 is 1.0 meters, then the corresponding gene value of the offspring chromosome will be 0.9 meters.

[0118] When performing mutation operations on cross-coding, the coefficient of variation is calculated based on the group action strength value. The coefficient of variation is directly proportional to the group action strength value; the higher the action strength value, the smaller the coefficient of variation, ensuring more stable support parameters in high-action-strength areas. The formula for calculating the coefficient of variation is the base mutation rate (e.g., 0.05) multiplied by (1 - group action strength value). For example, if the group action strength value of a grid is 0.7, then the coefficient of variation for the corresponding gene in that grid is 0.05 × (1 - 0.7) = 0.015. The support parameter values ​​in the chromosome coding are adjusted according to the coefficient of variation to obtain the mutated code. The mutation method uses Gaussian mutation, which adds a random number following a Gaussian distribution to the original gene value. The standard deviation of the random number is the original gene value multiplied by the coefficient of variation. For example, if the original gene value is 0.9 meters and the coefficient of variation is 0.015, then the mutated gene value may become 0.913 meters. The similarity between the mutated code and the chromosome code in the associated gene pool is calculated. The similarity calculation is based on Euclidean distance, treating the two chromosome codes as two points in a multidimensional space and calculating the normalized distance between them. When the parameter similarity is greater than a preset similarity threshold (e.g., 0.95), it indicates that the mutation effect is not significant, triggering a secondary mutation by increasing the mutation coefficient (e.g., doubling it) and repeating the mutation operation. This process continues until the parameter similarity is less than the threshold or the maximum number of mutations is reached, at which point the mutation result is determined.

[0119] The iterative process is repeated until the support optimization objective is met or the maximum number of iterations (e.g., 200) is reached. In each iteration, the support cost and response of the current solution are evaluated, compared with the optimization objective, and the optimal solution is updated. As iterations proceed, individuals in the population gradually move closer to the optimal solution. The final determined support parameter values ​​are used to guide the design and construction of the actual support system. For example, for a key protected area of ​​a historical building complex, the optimization results show that the spacing between support columns should be 0.6 meters, and the column cross-section size should be 12 cm × 12 cm; while for areas with less stress, the column spacing can be increased to 1.2 meters, and the cross-section size can be reduced to 8 cm × 8 cm, significantly reducing the support cost while ensuring safety.

[0120] Traditional support design typically employs uniform specifications and layout, failing to adequately consider the stress differences and structural characteristics of different areas, leading to wasted support materials or insufficient local support. Existing computer-aided optimization methods are mostly based on finite element analysis, which is computationally complex and struggles to handle the overall collaborative relationships of large-scale historical building complexes. Current optimization algorithms often use single objective functions and simple genetic algorithms, which are ineffective in handling the complex interrelationships and constraints between support parameters. This embodiment addresses the unique spatial distribution characteristics and group interaction relationships of historical building complex support systems, developing a more efficient and accurate method for optimizing support parameters. It introduces the concepts of group interaction intensity and gradient, quantifying the relationships between support groups as a guiding principle for support parameter optimization; it constructs a cluster-based association gene pool mechanism, making the crossover and mutation operations of the genetic algorithm more closely aligned with the physical nature of the support system; and it employs an adaptive mutation strategy, dynamically adjusting the mutation coefficient based on the group interaction intensity, improving the algorithm's convergence efficiency and solution quality.

[0121] like Figure 2 The diagram shown illustrates the logic flowchart for optimizing support parameters.

[0122] The collaborative optimization system for temporary support systems of historical building complexes in this embodiment of the invention includes:

[0123] The first unit is used to obtain multi-site scanning results of historical building complexes using three-dimensional laser scanning, and to determine point cloud data through point cloud registration and stitching processing.

[0124] The second unit is used to sample the density distribution of point cloud data and identify components. By analyzing the force transmission sequence, it analyzes the component types and spatial locations of historical building complexes and generates a component relationship matrix.

[0125] The third unit is used to partition the support of historical building complexes based on the component relationship matrix and using the spectral clustering algorithm. By calculating the spatial similarity and stress relationship between components, a support correlation matrix is ​​constructed to divide the support groups.

[0126] The fourth unit is used to obtain the group action intensity matrix based on the geometric similarity index and structural coupling index of the support group using the grey relational analysis method.

[0127] The fifth unit is used to construct a support layout spatial grid based on the group action intensity matrix, construct an optimization objective function based on the group action intensity value and action gradient, and use a genetic algorithm with a secondary mutation mechanism to optimize the support parameter values ​​of the support facilities and determine the support layout scheme.

[0128] A third aspect of the present invention provides an electronic device, comprising:

[0129] processor;

[0130] Memory used to store processor-executable instructions;

[0131] The processor is configured to invoke instructions stored in the memory to execute the aforementioned method.

[0132] A fourth aspect of the present invention provides a computer-readable storage medium having stored thereon computer program instructions that, when executed by a processor, implement the aforementioned method.

[0133] This invention can be a method, apparatus, system, and / or computer program product. The computer program product may include a computer-readable storage medium having computer-readable program instructions loaded thereon for performing various aspects of the invention.

[0134] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for collaborative optimization of temporary support system of historic building complex, characterized in that, The method comprises the following steps: Obtain multi-station scanning results of the historical building group by using three-dimensional laser scanning, and determine point cloud data by point cloud registration and splicing processing; Sample and identify the density distribution of the point cloud data, analyze the component type and spatial position of the historical building group by force transmission sequence, and generate a component relationship matrix; Based on the component relationship matrix, use spectral clustering algorithm to support partitioning of the historical building group, calculate the spatial similarity and force action relationship between components, construct a support correlation matrix, and divide support groups; Based on the geometric similarity index and structure coupling degree index of the support group, use gray correlation analysis method to obtain group action strength matrix; Based on the group action strength matrix, construct a support layout space grid, based on the group action strength value and action gradient, construct an optimization objective function, use genetic algorithm with secondary mutation mechanism to optimize the support parameter value of the support facility, and determine the support layout scheme.

2. The method of claim 1, wherein, Sampling and identifying the density distribution of the point cloud data, analyzing the component type and spatial position of the historical building group by force transmission sequence, and generating a component relationship matrix include: Calculate the density distribution map based on the point cloud data, determine the sampling radius according to the density distribution state in the density distribution map, resample the point cloud data using the sampling radius, and obtain the sampling point cloud; Extract component local structure feature values and overall contour feature values from the sampling point cloud, and perform weighted superposition to obtain component feature data; Input the component feature data into the recognizer, calculate the component type probability value of each sampling point, and determine the component type of the sampling point according to the component type probability value; Spatially cluster the sampling points with the same component type, and divide the sampling points with a spatial distance lower than a preset distance threshold into the same component individual; Fit the boundary of each component individual to obtain the spatial position and shape information of the component; Calculate the spatial shortest distance and projection overlap area between adjacent components to generate a distance matrix and an overlap matrix between components, judge the support relationship between components, and generate a component connection relationship table; According to the component connection relationship table, establish a force transmission sequence, determine the force transmission path between components, and identify the force convergence node component; Calculate the action relationship strength between components according to the connection relationship table and the force transmission sequence, and generate a component relationship matrix.

3. The method of claim 2, wherein, Calculating the spatial shortest distance and projection overlap area between adjacent components to generate a distance matrix and an overlap matrix between components, judging the support relationship between components, and generating a component connection relationship table include: Calculate the spatial shortest distance and projection overlap area between adjacent components to generate a distance matrix and an overlap matrix between components, and construct an initial component connection graph by taking each component as a node in the graph network; Extract the local neighborhood features of each node, including component type, relative position relationship, and overlap degree, transfer and converge node features in the neighborhood based on attention mechanism, and obtain fused environment node features; Input the fused environment node features into the strategy network, generate a connection probability distribution according to the spatial relationship between components, calculate the reward value based on the statics of the building and the basic construction rules, and optimize the connection judgment strategy by gradient update. Based on the optimized strategy network, the connection probability value of each pair of adjacent components in the initial component connection graph is calculated, the edge weight is updated according to the connection probability value, the connectivity analysis and stress verification are performed on the updated component connection graph, and the final component connection relationship table is generated.

4. The method of claim 1, wherein, Based on the component relationship matrix, the support partition of the historical building group is performed by using a spectral clustering algorithm, a support correlation matrix is constructed by calculating the spatial similarity and stress action relationship between components, and the support group is divided, including: The component relationship matrix of the historical building group is subjected to eigenvalue decomposition to obtain an eigenvalue array and a transformation basis array, a feature mapping space is constructed based on the eigenvalue array by selecting a transformation basis vector corresponding to the maximum eigenvalue according to a preset optimal number, and a component mapping vector is obtained by mapping the components in the component relationship matrix to the feature mapping space; Based on the component mapping vector, a component spatial similarity matrix is obtained by Gaussian kernel function operation, and a component stress action matrix is calculated based on the stress transmission path in the component relationship matrix; Based on the component spatial similarity matrix and the component stress action matrix, a support correlation matrix is generated by weighted fusion, an action intensity matrix is calculated according to the support correlation matrix, a Laplacian matrix is constructed by using the action intensity matrix and the support correlation matrix, a discrete spectrum sequence is obtained by solving the generalized eigenvalue equation of the Laplacian matrix, the number of groups is determined based on the discrete spectrum sequence, a mapping basis group of the Laplacian matrix is selected according to the number of groups, the mapping basis group is clustered by using K-means clustering operation, and a support unit division result is obtained to determine the support group.

5. The method of claim 1, wherein, Based on the geometric similarity index and the structure coupling degree index of the support group, the group action intensity matrix is obtained by using a grey correlation analysis method, including: The spatial barycentric coordinates and the projection area of the support group are calculated to obtain the group barycentric distance ratio, the boundary overlap degree and the morphological similarity, which constitute the geometric similarity index; the connection density index, the stress transmission coefficient and the node component proportion are obtained by counting the component connection number, the stress transmission path and the node component distribution of the support group, which constitute the structure coupling degree index; the geometric similarity index and the structure coupling degree index are normalized according to the index type and the value range to form a grey correlation factor sequence; The maximum value in the grey correlation factor sequence is constructed as a reference sequence, and the remaining index values are constructed as comparison sequences, and the absolute difference sequence of the reference sequence and the comparison sequence is calculated; the absolute difference sequence is partitioned by two-level resolution coefficients to obtain an association interval, in the association interval, the corresponding adaptive weight coefficient is set according to the difference degree, the local correlation degree and the overall correlation degree are calculated respectively, and the corresponding weighted result is taken as the grey correlation coefficient between groups; For each index in the grey correlation factor sequence, the initial weight interval is determined by the upper and lower limits of the interval, the comprehensive weight of each index is obtained by iteratively optimizing the initial weight interval by using the maximum deviation method, the group action intensity is obtained by weighted fusion of the grey correlation coefficient and the corresponding comprehensive weight, and the group action intensity matrix is formed by the group action intensity.

6. The method of claim 1, wherein, The support arrangement space grid is constructed based on the group action intensity matrix, the optimization objective function is constructed based on the group action intensity value and the action gradient, and the support parameter value of the support facility is optimized by using the genetic algorithm with the secondary mutation mechanism, including: The group action intensity matrix is converted into a grid subdivision weight matrix, a support arrangement space grid is constructed, the group action intensity value of the grid node is extracted, and the group action gradient is calculated by the change rate between adjacent nodes; The arrangement interval and the specification size of the support facility at each grid are collected, the support cost and the support response are calculated based on the group action intensity value and the action gradient, and the support optimization objective is constructed; The maximum change amount of the support parameter between adjacent grids is set based on the group action intensity difference, and the continuous change range of the support parameter is set based on the action intensity distribution in the local region, and the parameter constraint is determined; The iteration operation is performed based on the parameter constraint, including: the arrangement interval and the specification size of the support facility at each grid are constructed into chromosome coding, the chromosome coding is clustered and divided based on the group action intensity value, the chromosome coding in the same associated gene pool is stored, the chromosome coding is selected in the same associated gene pool for crossover operation to obtain crossover coding; the mutation operation is performed on the crossover coding, the mutation coefficient is calculated based on the group action intensity value, the support parameter value in the chromosome coding is adjusted according to the mutation coefficient to obtain mutation coding, and the parameter similarity of the mutation coding and the chromosome coding in the associated gene pool is calculated; when the parameter similarity is greater than a preset similarity threshold, the secondary mutation is triggered, and the mutation result is determined until the parameter similarity is greater than the similarity threshold; the iteration is repeated, and the final support parameter value is determined by the iteration operation until the support optimization objective is met.

7. A system for the coordinated optimization of temporary support systems for historic building complexes for implementing the method according to any one of the preceding claims 1 to 6, characterized in that, including: The first unit is configured to obtain multi-site scanning results of the historical building group by using three-dimensional laser scanning, and determine point cloud data by point cloud registration and splicing processing; The second unit is configured to perform density distribution sampling and component identification on the point cloud data, analyze the component type and spatial position of the historical building group through force transmission sequence, and generate a component relationship matrix; The third unit is configured to divide the support of the historical building group based on the component relationship matrix by using a spectral clustering algorithm, construct a support correlation matrix by calculating the spatial similarity and force action relationship between components, and divide the support groups; The fourth unit is configured to obtain a group action intensity matrix based on the geometric similarity index and the structure coupling degree index of the support groups by using a grey correlation analysis method; The fifth unit is configured to construct a support arrangement space grid based on the group action intensity matrix, construct an optimization objective function based on the group action intensity value and the action gradient, and optimize the support parameter value of the support facility by using a genetic algorithm with a secondary mutation mechanism to determine a support arrangement scheme.

8. An electronic device, comprising: including: A processor; A memory for storing processor-executable instructions; The processor is configured to invoke the instructions stored in the memory to execute the method of any one of claims 1 to 6.

9. A computer-readable storage medium having stored thereon computer program instructions, wherein, The computer program instructions are executed by the processor to implement the method of any one of claims 1 to 6.

Images