Urban landscape intelligent planning and decision-making method based on multi-objective optimization algorithm

Abstract

The application discloses a city style intelligent planning decision method based on a multi-objective optimization algorithm and belongs to the technical field of city planning. The method first acquires a three-dimensional space data voxelization cut of a planning area to construct a three-dimensional visual sensitive field; then, initial population is generated by encoding building group parameters, some samples are extracted to calculate style constraint indexes and an approximate mapping relationship between building parameters and style constraint indexes is fitted and established; on this basis, non-dominated sorting is carried out in combination with corresponding constraint conditions; further, elite individuals are extracted and three-dimensional ray tracing is used to check real indexes and correct fitness; finally, the approximate mapping relationship is updated by using the corrected elite individuals to output a scheme set. The application breaks the barrier between precision and efficiency through the cooperative mechanism of approximate prediction and accurate verification, can generate high-quality city planning schemes under complex spatial topological and visual field constraint conditions, and realizes the cooperative promotion of spatial quality and utilization efficiency under fine style control.

Description

Technical Field

[0001] This invention relates to the field of urban planning technology, and in particular to an intelligent urban landscape planning and decision-making method based on multi-objective optimization algorithms. Background Technology

[0002] Urban landscape is a comprehensive reflection of a city's history, culture, spatial pattern, architectural form, and regional characteristics. It is of great significance for continuing the city's cultural heritage, shaping its unique features, and enhancing its quality. With the continuous advancement of urban renewal, optimization of existing construction, and the requirements for high-quality development, urban planning is gradually shifting from incremental expansion to a development stage that emphasizes both improving the quality of existing resources and shaping distinctive features. The coordinated protection and construction of urban landscape has become an important aspect of planning decisions.

[0003] Existing urban landscape planning decision-making technologies typically begin by acquiring multi-source data on the planning area, including historical and cultural resources, architectural spatial forms, current land use, geospatial information, population and industrial development, and relevant planning and control requirements. This data is then organized, standardized, and managed uniformly. Subsequently, technologies such as knowledge graphs and feature extraction are used to identify and represent urban landscape elements and their relationships, transforming information such as architectural style, spatial pattern, landscape features, and historical resources into calculable feature data. Based on this, a landscape value assessment index system is constructed, and a case resource library is formed by combining historical planning cases. Finally, intelligent analysis models are used to comprehensively evaluate the landscape characteristics, spatial quality, and development characteristics of the planning area. The process involves several steps. First, an optimization model is established based on the planning objectives. This model considers factors such as landscape protection requirements, development needs, space utilization efficiency, public service allocation levels, and landscape harmony as optimization targets. Simultaneously, a constraint system is constructed by combining planning conditions such as building height control, development intensity control, land use constraints, and protection scope control. Then, a multi-objective optimization algorithm is used to iteratively solve the planning schemes. Through processes such as population initialization, fitness calculation, selection, crossover, mutation, and non-dominated sorting, multiple optimal solutions satisfying the constraints are obtained. Finally, the candidate schemes are comprehensively evaluated and compared, resulting in multiple sets of planning comparison results. This provides decision-making references for urban landscape planning, scheme demonstration, and planning management.

[0004] However, in the process of implementing the inventive technical solution in the embodiments of this application, it was found that the above-mentioned technology has at least the following technical problems:

[0005] In complex scenarios involving visual corridor protection and skyline control, landscape constraints often manifest as highly nonlinear geometric and topological coupling relationships based on spatial relative positions, such as visual cross-section control and contour correspondence. Faced with such constraints, existing decision-making methods typically employ dimensionality reduction strategies, conservatively transforming them into algebraic constraints such as absolute height envelopes. While this method ensures efficiency in decision-making, it fails to accurately characterize the dynamic changes in visual gradients, easily eliminating schemes that, while not actually damaging the visual corridor, exceed absolute height limits. This leads to a conservative approach in balancing landscape protection and development, limiting the synergistic improvement of spatial quality and utilization efficiency. Directly introducing 3D line-of-sight occlusion and topological fit calculations into the decision model to pursue constraint accuracy results in a sharp increase in the computational cost of fitness assessment, significantly reducing the convergence speed and solution timeliness of multi-objective optimization algorithms. Therefore, existing urban landscape planning decisions still face the limitation of balancing constraint modeling accuracy and solution efficiency when dealing with complex spatial topology and visual constraints, hindering the generation and decision-making of high-quality planning schemes in refined landscape management scenarios. Summary of the Invention

[0006] To address the technical problem in existing technologies where dimensionality reduction simplification fails to capture the dynamic characteristics of the field of view when dealing with landscape constraints involving highly nonlinear geometry and topological coupling, leading to a surge in computational power due to direct 3D calculations, it is difficult to balance modeling accuracy and solution efficiency. This results in conservative planning decisions or inefficient optimization solutions, hindering the generation of high-quality planning schemes and the synergistic improvement of space utilization efficiency under refined landscape management. This invention provides an intelligent urban landscape planning decision-making method based on a multi-objective optimization algorithm. The technical solution is as follows:

[0007] A smart urban landscape planning and decision-making method based on a multi-objective optimization algorithm includes: S1, acquiring 3D spatial data of the planning area and performing voxel segmentation; calculating the visual occlusion weight of each voxel grid on the visual corridor based on the spatial relative position between the preset viewpoint and the landscape target, and constructing a 3D visual sensitivity field; S2, encoding the spatial parameters of the building clusters in the planning area to generate an initial planning scheme population; extracting some individual schemes from the initial planning scheme population as reference scheme samples; calculating the 3D visual occlusion degree and the topological fit of the skyline outline of the reference scheme samples in the 3D visual sensitivity field as landscape constraint indicators; and fitting an approximate mapping relationship between the spatial parameters of the building clusters and the landscape constraint indicators based on the encoded features of the reference scheme samples and the landscape constraint indicators; S3, using the initial planning scheme population as the current population, predicting based on the approximate mapping relationship. The current population is used to calculate the target fitness of each planning scheme individual based on the landscape constraint index and the preset planning optimization objective. The current population is then non-dominated and sorted according to the target fitness and the constraints corresponding to the landscape constraint index. Population crossover and mutation operations are then performed to generate the next generation population. In step S4, during the iterative evolution of the population, elite individuals at the non-dominated frontier are extracted from the current population. A 3D ray tracing algorithm is used to perform 3D view occlusion and topological fit verification on these elite individuals to obtain the true landscape constraint index. The true landscape constraint index is then used to correct the target fitness of the elite individuals. In step S5, the corrected elite individuals are used as new samples to iteratively update the approximate mapping relationship. The next generation population is then used as the new current population. Step S3 is returned to be executed until the preset convergence condition is met, and the intelligent urban landscape planning decision scheme set is output.

[0008] The beneficial effects of the technical solutions provided in the embodiments of the present invention include at least the following:

[0009] This invention can achieve a balance between the accuracy of landscape constraint modeling and the efficiency of optimization solution in highly nonlinear geometric and topological coupling constraint scenarios such as complex view corridor protection and skyline control, thereby significantly improving the scientific, refined and intelligent level of planning decisions.

[0010] Specifically, this invention acquires three-dimensional spatial data of the planning area and constructs a voxelized three-dimensional visually sensitive field, transforming landscape control requirements such as visual corridor protection and landscape visibility into quantifiable spatial visibility information. This fully reflects the occlusion impact of different spatial locations on landscape targets and the degree of visual sensitivity. Compared to the dimensionality reduction simplification strategies or absolute height envelope constraints commonly used in existing technologies, this invention can realistically depict the dynamic changes in visual gradients and the correspondence between spatial contours, thereby avoiding the elimination of practically feasible high-quality planning schemes due to overly conservative height restrictions and improving the synergy between spatial quality and utilization efficiency.

[0011] Furthermore, this invention achieves rapid prediction of complex 3D view occlusion analysis and skyline topology evaluation by encoding the spatial parameters of building clusters and establishing an approximate mapping relationship between these parameters and landscape constraint indicators based on reference scheme samples. In existing technologies, directly introducing 3D view occlusion calculations into multi-objective optimization often leads to a sharp increase in fitness evaluation overhead and a decrease in convergence speed. This invention, however, significantly reduces computational costs while maintaining constraint representation capabilities through a surrogate model, improving the efficiency of planning scheme selection and optimization solutions, and providing reliable technical support for large-scale scheme search.

[0012] Furthermore, this invention combines the landscape constraint indicators predicted by approximate mapping with the planning objectives during the optimization process to perform non-dominated ranking and population evolution, achieving synergistic optimization of landscape protection objectives and construction and development objectives. Compared with the existing technology's simple dimensionality reduction approach to landscape constraints, this invention can preserve complex spatial relationships and dynamic features of the field of view, expand the search space for feasible solutions, and improve the rationality and innovation of the planning scheme.

[0013] Meanwhile, this invention extracts non-dominant frontier elite individuals during the iteration process, uses a three-dimensional ray tracing algorithm for real-world verification, and adjusts the fitness accordingly to ensure the accuracy of critical solution evaluation. Unlike existing technologies that involve excessive computational burden due to full-scale three-dimensional computation, this invention performs precise analysis only on high-value solutions, achieving an effective balance between computational accuracy and efficiency.

[0014] Finally, by continuously updating the approximate mapping relationship, this invention enables the surrogate model to adaptively learn new sample information during the optimization process, gradually improving prediction accuracy and optimization convergence stability, thereby significantly enhancing the reliability and intelligence level of the planning decision model. In summary, this invention not only solves the problem of balancing modeling accuracy and solution efficiency under complex landscape constraints in existing technologies, but also achieves a balanced optimization of landscape protection and urban construction, providing strong technical support for the generation of high-quality urban spatial planning schemes. Attached Figure Description

[0015] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0016] Figure 1 A flowchart illustrating the intelligent urban landscape planning and decision-making method based on a multi-objective optimization algorithm provided in this application embodiment;

[0017] Figure 2The overall logic diagram of the intelligent urban landscape planning and decision-making method based on multi-objective optimization algorithm provided in the embodiments of this application is shown. Detailed Implementation

[0018] Embodiments of the present disclosure will now be described in more detail with reference to the accompanying drawings. While some embodiments of the present disclosure are shown in the drawings, it should be understood that embodiments of the present disclosure may be implemented in various forms and should not be construed as limited to the embodiments set forth herein. Rather, these embodiments are provided to provide a more thorough and complete understanding of the present disclosure.

[0019] It should be understood that the accompanying drawings and embodiments of this disclosure are for illustrative purposes only and are not intended to limit the scope of protection of this disclosure. In the description of the embodiments of this disclosure, the term "comprising" and similar terms should be understood as open-ended inclusion, i.e., "including but not limited to". The term "based on" should be understood as "at least partially based on". The term "one embodiment" or "this embodiment" should be understood as "at least one embodiment". The terms "first", "second", etc., may refer to different or the same objects.

[0020] To make the technical problems, technical solutions and advantages of the present invention clearer, a detailed description will be given below in conjunction with the accompanying drawings and specific embodiments.

[0021] With the continuous advancement of urbanization and the arrival of the era of stock renewal, urban planning has gradually shifted from a scale expansion-oriented approach to a quality improvement-oriented approach. Urban landscape, as an important carrier reflecting urban characteristics, historical context, and spatial quality, has become a key aspect of urban planning and construction management. In actual planning, landscape control requirements such as visual corridor protection, skyline control, coordination of important landscape nodes, and shaping of the overall urban spatial form typically require comprehensive consideration of building height, volume, layout, and spatial relationships with surrounding landscape resources to ensure the accessibility and aesthetic appeal of urban landscape resources and improve the quality of the urban spatial environment. Therefore, how to achieve a coordinated unity between urban landscape protection and efficient utilization of spatial resources while meeting the needs of urban construction and development has become a crucial issue that urgently needs to be addressed in the field of urban planning decision-making.

[0022] Current urban landscape planning decision-making methods mostly rely on planning experience, rule constraints, or evaluation methods based on indicator systems to screen and optimize planning schemes. With the application of multi-objective optimization algorithms in the field of urban planning, planning schemes can be synergistically optimized among multiple objectives such as construction intensity, economic benefits, and public service levels. However, constraints related to urban landscape often involve complex spatial relationships between sight corridors, landscape interfaces, skyline outlines, and the spatial forms of building clusters. These are essentially highly nonlinear geometric and topological coupling constraints, which are difficult to accurately describe using traditional algebraic models.

[0023] To reduce the difficulty of optimization solutions, existing technologies typically simplify landscape constraints such as visual corridor protection and skyline control, for example, by using absolute height control lines, height zoning management, or envelope restrictions. While such methods can improve the efficiency of planning scheme evaluation, they tend to exclude some planning schemes that actually meet landscape requirements because they ignore the dynamic characteristics of visual obstruction caused by changes in spatial location and the influence of the overall outline relationship of building groups. This leads to conservative planning decisions, limiting the rational development and utilization of urban spatial resources and the improvement of spatial quality.

[0024] On the other hand, to improve the accuracy of landscape constraints, some studies have attempted to directly introduce methods such as 3D line-of-sight analysis, ray tracing calculations, and spatial topological relationship analysis into the planning optimization process to conduct refined evaluations of planning schemes. However, these methods require frequent and complex 3D spatial calculations during the optimization iteration process. As the planning area expands and the number of schemes increases, the computational cost shows a significant upward trend, which can easily lead to a decrease in optimization solution efficiency and a slower convergence speed, making it difficult to meet the timeliness requirements of actual planning decisions.

[0025] Therefore, how to balance the accuracy of expressing complex landscape constraints with the efficiency of multi-objective optimization in the process of urban landscape planning and decision-making, accurately reflect spatial landscape requirements such as visual corridor protection and skyline control, while reducing the computational power consumption caused by large-scale 3D calculations, and achieving coordinated optimization among urban landscape protection, construction and development, and space utilization efficiency, has become an urgent technical problem to be solved in the field of intelligent urban planning and decision-making. To address these needs, it is necessary to propose an intelligent urban landscape planning and decision-making method that can balance the accuracy of landscape constraint modeling with the efficiency of optimization solutions, so as to support the generation and decision-making of high-quality planning schemes under the conditions of refined landscape management.

[0026] like Figure 1 The diagram shown is a flowchart of an intelligent urban landscape planning and decision-making method based on a multi-objective optimization algorithm provided in this application embodiment. Figure 1 As can be seen, the method of this invention first acquires three-dimensional spatial data of the planning area and performs voxelization processing. Combining this with the spatial relationship between preset viewpoints and landscape targets, a three-dimensional visually sensitive field is constructed that reflects the degree of visual obstruction, thereby achieving a refined and quantitative expression of landscape control requirements such as visual corridor protection. Based on this, the spatial parameters of the building complex are encoded to generate an initial planning scheme population. Furthermore, by extracting some reference schemes and calculating their three-dimensional visual obstruction degree and skyline topological fit, an approximate mapping relationship is established between building spatial parameters and landscape constraint indicators, enabling rapid prediction of complex landscape constraints.

[0027] Furthermore, in the multi-objective optimization process, the landscape constraint indicators of each planning scheme are predicted using approximate mapping relationships. Fitness is calculated in conjunction with preset planning objectives, and new scheme populations are continuously generated through non-dominated ranking and crossover mutation operations. This achieves synergistic optimization of landscape protection and development goals while ensuring computational efficiency. Simultaneously, elite individuals on the non-dominated frontier are extracted during the iteration process, and their real landscape constraints are verified using a three-dimensional ray tracing algorithm. Fitness is then adjusted accordingly to ensure the accuracy and reliability of the key scheme evaluation results. Finally, the verified elite individuals are used as new samples to continuously update the approximate mapping relationships, enabling the model to continuously improve prediction accuracy and gradually converge during the iteration process, ultimately outputting a set of urban planning schemes that meet complex landscape constraints.

[0028] It should also be understood that, through the above technical solution, the present invention avoids the problems of insufficient landscape expression due to dimensional reduction simplification or low efficiency due to full three-dimensional calculation in the prior art, while achieving an effective balance between the accuracy of landscape constraint modeling and the efficiency of optimization solution. It can not only improve the convergence speed of multi-objective optimization and the efficiency of planning decision-making, but also avoid the elimination of high-quality solutions by excessively conservative constraints, thereby significantly improving the scientificity, rationality and space utilization efficiency of urban landscape planning.

[0029] refer to Figure 1 Specifically, the first step of the intelligent urban landscape planning and decision-making method based on multi-objective optimization algorithms is as follows: S1, acquire the three-dimensional spatial data of the planning area and perform voxel segmentation. Based on the spatial relative position between the preset viewpoint and the landscape target, calculate the visual occlusion weight of each voxel grid on the visual corridor and construct a three-dimensional visual sensitive field. By establishing a three-dimensional visual sensitive field, the traditional qualitative landscape requirements are transformed into quantifiable and computable spatial constraint information, which can accurately identify the importance of different spatial locations for the protection of the visual corridor, provide a unified evaluation basis for the landscape evaluation of subsequent planning schemes, and improve the scientific and refined level of landscape control.

[0030] As a further solution, in S1, a three-dimensional visual sensitive field is constructed, specifically including:

[0031] Step S11: Generate the view cross section and the three-dimensional envelope: Based on the spatial coordinates of preset viewpoints (such as important urban viewpoints, road viewpoints, etc.) and the boundary contours of landscape targets (such as historical landmarks, natural mountains, etc.), generate a spatial connection line from the preset viewpoint to the landscape target; using the spatial connection line as the central axis of the view corridor, combined with the preset view opening angle or the projection boundary of the landscape target, extend it into space to generate the view cross section; then, perform lofting or sweeping operations on all view cross sections along the central axis, calculate and generate the three-dimensional envelope that encloses the space of the view corridor, and the three-dimensional envelope defines the potential spatial range of the view corridor being occluded.

[0032] Step S12, Voxelization segmentation and basic view occlusion sensitivity calculation:

[0033] The 3D envelope is voxelized and segmented according to a preset spatial resolution (e.g., 5m×5m×5m). Voxels located outside the 3D envelope are filtered out, while voxel meshes within the 3D envelope are retained. For any retained voxel mesh... Obtain the coordinates of its center point. Calculate the center point The shortest spatial distance to the space connection Simultaneously, calculate the center point. A first direction vector pointing to a preset viewpoint and a second direction vector connecting them in space are given, and the spatial angle between the first and second direction vectors is calculated. Based on the shortest spatial distance Angle with space Computational voxel mesh Basic visual obstruction sensitivity The calculation formula is as follows: ;in, This is a preset dimension adjustment coefficient used to adjust the calculation results to a preset order of magnitude. To prevent extremely small positive numbers with a denominator of zero, the preferred value range is 0.01 to 0.1. This formula indicates that the closer a voxel is to the central axis of the visual corridor and the better its alignment with the viewpoint, the higher its occlusion threat to the visual corridor and the greater its basic sensitivity. The larger.

[0034] Step S13, combining viewpoint importance normalization processing:

[0035] Assign weight coefficients to each preset viewpoint To reflect the differences in viewing levels (e.g., a city-level core viewing point has a weight of 1.0, while a district-level viewing point has a weight of 0.6), the voxel grid is used. Basic visual obstruction sensitivity The importance weight of its viewpoint The weighted sensitivity is obtained by multiplication; then, the weighted sensitivity of all voxel meshes within the 3D envelope is normalized using the Min-Max normalization method, and the result is mapped to the [0,1] interval to obtain the view occlusion weight of each voxel mesh for a single viewpoint. The calculation formula is as follows: ;in, and These are the maximum and minimum weighted base sensitivities of all voxel meshes within the current 3D envelope, respectively. If the same voxel mesh is simultaneously located within the 3D envelope of multiple preset viewpoints, then the view occlusion weights calculated for that voxel mesh at each viewpoint are compared. The maximum value among them is taken as the final view occlusion weight of the voxel mesh. ,Right now To ensure the landscape control requirements of high-level viewpoints.

[0036] Step S14: Construct a three-dimensional visual sensitive field:

[0037] The calculated view occlusion weight Assigned as an attribute value to the corresponding voxel mesh, carrying the view occlusion weight attribute across the entire planning area. The set of voxel meshes together constitutes a three-dimensional visual sensitive field, which is stored in a three-dimensional raster data structure.

[0038] Existing technologies often simplify the visual corridor constraint to an absolute height envelope, leading to the inadvertent deletion of solutions that exceed the height limit but do not substantially obstruct the view. This invention introduces "spatial distance to the line of sight" and "normal angle" to construct a basic sensitivity formula, accurately distinguishing the differentiated impact of different spatial locations within the visual corridor on line of sight obstruction (e.g., voxels deviating from the center of the visual corridor or facing away from the viewpoint pose a low threat of obstruction). This achieves dynamic, non-equivalent quantification of visual gradient sensitivity, effectively avoiding the conservative decision-making problems caused by the traditional "one-size-fits-all" approach of absolute height, and providing a refined basis for balancing landscape protection and construction development.

[0039] The second step of the intelligent urban landscape planning and decision-making method based on multi-objective optimization algorithms is as follows: S2, the spatial parameters of the building clusters in the planning area are encoded to generate an initial planning scheme population; some individual schemes are extracted from the initial planning scheme population as reference scheme samples, and the 3D view occlusion degree and skyline contour topological fit degree of the reference scheme samples in the 3D visual sensitive field are calculated as landscape constraint indicators. Based on the encoded features of the reference scheme samples and the landscape constraint indicators, an approximate mapping relationship between the building cluster spatial parameters and the landscape constraint indicators is fitted and established. This step enables complex landscape constraints to be quickly estimated through approximate mapping, avoiding frequent high-cost 3D simulation calculations during the optimization process, thereby significantly improving the efficiency of planning scheme evaluation and providing support for large-scale scheme optimization.

[0040] It should be noted that in S2, the spatial parameters of the building complex in the planning area are encoded to generate an initial planning scheme population, specifically including:

[0041] Each building plot within the planning area is treated as an independent coding unit. The building height, base area ratio, building setback distance, and volume rotation angle of each building plot are extracted as spatial control parameters. The value range of each spatial control parameter is determined according to the control detailed plan of the planning area. Using real number coding, the spatial control parameters of all building plots are sequentially concatenated and combined into multi-dimensional decision vectors according to the plot number to form a chromosome sequence representing the individual initial planning schemes. Multiple multi-dimensional decision vectors are randomly generated within the value range to form the initial planning scheme population.

[0042] It should also be noted that in S2, the 3D view occlusion is calculated using the following methods:

[0043] Based on a unified spatial coordinate system for the planning area, and with the same spatial resolution and origin coordinates as the three-dimensional visual sensitive field in step S1, the three-dimensional model of the building complex corresponding to the reference scheme sample is transformed into voxels to obtain the set of entity voxel meshes occupied by the building complex.

[0044] Traverse the set of entity voxel meshes, match them in the three-dimensional visual sensitive field according to the spatial coordinates of the voxel meshes, identify the building entity voxels that fall within the range of the three-dimensional visual sensitive field, and accumulate the line-of-sight occlusion weights corresponding to the identified building entity voxels to obtain the cumulative occlusion weights.

[0045] The ratio of the cumulative occlusion weight to the sum of the line-of-sight occlusion weights of all voxel meshes in the 3D visually sensitive field is calculated, and the calculated ratio is used as the 3D visual field occlusion degree.

[0046] In addition, in S2, the topological fit of the skyline outline is calculated in the following ways:

[0047] Using the line of sight from the preset viewpoint to the landscape target as the projection direction, the outline of the three-dimensional model of the building complex in the reference scheme sample is orthographically projected onto a two-dimensional projection plane perpendicular to the projection direction, and the upper edge line of the projection is extracted as the generated skyline; the skyline of the historical and cultural protection area corresponding to the landscape target is extracted from the urban historical and cultural protection planning documents or the existing three-dimensional geographic information data as the reference skyline.

[0048] The generated skyline and the reference skyline are unified to the same two-dimensional coordinate system. The Douglas-Puk algorithm is used to extract the inflection points of the generated skyline and the reference skyline as key turning points. The local elevation maximum point of each skyline is extracted as the peak node and the local elevation minimum point is extracted as the trough node.

[0049] Based on the principle of nearest neighbor in horizontal coordinates, the corresponding key turning points and peak and valley nodes are matched and determined on the reference skyline to generate each key turning point and peak and valley node of the skyline.

[0050] Calculate the curvature values ​​of the generated skyline and the reference skyline at the corresponding key inflection points, and calculate the sum of the absolute differences of the curvature values ​​at the corresponding key inflection points to obtain the curvature difference degree; calculate the Euclidean distance between the corresponding peak and trough nodes of the generated skyline and the reference skyline on the two-dimensional projection plane, as the spatial misalignment distance.

[0051] The topological fit of the skyline contour is calculated based on the curvature difference and spatial misalignment distance, using the following formula: ;in, For the topological fit of the skyline, and These are the curvature values ​​of the generated skyline and the reference skyline at the corresponding key inflection points, respectively. This corresponds to the spatial misalignment distance between peak and trough nodes. and The preset weighting coefficients and (Set according to the emphasis on contour undulation and skeleton offset based on the landscape control). and The preset scaling factor (calibrated based on the curvature and spatial scale of the skyline of the historical district, used to normalize the exponential term). The larger the value, the better the shape fit between the generated skyline and the reference skyline.

[0052] As a further step, in S2, an approximate mapping relationship between the spatial parameters of the building complex and the style constraint index is established by fitting, specifically including:

[0053] A Latin hypercube sampling method is used to uniformly extract a predetermined number (preferably 10%~20% of the population size, but no less than 50) of individual schemes from the initial planning scheme population as reference scheme samples to ensure the uniformity of sample distribution in the multidimensional decision space. The multidimensional decision vector of each reference scheme sample is used as the input feature, and the two-dimensional vector composed of the calculated three-dimensional view occlusion degree and the topological fit of the skyline contour is used as the output label to construct a training set. A surrogate model based on a radial basis function neural network is constructed. The number of nodes in the input layer of the surrogate model is equal to the dimension of the multidimensional decision vector, and the number of nodes in the output layer is 2 (corresponding to the three-dimensional view occlusion degree and the topological fit of the skyline contour, respectively). The hidden layer uses a Gaussian kernel function.

[0054] The surrogate model is trained using the training set. The center vector and width parameters of the Gaussian kernel function in the hidden layer are determined by the K-means clustering algorithm. The weight matrix from the hidden layer to the output layer is calculated by the least squares method.

[0055] During training, the mean squared error (MSE) on the training set is used as the loss function. Training stops when the loss function value is less than a preset threshold (such as 0.001) or when the maximum number of iterations is reached. The surrogate model after training convergence is used as an approximate mapping relationship between the spatial parameters of the building complex and the style constraint index.

[0056] This step involves encoding the spatial parameters of the building complex to generate an initial population of planning schemes, and using reference scheme samples to establish an approximate mapping relationship between the spatial parameters of the building complex and the landscape constraint indicators. This achieves a digital expression of the spatial form of the planning schemes and rapid prediction of landscape constraints. Simultaneously, by constructing quantitative indicators such as three-dimensional view occlusion and skyline topological fit, the visual corridor protection effect and skyline harmony of the planning schemes are objectively evaluated, transforming traditional experience-based landscape evaluation into calculable and comparable constraints. Employing surrogate models to replace some high-cost three-dimensional analysis calculations significantly improves the efficiency of planning scheme evaluation while ensuring the accuracy of landscape assessment, providing efficient and reliable landscape constraint support for subsequent multi-objective optimization.

[0057] The third step of the intelligent urban landscape planning and decision-making method based on multi-objective optimization algorithm is: S3, taking the initial planning scheme population as the current population, predicting the landscape constraint index of each planning scheme individual in the current population based on the approximate mapping relationship, calculating the target fitness of each planning scheme individual in combination with the preset planning optimization objective, performing non-dominated sorting of the current population according to the target fitness and the constraint conditions corresponding to the landscape constraint index, and performing population crossover and mutation operations to generate the next generation population; this step realizes the synergistic optimization of landscape protection objectives and urban development objectives, and can continuously explore better solution space while meeting landscape constraint requirements, improve the diversity and optimization efficiency of planning schemes, and thus provide more candidate schemes with practical value.

[0058] As a further solution, in S3, the current population is non-dominatedly ranked according to the target fitness and the constraints corresponding to the landscape constraint index, specifically including:

[0059] S31, Calculate the preset planning optimization objective function value:

[0060] The preset planning optimization objectives include maximizing the total amount of development and construction and maximizing the uniformity of spatial layout; for any individual planning scheme in the current population, based on its multi-dimensional decision vector (building height of each plot)... Base area ratio Building setback distance and volume rotation angle Calculate the building height of each plot: With base footprint and the pre-set land area The products are summed to obtain the first objective function value representing the total amount of development and construction. .

[0061] Based on the building setback distance of each building plot Rotation angle of volume The spatial spacing between adjacent buildings is calculated, and the proportion of building pairs that meet the preset daylighting spacing standard to the total number of adjacent building pairs is used as the second objective function value to characterize the uniformity of the spatial layout. ;Depend on and The vector of objective function values ​​that constitutes this individual.

[0062] S32, Calculate the degree of violation of the landscape constraint index:

[0063] Obtain the 3D view occlusion degree of each planning scheme individual based on the prediction of approximate mapping relationship. Topological fit with skyline outline and the preset upper limit threshold for view occlusion. Lower limit threshold of topological fit The excessive values ​​of each constraint index are normalized and summed to calculate the total constraint violation degree. The calculation formula is: ;in, To prevent extremely small positive numbers with a denominator of zero (such as 0.001), if CV=0, the individual planning scheme is considered to satisfy the landscape constraints and is therefore a feasible solution; if CV>0, the individual planning scheme is considered to violate the landscape constraints and is therefore an infeasible solution, and the larger the CV value, the more severe the violation.

[0064] S33, Perform a constraint-based non-dominated sort:

[0065] In the current population, for any two planning schemes, individuals p and q, determine whether individual p dominates individual q according to the following rules:

[0066] Rule 1: If individuals p and q both satisfy the constraints (i.e. and If individual p is not inferior to individual q on all optimization objectives and is strictly superior to individual q on at least one optimization objective, then individual p is determined to dominate individual q.

[0067] Rule 2: If individual p satisfies the constraints while individual q violates the constraints (i.e., ... and If the constraint is satisfied, then individual p dominates individual q that violates the constraint.

[0068] Rule 3: If both individuals p and q violate the constraints (i.e. and Then compare the total degree of constraint violation between the two. If If the individual p, whose violation is less severe, is determined to dominate the individual q, whose violation is more severe.

[0069] Based on the above domination relationship, the number of individuals dominated by each individual in the population is calculated. Individuals not dominated by any other individual are assigned to the first non-dominated frontier. Subsequently, the remaining individuals are recursively assigned to subsequent non-dominated frontiers, and each individual is assigned a non-dominated rank.

[0070] S34, Calculate congestion distance:

[0071] For individuals within the same non-dominated frontier, the normalized distance between them and their neighboring individuals is calculated on each dimension of the objective function value. The sum of the distances in each dimension is used as the crowding distance CD of the individual to evaluate the sparsity of the individual's distribution in the objective space.

[0072] S35, perform population crossover and mutation operations:

[0073] The fitness evaluation criterion is a combination of an individual's non-dominated hierarchy rank and crowding distance (CD) (rank has higher priority than CD; the smaller the rank and the larger the CD, the better the individual). A tournament selection strategy is used to select excellent individuals. A simulated binary crossover operator (SBX) is used to perform a crossover operation on the selected individuals, with a crossover probability set to 1. (Value range 0.8~0.9), distribution index set to A polynomial mutation operator is used to mutate the crossover individuals, with the mutation probability set to 1. (The value is the reciprocal of the decision vector dimension), and the distribution index is set to... This generates the next generation of the population.

[0074] Step S3 rapidly predicts the landscape constraint indicators for each planning scheme based on approximate mapping relationships, and calculates the target fitness by combining the preset planning optimization objectives. By incorporating target optimization and landscape constraints into the non-dominated ranking process, multi-objective collaborative optimization of planning schemes is achieved. Specifically, by introducing a constraint violation determination mechanism, planning schemes that meet the landscape constraints are prioritized during non-dominated ranking, while schemes that violate the constraints are selectively chosen, ensuring that the optimization process always evolves towards meeting landscape requirements. Simultaneously, the solution space is continuously expanded using crossover and mutation operations, which helps improve population diversity and global search capabilities, preventing the optimization process from getting trapped in local optima. This achieves a balance between urban landscape protection requirements and planning and construction needs while considering planning objectives such as development intensity and space utilization efficiency, thereby improving the rationality of the optimization results and the quality of decision-making.

[0075] The fourth step of the intelligent urban landscape planning and decision-making method based on multi-objective optimization algorithm is as follows: S4, during the iterative evolution of the population, elite individuals in the current population that are not at the dominant frontier are extracted. A three-dimensional ray tracing algorithm is used to perform three-dimensional view occlusion and topological fit verification on the elite individuals to obtain the real landscape constraint indicators. The target fitness of the elite individuals is then corrected using the real landscape constraint indicators. By performing high-precision verification on the optimal candidate scheme, the prediction error generated by the approximate mapping model can be effectively eliminated, ensuring that the optimization direction is consistent with the real landscape constraints, thereby improving the reliability, accuracy and engineering application value of the planning results.

[0076] It should be further explained that in S4, a 3D ray tracing algorithm is used to perform 3D view occlusion and topological fit verification on elite individuals to obtain realistic landscape constraint indicators, specifically including:

[0077] First, a spatial hierarchical bounding box data structure is constructed: multi-dimensional decision vectors corresponding to elite individuals in the current population that are at the non-dominant frontier are extracted to generate a three-dimensional geometric model of the building complex; for all triangular faces in the three-dimensional geometric model of the building complex, a hierarchical bounding box tree data structure based on axis-aligned bounding boxes is constructed to accelerate the subsequent ray and face intersection calculation process.

[0078] Secondly, 3D ray tracing calculates the true 3D view occlusion:

[0079] Using a preset viewpoint as the origin of ray emission, a cluster of three-dimensional rays is emitted uniformly within the field of view frustum formed by the preset horizontal and vertical field of view angles, with equal angular step sizes. The distance value closest to the viewpoint on the bounding box of the landscape target boundary is obtained as the reference occlusion distance. .

[0080] Traverse each ray in the 3D ray cluster, and use a spatial hierarchical bounding box tree data structure to accelerate the calculation of intersection between the ray and the triangular facets of the building complex; if the ray reaches the reference occlusion distance Previously, it intersected with the building facade, and the distance from the intersection point to the viewpoint was... If the ray is blocked, it is determined that the ray is blocked; otherwise, it is determined that the ray is not blocked.

[0081] Count the number of blocked rays The total number of three-dimensional ray clusters Calculate the true 3D view occlusion. .

[0082] Next, the real skyline is reconstructed and the topological fit of the real skyline outline is calculated:

[0083] The horizontal field of view of the viewing cone is divided into multiple horizontal azimuth angles according to equal angular step sizes. For each horizontal azimuth angle, a vertical scanning ray is emitted in the vertical field of view. The spatial hierarchy bounding box tree is used to find the intersection point of the building facets closest to the viewpoint on the ray as the first visible intersection point. The elevation value of the first visible intersection point is extracted as the elevation of the skyline outline point under the horizontal azimuth angle. If there is no building obstruction at the azimuth angle, the elevation of the terrain surface is extracted as the elevation of the outline point.

[0084] Project all skyline contour points under the horizontal azimuth angle onto a two-dimensional coordinate system composed of the horizontal azimuth angle and the elevation, sort them according to the horizontal azimuth angle and connect them in sequence to reconstruct the two-dimensional real skyline.

[0085] Calculate the discrete Frescher distance between the real 2D skyline and a reference skyline projected onto the same 2D coordinate system. Discrete Fréchet distance Mapping to the topological fit of the true skyline outline The calculation formula is: ,in This is the preset distance attenuation coefficient; The smaller, The closer it is to 1, the higher the topological fit.

[0086] As a further solution, in S4, the target fitness of elite individuals is corrected using real-world appearance constraint indicators, specifically including:

[0087] First, replace the landscape constraint indicators:

[0088] The topological fit between the 3D view occlusion degree predicted by elite individuals based on an approximate mapping relationship and the skyline contour is replaced with the actual 3D view occlusion degree obtained through 3D ray tracing verification. Topological fit with the real skyline .

[0089] Secondly, calculate the deviation magnitude and the fitness penalty factor:

[0090] Based on a preset upper limit threshold for view occlusion Lower limit threshold of topological fit To determine whether the actual landscape constraint indicators violate the preset constraint conditions.

[0091] like and If the constraint is satisfied, the fitness penalty factor is applied. If the constraints are violated, a fitness penalty factor is dynamically generated based on the magnitude of the deviation exceeding the constraint range. The calculation formula is: ;in, and The preset penalty coefficient is determined based on the importance of view occlusion and skyline features; when the constraint is violated, And the larger the deviation, The smaller the value.

[0092] Next, the target fitness is adjusted by adding penalties:

[0093] Fitness penalty factor The initial target fitness is applied to the elite individual; since the planning optimization objective of this invention is to maximize the total amount of development and construction. And maximizing the uniformity of spatial layout The corrected target fitness values ​​are respectively and By reducing the target fitness value of elite individuals who violate the constraints, they are disadvantaged in the non-dominant ranking and tournament selection of subsequent population evolution, thereby reducing their survival probability.

[0094] This step involves performing 3D ray tracing verification on elite individuals in the current population that are not at the dominant front, achieving high-precision verification of landscape constraint indicators and correction of target fitness. During verification, spatial hierarchical bounding boxes are used to accelerate the calculation of intersections between rays and the 3D model of the building complex, thereby accurately obtaining the 3D visual field occlusion degree of elite individuals. The topological fit of the skyline contour is quantified by calculating the Fraser distance through skyline reconstruction. Subsequently, the verified true landscape constraint indicators replace the approximate predicted values, and fitness is dynamically adjusted according to the degree of deviation from the constraints. A penalty mechanism is introduced to reduce the survival probability of individuals that violate landscape requirements in the population. This method effectively corrects approximate mapping prediction errors, ensures the directionality and reliability of the optimization process, and improves the accuracy and feasibility of the final planning scheme in terms of visual corridor protection and skyline coordination, achieving efficient coordination between landscape protection and multi-objective optimization.

[0095] The fifth step of the intelligent urban landscape planning and decision-making method based on multi-objective optimization algorithms is as follows: S5, the corrected elite individuals are used as new samples to iteratively update the approximate mapping relationship, and the next generation of population is used as the new current population. The process returns to step S3 until the preset convergence condition is met, outputting a set of intelligent urban landscape planning and decision-making schemes. By continuously updating the mapping model through feedback from elite samples, the accuracy of landscape constraint prediction is dynamically improved, ensuring that the optimization process balances computational efficiency and evaluation accuracy. Ultimately, a high-quality planning scheme that simultaneously meets landscape protection requirements, space utilization needs, and development and construction goals is obtained.

[0096] As a further step, in S5, the refined elite individuals are used as new samples to iteratively update the approximate mapping relationship, specifically including:

[0097] S51, Construct and add new samples: Extract the multi-dimensional decision vector of the elite individuals after fitness correction as input features, and extract the real three-dimensional view occlusion degree and the topological fit of the skyline contour obtained by the three-dimensional ray tracing verification in S4 as output labels; combine the input features and output labels into new sample data pairs and add them to the historical sample library used to train the surrogate model; the historical sample library contains the initial training set samples generated in S2 and the verification samples returned from each iteration.

[0098] S52, Monitor sample library capacity and trigger pruning: Monitor the total number of samples in the historical sample library in real time, and trigger pruning when the total number of samples exceeds the preset sample capacity threshold. At that time, the sample library pruning mechanism is triggered.

[0099] S53, Redundant Sample Removal Based on K-means Clustering: Input features (i.e., multidimensional decision vectors) of all samples in the historical sample database are extracted to form a feature matrix. The Min-Max normalization method is then used to normalize the feature matrix. The calculation formula is as follows: ,in These are the original eigenvalues. and These represent the maximum and minimum values ​​on the corresponding feature dimensions. These are the normalized eigenvalues.

[0100] The K-means clustering algorithm is used to perform cluster analysis on the normalized feature matrix, and the target number of clusters K is set to be less than the sample size threshold. integers (preferably) (70% to 80% of the sample bank capacity is pruned to ensure that the sample bank capacity returns to a safe range after pruning, and to leave room for subsequent addition of samples).

[0101] For each cluster generated by clustering, calculate the Euclidean distance from the normalized feature vector of each sample in the cluster to the center of the cluster. Keep the sample closest to the cluster center as the representative sample and remove the other redundant samples in the same cluster. After pruning, the total number of samples retained in the historical sample library is equal to the number of clusters K.

[0102] S54, Retrain and Update Approximate Mapping Relationship: Using the input features and output labels in the sample library retained after pruning, based on the radial basis function neural network or kriging interpolation model structure and hyperparameter settings determined in S2, the surrogate model is retrained and the model weight parameters are updated to complete the iterative update of the approximate mapping relationship.

[0103] It should also be noted that in S5, until the preset convergence conditions are met, a set of intelligent urban landscape planning decision schemes is output, specifically including:

[0104] Using the next generation of the population as the new current population, return to step S3 for iterative execution; at the end of each iteration, calculate the average rate of change of target fitness of all feasible individuals satisfying the constraints on the first non-dominated front of the current population. The calculation formula is: ;in, Let the current iteration algebra be... To optimize the total number of objectives, For the first The first non-dominated frontier feasible individual in the first generation The average fitness value on each objective, To prevent extremely small positive numbers with a denominator of zero (values ​​of...) ); Determine if the current iteration number has reached the maximum iteration number. or the average rate of change of target fitness The convergence accuracy is less than the preset value for g consecutive generations. .

[0105] If so, the preset convergence condition is met, and the iteration is terminated. Extract all elite individuals in the current population that meet the landscape constraints and are in the first non-dominant frontier. According to the real number encoding rules in S2, decode their multidimensional decision vectors into specific spatial control parameters for each building plot, such as building height, base area ratio, building setback distance, and volume rotation angle. This constitutes a set of intelligent urban landscape planning decision schemes for output.

[0106] Step S5 continuously feeds back elite individuals, validated by real-world landscape constraints and corrected for fitness, to the surrogate model training process. This enables adaptive iterative updates of the approximate mapping relationship, allowing the surrogate model to continuously learn from high-quality sample information generated during optimization, thereby gradually improving the accuracy and reliability of landscape constraint index predictions. Simultaneously, by clustering and pruning the historical sample database, representative samples are retained while redundant data is removed. This effectively controls the sample size while ensuring sample diversity and model training quality, reducing the computational overhead of model updates. Through the co-evolution of the surrogate model and the optimization process, search efficiency and optimization accuracy are continuously improved, accelerating algorithm convergence and ultimately obtaining a high-quality intelligent urban landscape planning decision set that balances planning objectives and landscape protection requirements.

[0107] From the overall process perspective, the steps of this invention are not executed independently, such as... Figure 2The diagram shown is an overall logic diagram of the intelligent urban landscape planning and decision-making method based on multi-objective optimization algorithm provided in this application embodiment. In this invention, steps S1 to S5 are interconnected and work together to form a closed-loop optimization mechanism of "spatial perception modeling - landscape constraint learning - scheme optimization search - accurate verification feedback - model adaptive update". Step S1 constructs a three-dimensional visual sensitive field to provide a unified spatial expression basis for the protection of visual corridors and the analysis of landscape visibility in the planning area. Step S2 obtains landscape constraint indicators based on the three-dimensional visual sensitive field and establishes an approximate mapping relationship between the spatial parameters of building groups and landscape constraint indicators to achieve rapid prediction of complex landscape constraints and provide support for the evaluation of subsequent large-scale planning schemes. Step S3 uses the approximate mapping relationship to quickly evaluate the planning schemes and conducts multi-objective optimization search in conjunction with the planning objectives to achieve synergistic optimization of landscape protection requirements and construction and development needs. Step S4 performs high-precision three-dimensional ray tracing verification on the non-dominant frontier elite individuals generated during the optimization process to correct the landscape constraint evaluation results, thereby ensuring the accuracy and reliability of the evaluation results of key planning schemes. Step S5 further feeds back the real verification results to update the approximate mapping relationship, so that the model can continuously learn new sample information and continuously improve the prediction accuracy during the iteration process, thereby having a feedback effect on the subsequent optimization process.

[0108] Therefore, this invention achieves the organic integration of three-dimensional spatial analysis, landscape constraint modeling, multi-objective optimization solution, and model self-learning update through data transfer, result feedback, and iterative updates between steps S1 to S5. Compared to the existing technology where landscape analysis and planning optimization are separated, and constraint evaluation and scheme generation are carried out independently, this invention can significantly reduce computational overhead and improve planning decision-making efficiency while ensuring the authenticity of landscape constraints and the reliability of planning results. It achieves a dynamic balance between urban landscape protection goals and urban construction and development needs, and has stronger engineering practicality and intelligent decision-making capabilities.

[0109] Overall, addressing the technical challenges of existing technologies in handling highly nonlinear landscape constraints such as visual corridor protection and skyline control—namely, the difficulty in balancing constraint modeling accuracy and solution efficiency, and the tendency for planning decisions to be conservative or time-consuming—this invention uses the precise quantification of dynamic features of the visual gradient in a three-dimensional visual sensitive field as a spatial foundation. It achieves efficient global optimization by constructing an approximate mapping relationship between building cluster spatial parameters and landscape constraint indicators. Furthermore, it innovatively introduces an elite verification and closed-loop feedback update mechanism based on ray tracing to ensure high fidelity in the optimization direction. This invention effectively overcomes the conservatism in decision-making caused by traditional dimensionality reduction simplification and resolves the computational bottleneck caused by direct three-dimensional calculations. While ensuring the accuracy of refined landscape management, it maximizes the potential of construction space, achieving a synergistic improvement in urban landscape protection and space utilization efficiency. This provides scientific and efficient technical support for the intelligent generation and decision-making of high-quality urban landscape planning schemes.

[0110] Through the above description of the implementation methods, those skilled in the art can clearly understand that, for the sake of convenience and brevity, only the division of the above functional modules is used as an example. In actual applications, the above functions can be assigned to different functional modules as needed, that is, the internal structure of the above functions can be divided into different functional modules to complete all or part of the functions described above.

[0111] In the embodiments provided in this application, it should be understood that the disclosed systems and methods can be implemented in other ways. For example, the embodiments described above are merely illustrative; for instance, the division of modules or units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another device, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces, or indirect coupling or communication connection between devices or units, and may be electrical, mechanical, or other forms.

[0112] The units described as separate components may or may not be physically separate. A component shown as a unit can be one or more physical units, located in one place or distributed in multiple different locations. Some or all of the units can be selected to achieve the purpose of this embodiment, depending on actual needs.

[0113] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0114] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a readable storage medium. Based on this understanding, the technical solutions of the embodiments of this application, or the parts that contribute to the solution, or all or part of the technical solution, can be embodied in the form of a software product. This software product is stored in a storage medium and includes several instructions to cause a device (which may be a microcontroller, chip, etc.) or processor to execute all or part of the steps of the methods of the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0115] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions within the technical scope disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A smart urban landscape planning and decision-making method based on multi-objective optimization algorithms, characterized in that, Includes the following steps: S1. Acquire the three-dimensional spatial data of the planning area and perform voxel segmentation. Based on the spatial relative position between the preset viewpoint and the landscape target, calculate the occlusion weight of each voxel grid on the visual corridor and construct a three-dimensional visual sensitive field. S2, Encode the spatial parameters of the building complex in the planning area to generate an initial planning scheme population; A portion of individual schemes were extracted from the initial planning scheme population as reference scheme samples. The 3D visual field occlusion degree and the topological fit of the skyline contour of the reference scheme samples under the 3D visual sensitive field were calculated as style constraint indicators. Based on the coding features of the reference scheme samples and the style constraint indicators, an approximate mapping relationship between the spatial parameters of the building group and the style constraint indicators was established. S3, take the initial planning scheme population as the current population, predict the landscape constraint index of each planning scheme individual in the current population based on the approximate mapping relationship, calculate the target fitness of each planning scheme individual in combination with the preset planning optimization goal, perform non-dominated sorting of the current population according to the target fitness and the constraint conditions corresponding to the landscape constraint index, and perform population crossover and mutation operations to generate the next generation population. S4. During the iterative evolution of the population, the elite individuals in the current population that are not at the dominant frontier are extracted. The three-dimensional ray tracing algorithm is used to perform three-dimensional view occlusion and topological fit verification on the elite individuals to obtain the real landscape constraint index. The real landscape constraint index is then used to correct the target fitness of the elite individuals. S5, the corrected elite individuals are used as new samples to iteratively update the approximate mapping relationship, and the next generation population is used as the new current population. Then, return to step S3 until the preset convergence condition is met, and output the set of intelligent urban landscape planning decision schemes.

2. The intelligent urban landscape planning and decision-making method based on multi-objective optimization algorithm as described in claim 1, characterized in that, In S1, the construction of the three-dimensional visual sensitive field specifically includes: A viewpoint section is generated based on the connection between the preset viewpoint and the landscape target, and the three-dimensional envelope of the viewpoint section is calculated. The three-dimensional envelope is divided into voxels. Based on the spatial distance from the center point of each voxel mesh to the connecting line and the normal angle, the basic line-of-sight occlusion sensitivity of each voxel mesh is calculated. By combining the importance weights of preset viewpoints, the basic line-of-sight occlusion sensitivity is normalized to obtain the line-of-sight occlusion weights of each voxel mesh. A three-dimensional visual sensitive field is formed by voxel meshes carrying view occlusion weights.

3. The intelligent urban landscape planning and decision-making method based on multi-objective optimization algorithm as described in claim 1, characterized in that, In S2, encoding the spatial parameters of the building clusters in the planning area to generate an initial planning scheme population specifically includes: Each building plot within the planning area is treated as an independent coding unit; The building height, base area ratio, building setback distance, and volume rotation angle of each building plot are extracted as spatial control parameters. Using a real-number encoding method, the spatial control parameters of all building plots are sequentially spliced ​​and combined into a multi-dimensional decision vector to form a chromosome sequence representing the individual initial planning schemes.

4. The intelligent urban landscape planning and decision-making method based on multi-objective optimization algorithm as described in claim 1, characterized in that, In S2, the fitting process to establish an approximate mapping relationship between the spatial parameters of the building complex and the style constraint index specifically includes: The Latin hypercube sampling method was used to uniformly select individuals from the initial planning scheme population as a reference scheme sample. The multidimensional decision vectors of each reference scheme sample are used as input features, and the calculated three-dimensional view occlusion degree and the topological fit of the skyline contour are used as output labels. Construct a surrogate model based on radial basis function neural network or kriging interpolation, use input features and output labels to train the surrogate model, and use the converged surrogate model as an approximate mapping relationship.

5. The intelligent urban landscape planning and decision-making method based on multi-objective optimization algorithm as described in claim 4, characterized in that, The calculation method for the three-dimensional view occlusion degree includes: Obtain the solid voxel mesh occupied by the 3D building complex model corresponding to the reference scheme sample in the 3D visual sensitive field; The cumulative occlusion weights are calculated by summing the line-of-sight occlusion weights corresponding to the entity voxel meshes. The ratio of the cumulative occlusion weight to the total weight of all voxel meshes in the 3D visual sensitive field is calculated, and the calculated ratio is used as the 3D view occlusion degree.

6. The intelligent urban landscape planning and decision-making method based on multi-objective optimization algorithm as described in claim 4, characterized in that, The calculation method for the topological fit of the skyline contour includes: Extract the edge projection lines of the 3D model outline of the building complex in the reference scheme sample onto the preset projection plane, and use them as the generated skyline; Obtain the skyline of the historical district corresponding to the landscape target as a reference skyline; Calculate the similarity of curvature changes between the generated skyline and the reference skyline at key inflection points, as well as the spatial misalignment distance between peak and trough nodes; By combining the similarity of curvature changes and the spatial misalignment distance, the morphological differences between the two are quantitatively assessed, and the topological fit of the skyline outline is obtained.

7. The intelligent urban landscape planning and decision-making method based on multi-objective optimization algorithm as described in claim 1, characterized in that, In S3, the non-dominated ranking of the current population based on the target fitness and the constraints corresponding to the landscape constraint index specifically includes: Based on the predicted landscape constraint index, assess the degree of violation of the constraints by individuals of each planning scheme in the current population. When performing a non-dominated ranking comparison of any two individuals, if both meet the constraints, a Pareto dominance comparison is performed based on the target fitness. If one individual satisfies the constraint while the other violates the constraint, then the individual that satisfies the constraint is determined to dominate the individual that violates the constraint. If both violate the constraints, the individual with the smaller degree of violation is determined to dominate the individual with the larger degree of violation.

8. The intelligent urban landscape planning and decision-making method based on multi-objective optimization algorithm as described in claim 1, characterized in that, In S4, the use of a 3D ray tracing algorithm to perform 3D view occlusion and topological fit verification on elite individuals to obtain accurate landscape constraint indicators specifically includes: Construct a spatial hierarchical bounding box data structure for the 3D geometric model of the building complex corresponding to elite individuals; Using a preset viewpoint as the origin of ray emission, a cluster of three-dimensional rays is uniformly emitted towards the landscape target. The spatial hierarchy bounding box data structure is used to accelerate the process of calculating the intersection of the three-dimensional ray clusters with the surface patches of the three-dimensional geometric model of the building complex. The true occlusion degree of the three-dimensional viewport is determined based on the proportion of blocked rays. The real skyline is reconstructed based on the 3D coordinates of the intersection points of the 3D ray clusters and the building complex, and the Fraser distance between the real skyline and the reference skyline is calculated. The Fraser distance is then mapped to the topological fit of the real skyline contour.

9. The intelligent urban landscape planning and decision-making method based on multi-objective optimization algorithm as described in claim 1, characterized in that, In S4, the method of using real appearance constraint indicators to correct the target fitness of elite individuals specifically includes: Replace the landscape constraint indicators predicted by elite individuals based on approximate mapping relationships with the real landscape constraint indicators obtained through verification. Determine whether the actual landscape constraint indicators violate the preset constraint conditions. If they do, dynamically generate a fitness penalty factor based on the deviation range exceeding the constraint range. A fitness penalty factor is applied to the original target fitness of the elite individual to make a penalty correction, thereby reducing the survival probability of the elite individual that violates the constraint in subsequent population evolution.

10. The intelligent urban landscape planning and decision-making method based on multi-objective optimization algorithm as described in claim 1, characterized in that, In S5, the step of iteratively updating the approximate mapping relationship by using the corrected elite individuals as new samples specifically includes: The encoded features of elite individuals that have completed fitness correction are combined with real appearance constraint indicators to form new samples, which are then added to the historical sample library used to train the surrogate model. The total number of samples in the historical sample database is monitored in real time. When the total number of samples exceeds a preset threshold, the K-means clustering algorithm is used to perform cluster analysis on the sample features in the historical sample database. The representative samples closest to each cluster center are retained, and the remaining redundant samples within the same cluster are removed, in order to prune the historical sample database. The surrogate model is retrained using the sample library retained after pruning, and the approximate mapping relationship is updated.

Images