Vector building element overlapping conflict processing method based on deep reinforcement learning

Abstract

The present invention belongs to the field of data processing technology and provides a method for handling overlapping conflicts of vector building elements based on deep reinforcement learning, comprising: determining and constructing a constrained delaunary triangulation and skeleton line segments between each vector building element in a target spatial area to generate a spatial conflict detection field model; connecting each vector building element with the center point of other surrounding vector building elements, and combining the spatial conflict detection field model to determine abnormal vector building elements with overlapping conflicts; determining the current state information of the abnormal vector building elements, grouping the data according to their area size to set constraint condition parameters and action reward information for each group, and combining the DQN model to obtain action prediction information; and based on the action prediction information, controlling the abnormal vector building elements to execute a position movement action until they reach the optimal position. The solution provided by the present invention improves the efficiency, accuracy, and feasibility of handling overlapping conflicts.

Description

Technical Field

[0001] The present invention relates to the field of data processing technology, and in particular to a method for handling overlapping conflicts of vector building elements based on deep reinforcement learning. Background Art

[0002] In the field of cartography and geographic information systems, to accommodate the display needs of maps at different scales, multiple cartographic synthesis algorithms and diverse cartographic symbols are required. However, these operations often lead to spatial overlap between geographic features such as vector buildings, resulting in confusing map representations.

[0003] Traditional shift algorithms for addressing spatial overlap conflicts have numerous drawbacks in practical applications. On the one hand, they require large amounts of computation and suffer from low processing efficiency, making them inefficient for rapidly producing large amounts of data. On the other hand, they are unable to effectively resolve the secondary conflicts caused by shift operations, presenting significant limitations in their applicability and preventing the rational allocation of map elements.

[0004] Therefore, the traditional vector building element overlapping conflict processing solution has technical problems such as low processing efficiency, insufficient processing accuracy and poor feasibility. Summary of the Invention

[0005] The present invention provides a method for handling overlapping conflicts of vector building elements based on deep reinforcement learning, which is used to solve the defects of traditional vector building element overlapping conflict handling solutions, such as low processing efficiency, insufficient processing accuracy and poor feasibility.

[0006] The present invention provides a method for handling overlapping conflicts of vector building elements based on deep reinforcement learning, the method comprising:

[0007] Determine and construct the constraint Delaunary triangulation and skeleton segments between each vector building element in the target spatial area to generate a spatial conflict detection field model;

[0008] Connect each vector building feature with the center points of other surrounding vector building features to obtain connecting line segments, and combine them with the spatial conflict detection field model to determine abnormal vector building features with overlapping conflicts;

[0009] Determine the current state information of the abnormal vector building elements, group the data according to the area size of the abnormal vector building elements to set constraint parameters and action reward information for each group, and combine with the pre-built DQN model to obtain action prediction information;

[0010] According to the motion prediction information, the abnormal vector building element is controlled to perform a position movement action until it reaches an optimal position.

[0011] According to the vector building element overlapping conflict processing method based on deep reinforcement learning provided by the present invention, the constrained Delaunary triangulation between each vector building element in the target space area is determined, including: node encryption processing of all edges existing in the target space area, and plane Delaunary triangulation construction between each vector building element to obtain a constrained Delaunary triangulation.

[0012] According to the vector building element overlapping conflict processing method based on deep reinforcement learning provided by the present invention, the skeleton line segments between each vector building element in the target space area are determined, including: determining the number of adjacent triangles of each triangle in the constrained Delaunary triangulation network, and dividing all triangles in the constrained Delaunary triangulation network into three types based on the number of adjacent triangles; traversing all triangles, and determining the segmentation curve of the blank area between each vector building element according to the type corresponding to each triangle, and using the segmentation curve as the skeleton line segment in the target space area.

[0013] According to the vector building element overlapping conflict processing method based on deep reinforcement learning provided by the present invention, the abnormal vector building element with overlapping conflict is determined, including: judging whether the connecting line segment intersects with the skeleton line segment in the spatial conflict detection field model, or whether it completely intersects with any triangle in the constrained Delaunary triangulation in the spatial conflict detection field model, to obtain a first judgment result; based on the first judgment result, the abnormal vector building element with overlapping conflict is determined.

[0014] According to the method for handling overlapping conflicts of vector building elements based on deep reinforcement learning provided by the present invention, based on the first judgment result, abnormal vector building elements with overlapping conflicts are determined, including: if the first judgment result is not completely yes, it is determined that there is an overlapping conflict between the vector building elements, and the abnormal vector building elements with overlapping conflicts are obtained.

[0015] According to the deep reinforcement learning-based vector building element overlapping conflict processing method provided by the present invention, the DQN model includes: an online network, which is used to predict the expected reward value corresponding to each alternative action through the original fixed parameters, and output action prediction information based on the expected reward value; a target network, which is used to generate a target reward value through the current fixed parameters, and the parameters of the target network are copied from the online network according to a set period.

[0016] According to the method for handling overlapping conflicts of vector building elements based on deep reinforcement learning provided by the present invention, the constraint parameters of each group are set, including: determining the map boundary constraint based on the map boundary corresponding to the target spatial area; determining the moving buffer constraint based on the distance between the abnormal vector building element and the surrounding building complex and the distribution pattern of the surrounding building complex; determining the standard distance range between the abnormal vector building element and other buildings and determining the moving distance constraint based on the provisions of the mapping specifications; determining the secondary conflict constraint based on the state of overlapping conflicts between the abnormal vector building element and other vector building elements during the movement process; and using the map boundary constraint, the moving buffer constraint, the moving distance constraint and the secondary conflict constraint as the constraint parameters of each group.

[0017] According to the vector building element overlapping conflict processing method based on deep reinforcement learning provided by the present invention, a moving buffer constraint condition is determined based on the distance between the abnormal vector building element and the surrounding building complex and the distribution pattern of the surrounding building complex, including: if the distribution pattern of the surrounding building complex is a regular pattern, the minimum distance between the abnormal vector building element and the adjacent building elements in the surrounding building complex is determined, and a first moving buffer with a side length of three times the minimum distance and capable of accommodating twice the size of the abnormal vector building element itself is constructed to obtain a moving buffer constraint condition under a regular distribution; if the distribution pattern of the surrounding building complex is an irregular pattern, the maximum distance between the abnormal vector building element and the adjacent building elements in the surrounding building complex is determined, and a second moving buffer with a side length of four times the maximum distance and capable of accommodating twice the size of the abnormal vector building element itself is constructed to obtain a moving buffer constraint condition under an irregular distribution.

[0018] According to the vector building element overlapping conflict processing method based on deep reinforcement learning provided by the present invention, the action reward information of each group is set, including: judging whether the abnormal vector building element satisfies the map boundary constraint condition during the movement process, obtaining a second judgment result, and calculating the boundary constraint reward value in the current state according to the second judgment result; judging whether the abnormal vector building element satisfies the moving buffer constraint condition during the movement process, obtaining a third judgment result, and calculating the buffer constraint reward value in the current state according to the third judgment result; judging whether the abnormal vector building element satisfies the moving distance constraint condition during the movement process, obtaining a fourth judgment result, and calculating the distance constraint reward value in the current state according to the fourth judgment result; judging whether the abnormal vector building element satisfies the secondary conflict constraint condition during the movement process, obtaining a fifth judgment result, and calculating the secondary constraint reward value in the current state according to the fifth judgment result; summing the boundary constraint reward value, the buffer constraint reward value, the distance constraint reward value and the secondary constraint reward value to obtain the action reward information of each group.

[0019] According to the vector building element overlapping conflict processing method based on deep reinforcement learning provided by the present invention, the abnormal vector building element is controlled to perform a position movement action until the optimal position according to the action prediction information, including: determining the target action of the abnormal vector building element according to the action prediction information; wherein the target action is one of a plurality of alternative movement schemes obtained by pre-uniform division; according to the target action, the abnormal vector building element is controlled to move at equal distances until the optimal position.

[0020] The present invention provides a method for handling overlapping conflicts of vector building elements based on deep reinforcement learning. The method generates a spatial conflict detection field model by determining and constructing a constrained delaunary triangulation and skeleton line segments between each vector building element in a target spatial area. Each vector building element is connected to the center point of other surrounding vector building elements to obtain connecting line segments. In combination with the spatial conflict detection field model, overlapping conflicts are detected for the vector building elements to identify abnormal vector building elements with overlapping conflicts. The current state information of the abnormal vector building elements is determined, and data is grouped according to their area size to set constraint parameters and action reward information for each group. In combination with a pre-built DQN model, action prediction information is obtained. Based on the action prediction information, the abnormal vector building elements are controlled to execute position movement actions until they are at the optimal position. Since the abnormal vector building elements are treated as intelligent agents in the overlapping conflict handling process, the DQN model enables the abnormal vector building elements to achieve efficient resolution of overlapping conflicts through autonomous interaction with the environment. This not only improves the efficiency and accuracy of overlapping conflict handling, but also makes the overlapping conflict handling process easier to implement, thereby improving the feasibility of the overlapping conflict handling process. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] In order to more clearly illustrate the technical solutions in the present invention or the prior art, a brief introduction is given below to the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0022] Figure 1 1 is a flow chart of a method for handling overlapping conflicts of vector building elements based on deep reinforcement learning provided by an embodiment of the present invention;

[0023] Figure 2 It is a schematic diagram of the structure of the constrained Delaunary triangulation and skeleton segments in different situations;

[0024] Figure 3 It is a schematic diagram of the overlap conflict demonstration, conflict detection representation and conflict resolution;

[0025] Figure 4 This is a schematic diagram of the implementation principle of the entire capping conflict processing link;

[0026] Figure 5 is a schematic diagram of the agent's movement direction;

[0027] Figure 6 It is a schematic diagram of four buffer zones obtained based on the four neighboring building features around the intelligent agent;

[0028] Figure 7 This is a schematic diagram of the principle of determining the first moving buffer zone in the rule mode;

[0029] Figure 8 This is a schematic diagram of the principle of determining the second moving buffer zone in the irregular mode;

[0030] Figure 9 It is a schematic diagram of the identification status of the standard distance range;

[0031] Figure 10 This is a schematic diagram of the principle for determining the maximum distance before and after the agent moves;

[0032] Figure 11 It is a schematic diagram of sub-regions obtained by dividing a target space area;

[0033] Figure 12 This is a schematic diagram of the overlapping conflict detection results of the three sub-areas;

[0034] Figure 13 This is a schematic diagram of the results of screening and numerically labeling the overlapping conflicts detected in each sub-area;

[0035] Figure 14 This is a schematic diagram showing the results of the two intelligent entities with the largest area among the six groups in the entire study area;

[0036] Figure 15 This is a diagram showing the DQN calculation results for the two agents with the largest area among the six groups in the entire study area;

[0037] Figure 16 This is a diagram showing the changes in the reward function obtained by statistically analyzing the reward values ​​obtained during the DQN model calculation process.

[0038] Figure 17 This is a diagram showing the change in the loss function obtained by statistically analyzing the loss values ​​obtained during the DQN model calculation process;

[0039] Figure 18 This is a schematic diagram of the results of extracting area features in the study area and performing topological analysis on existing spatial overlapping conflicts using the ArcGIS 10.8 software platform.

[0040] Figure 19 yes Figure 18 Schematic diagram of the selected area;

[0041] Figure 20 Yes Figure 18 Schematic diagram of the result of spatial overlapping conflict processing on the selected part;

[0042] Figure 21It is a schematic diagram showing the existence of multiple overlapping conflicts between single buildings and the elimination of corresponding overlapping conflicts. DETAILED DESCRIPTION

[0043] To make the objectives, technical solutions, and advantages of the present invention more clear, the technical solutions of the present invention will be clearly and completely described below in conjunction with the accompanying drawings. Obviously, the embodiments described are only some of the embodiments of the present invention, not all of them. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts shall fall within the scope of protection of the present invention.

[0044] The following combination Figures 1 to 21 Describe the details of the method for handling overlapping conflicts of vector building elements based on deep reinforcement learning provided by an embodiment of the present invention.

[0045] Figure 1 It is a flow chart of a method for handling overlapping conflicts of vector building elements based on deep reinforcement learning provided by an embodiment of the present invention.

[0046] like Figure 1 As shown, the method for handling overlapping conflicts of vector building elements based on deep reinforcement learning provided by an embodiment of the present invention mainly includes the following steps:

[0047] Step 101: Determine and construct the constraint delaunary triangulation and skeleton line segments between each vector building element in the target spatial area to generate a spatial conflict detection field model.

[0048] It can be understood that the construction of the spatial conflict detection field model provides an effective processing basis for the detection of spatial overlapping conflicts.

[0049] Step 102: Connect each vector building element with the center points of other surrounding vector building elements to obtain connecting line segments, and combine them with the spatial conflict detection field model to determine abnormal vector building elements with overlapping conflicts.

[0050] In this embodiment, the geometric positions of the connecting line segments and the triangles and skeleton line segments in the spatial conflict detection field model may be compared, thereby realizing overlapping conflict detection.

[0051] Step 103: Determine the current state information of the abnormal vector building element, group the data according to the area size of the abnormal vector building element to set the constraint parameters and action reward information of each group, and combine the pre-built DQN model to obtain action prediction information.

[0052] In this embodiment, data can be grouped according to the area scale of abnormal vector building elements, and then related constraint parameters and action reward information can be set.

[0053] Step 104: Based on the motion prediction information, control the abnormal vector building element to perform a position movement action until it reaches the optimal position.

[0054] In one embodiment, determining the constrained delaunary triangulation between the vector building elements in the target spatial area specifically includes:

[0055] First, the nodes of the edges of each vector building element in the target spatial area are encrypted, and then a planar Delaunary triangulation is constructed between each vector building element to obtain a constrained Delaunary triangulation.

[0056] It's understandable that the spatial conflict detection field model is based on a constrained Delaunay triangulation (TDM) from computational geometry. This TDM involves two key aspects: First, the circumcircle of each triangle must not contain any other points within the surface. Second, the diagonals of the convex quadrilateral formed by every two adjacent triangles must be interchanged so that the minimum of the six interior angles does not increase.

[0057] In practice, too few points will result in the appearance of narrow triangles in the constructed Delaunay triangulation, for example Figure 2 In the case shown in sub-image a, the proximity relationship between vector building features cannot be correctly detected.

[0058] To avoid the above defects, this embodiment needs to encrypt the nodes of the edges of the vector building elements before constructing the constrained Delaunary triangulation. For details, please refer to Figure 2 In the b subgraph, this rule avoids the generation of narrow triangles in the triangle generation. The constrained Delaunary triangulation generated after node encryption can be seen in Figure 2 Subgraph c in the figure.

[0059] In one embodiment, determining skeleton line segments between vector building elements in the target spatial region specifically includes:

[0060] First, the number of adjacent triangles of each triangle in the constrained Delaunary triangulation is determined. Based on the number of adjacent triangles, all triangles in the constrained Delaunary triangulation are divided into three types, namely, triangles adjacent to one triangle, two triangles, and three triangles. Different types of triangles construct different skeleton segments.

[0061] Then, all triangles are traversed, and according to the type of each triangle, the segmentation curve of the blank area between each vector building element is determined, and the segmentation curve is used as the skeleton segment of the target space.

[0062] In one embodiment, determining abnormal vector building elements with overlapping conflicts specifically includes:

[0063] First, it is determined whether the connecting line segment intersects with the skeleton line segment in the spatial conflict detection field model, or whether it completely intersects with any triangle in the constrained Delaunary triangulation in the spatial conflict detection field model, to obtain a first judgment result.

[0064] Finally, based on the first judgment result, abnormal vector building elements with overlapping conflicts are determined.

[0065] In a specific implementation, determining abnormal vector building elements with overlapping conflicts based on the first judgment result specifically includes:

[0066] If the first judgment result is not completely yes, it is determined that there is an overlapping conflict between the vector building elements, and an abnormal vector building element with an overlapping conflict is obtained.

[0067] It is understandable that for the overlapping conflicts existing in the space, such as Figure 3 In the case shown in sub-graph e, for the extraction of conflicting line segments, if the connecting line segment between the two center points does not intersect with the skeleton line segment or does not completely intersect with the triangle, it can be determined that there is an overlapping conflict between the abnormal vector building elements corresponding to the two center points. The detection process can be seen in Figure 3 The situation after resolving the conflict is as follows Figure 3 As shown in sub-figure g.

[0068] In one embodiment, the DQN model specifically includes:

[0069] The online network is used to predict the expected return value corresponding to each alternative action through the original fixed parameters, and output action prediction information based on the expected return value.

[0070] The target network is used to generate the target reward value using the current fixed parameters, reducing the correlation between the predicted value and the target value during training. The parameters of the target network are copied from the online network at a set period (such as several rounds of iterations) to maintain the relative stability of the target value and achieve the purpose of stabilizing the training process.

[0071] In this embodiment, the DQN model can use the dual neural network structure to calculate the corresponding reward value for the action performed by the intelligent agent, store the experience gained from each step of the action, and finally update and fit the parameters of the dual network through experience replay and calculation of the loss function, thereby making the best judgment and guidance for the action of the intelligent agent, thereby achieving effective resolution of spatial overlapping conflicts.

[0072] The implementation principle of the entire capping conflict handling link can be found in Figure 4,like Figure 4 As shown in the figure, in the early stages, the field model is constructed using constrained delaunary triangulation and skeleton segments to obtain abnormal vector building features. Then, multiple constraints, such as map boundary constraints, movement buffer constraints, movement distance constraints, and secondary conflict constraints, as well as corresponding reward functions, are determined to guide the agent's movement and decision-making.

[0073] At time t, the agent is in the initial state S t Based on this, perform action a t After execution, it is necessary to determine whether the above constraints are met. If the above constraints are met, stop the next action, otherwise continue to select actions, and then calculate the total reward value r t and the next moment state information S t+1 , the experience fragment (S t ,a t ,r t ,S t+1 ) is stored in the experience pool for subsequent experience replay. During experience replay, experience fragments can be randomly extracted from the experience pool and provided to the current network and the target network for updating and fitting.

[0074] Then, the loss function is set Measure the difference between the expected return value and the target return value, and use the gradient descent method to update the parameters of the online network to reduce the loss and continuously optimize the current network prediction ability.

[0075] In one embodiment, based on the motion prediction information, controlling the abnormal vector building element to perform a position movement action until it reaches the optimal position specifically includes:

[0076] First, the target action of the abnormal vector building element is determined based on the action prediction information; wherein the target action is one of multiple alternative movement plans obtained by uniform division in advance.

[0077] Then, according to the target action, the abnormal vector building elements are controlled to move at equal distances until they reach the optimal position.

[0078] In this embodiment, to ensure that no deformation occurs during the movement, see Figure 5 In this embodiment, the building as a whole is regarded as the intelligent agent S, and its actions are deterministic. In this embodiment, the actions of the intelligent agent S are equidistant actions in eight uniform directions, as follows: Figure 5 As shown by the middle arrow, the distance is determined according to the drawing specifications.

[0079] At the same time, with the help of the DQN model The greedy strategy balances the proportion of exploration and utilization in the action. The agent takes each action with The probability of random exploration is The probability of is maximized with the help of the results predicted by the neural network, namely:

[0080] (1)

[0081] Among them, a t Represents the action of the agent in the current state, Indicates randomly selecting an action from the action space A. represents the probability, is a constant between 0 and 1, Indicates choosing the action that maximizes the Q (reward) value in the current state.

[0082] In this embodiment, the DQN model introduces a dual-network structure: an online network and a target network. This network predicts the agent's actions and continuously matches them to the actual action, thereby guiding each action. The online network is the neural network used to generate action selection probabilities during the agent's training process, providing a numerical basis for the agent's action selection.

[0083] The online network uses the experience replay mechanism to continuously extract a certain batch of experience data in each round to participate in the parameter update of gradient descent, and Controls the update step size.

[0084] The target network is a copy introduced in the DQN model and is mainly used to calculate the target Q value. The calculation formula is as follows:

[0085] (2)

[0086] Where, is the target Q value, The next state S for the target network t+1 The predicted maximum Q value, r t The total reward value obtained for the current action, is the discount factor used to control the weight of future rewards, Indicates the action at the next moment.

[0087] Using the different Q values ​​calculated by the dual networks, the loss function is calculated between the online network and the target network, continuously minimizing the error to gradually fit the true target Q value. The target network parameters are regularly updated synchronously with the online network to avoid training oscillations caused by frequent target value updates.

[0088] Constraints are limiting rules used to determine the optimal position of the agent during deep reinforcement learning. By setting multiple constraints on the DQN model, it is determined whether the agent's movement results reach the optimal position. The constraint parameters of this embodiment specifically include: map boundary constraints, movement buffer constraints, movement distance constraints, and secondary conflict constraints.

[0089] In one embodiment, taking any group as an example, the constraint condition parameters of each group are set, specifically including:

[0090] First, the map boundary constraints are determined based on the map boundary corresponding to the target spatial area.

[0091] The map sheet is the carrier of map data. During the movement of the intelligent agent, it is necessary to ensure that the intelligent agent is within the range of the map sheet and does not exceed the boundary of the map sheet.

[0092] Secondly, the moving buffer constraint conditions are determined based on the distance between the abnormal vector building elements and the surrounding buildings and the distribution pattern of the surrounding buildings.

[0093] Maintaining the distribution pattern between vector building elements is an important constraint in resolving overlapping conflicts. This embodiment mainly constrains the movement range of vector building elements in the target space area to ensure that the distribution of vector building elements after movement is within a reasonable range, avoiding excessive exploration of the space by the intelligent agent.

[0094] The specific approach is to detect multiple distances from each adjacent building element in the building complex around the intelligent agent to the intelligent agent, and select one of the minimum distances as the basis to create a buffer zone as the range of intelligent agent movement, and the range can accommodate at least two intelligent agents themselves, such as Figure 6 As shown, four buffer zones can be designed based on the four adjacent building elements around the agent, namely, the four areas numbered 201, 202, 203, and 204. However, only the buffer zone numbered 201 cannot accommodate the size of two agents in all directions. Therefore, the buffer zone numbered 201 is not constructed, while the other buffer zones are constructed. This embodiment adopts different calculation methods for regular and irregular building cluster distribution patterns.

[0095] In a specific implementation, the moving buffer constraint conditions are determined based on the distance between the abnormal vector building element and the surrounding buildings and the distribution pattern of the surrounding buildings, including:

[0096] If the distribution pattern of the surrounding building complex is a regular pattern, the movement distance and range of the vector building element are highly constrained, and the establishment of the buffer zone follows the principle of keeping the range as small as possible. The minimum distance between the abnormal vector building element and the adjacent building elements in the surrounding building complex is determined, and a first moving buffer zone is constructed with a side length of three times the minimum distance and capable of accommodating twice the size of the abnormal vector building element itself. The buffer zone range of this embodiment is determined by the intelligent agent with the largest single area and spatial overlapping conflict in the target spatial area, so as to ensure that the buffer zone established by the intelligent agent can be applied to other elements, thereby improving the operation efficiency of the algorithm.

[0097] like Figure 7 As shown in the figure, the buffer size is determined by the minimum distance d between the intelligent agent and the adjacent building elements in the surrounding building complex. A rectangular area with a side length of 2d is constructed with this minimum distance, and then expanded outward by 1.5 times on the basis of this rectangular area to construct a rectangular area with a side length of 3d that can accommodate twice the size of the abnormal vector building element itself as the first moving buffer.

[0098] If the distribution pattern of the surrounding buildings is irregular, the distribution position of the intelligent body after movement does not need to be too precise. In this case, the maximum distance between the abnormal vector building element and the adjacent building elements in the surrounding buildings is determined, and a second moving buffer is constructed with a side length of four times the maximum distance and capable of accommodating twice the size of the abnormal vector building element itself.

[0099] like Figure 8 As shown, for Figure 8 Among d1, d2, d3 and d4, the largest distance d1 is used as the basis for determining the buffer zone, and a rectangular buffer zone with a side length twice that of d1 is constructed. At the same time, it is expanded outward by 2 times on this basis to construct a rectangular buffer zone with a side length four times that of d1 and which can accommodate twice the size of the abnormal vector building element itself, namely the second moving buffer zone, so as to ensure that this range is suitable for most building complexes.

[0100] Thirdly, according to the provisions of the mapping specifications, the standard distance range between abnormal vector building elements and other buildings is determined, and the movement distance constraint conditions are determined.

[0101] In this embodiment, according to the cartographic standards, the spacing between graphic elements (such as lines and symbols) depends on the scale and complexity of the drawing. Usually, the standard distance range can be 3-5mm, so as to ensure that the map elements are clear and do not overlap at different scales. Figure 9 shown.

[0102] Therefore, during the movement process, this embodiment further constrains the distance of the intelligent agent's movement to ensure that the distance between the vector building elements after movement does not exceed the limit of the map drawing distance. At the same time, considering the timeliness and complexity of the algorithm operation, only the intelligent agent with overlapping conflicts and the largest area is calculated on the movement distance constraint so that it can meet the movement requirements of most intelligent agents.

[0103] In this embodiment, the rules for establishing the movement distance constraint are as follows: First, the maximum distance before and after the movement is calculated. The maximum distance D is composed of the distance d5 from the geometric center of the intelligent body to its farthest boundary and a value d6 in the standard distance range, as shown in the following example: Figure 10 As shown in , the maximum distance D before and after the agent moves can be specifically expressed as follows:

[0104] (3)

[0105] Fourthly, the secondary conflict constraint conditions are determined based on the state of overlapping conflicts between abnormal vector building elements and other vector building elements during movement.

[0106] When moving an intelligent agent with a spatial overlapping conflict, it will inevitably overlap with other surrounding intelligent agents to cause new spatial conflicts. Therefore, this embodiment constructs secondary conflict constraints for the intelligent agent and regards secondary conflicts as the most stringent constraints.

[0107] Finally, the map boundary constraints, moving buffer constraints, moving distance constraints, and secondary conflict constraints are used as constraint parameters for each group.

[0108] In one embodiment, taking any group as an example, action reward information for each group is set, specifically including:

[0109] First, it is determined whether the abnormal vector building element meets the map boundary constraint conditions during the movement process, and a second judgment result is obtained. The boundary constraint reward value in the current state is calculated according to the second judgment result.

[0110] In this embodiment, if the agent exceeds the map boundary, the reward value is set to -3, and if it does not exceed the boundary, the reward value is set to 0. In other words, the boundary constraint reward value can be expressed as follows:

[0111] (4)

[0112] Secondly, it is determined whether the abnormal vector building element meets the moving buffer constraint condition during the movement process, and a third judgment result is obtained, and the buffer constraint reward value in the current state is calculated according to the third judgment result.

[0113] Since the distribution pattern ensures the correctness and authenticity of the map data, once the agent's movement exceeds the range of the movement buffer, the reward value is set to -7 to guide it to continue moving to avoid exceeding the movement buffer. If it does not exceed the movement buffer, the reward value can be set to 0. Therefore, the buffer constraint reward value can be expressed as follows:

[0114] (5)

[0115] Thirdly, it is determined whether the abnormal vector building element satisfies the moving distance constraint condition during the moving process, and a fourth judgment result is obtained, and the distance constraint reward value in the current state is calculated according to the fourth judgment result.

[0116] If the agent moves beyond the specified distance after taking action, that is, if the movement distance constraint is not met, the reward value is set to -3. Otherwise, the reward value is set to 0. Therefore, the distance constraint reward value can be expressed as:

[0117] (6)

[0118] Fourthly, it is determined whether the abnormal vector building element satisfies the secondary conflict constraint condition during the movement process, and a fifth judgment result is obtained, and the secondary constraint reward value in the current state is calculated according to the fifth judgment result.

[0119] In this embodiment, the absolute value of the secondary generation reward value is greater than the reward value of other constraint items, so as to guide the agent to quickly learn to avoid the occurrence of secondary conflicts. When an overlapping conflict occurs, the reward value is set to -10, otherwise, the reward value is set to 0. Therefore, the secondary constraint reward value can be specifically expressed as:

[0120] (7)

[0121] Finally, the boundary constraint reward value, buffer constraint reward value, distance constraint reward value, and secondary constraint reward value are summed to obtain the action reward information of each group.

[0122] In this embodiment, all reward values ​​are added together. When all constraints are met, that is, when the constraint reward value obtained by the action is 0, the correct result can be output. Specifically, the action reward information (that is, the total reward value) It can be expressed as follows:

[0123] (8)

[0124] in, represents the boundary constraint reward value, represents the buffer constraint reward value, represents the distance constraint reward value, Represents the secondary constraint reward value.

[0125] The following takes a target spatial area as an example to explain the implementation process of the above scheme in detail. The spatial characteristics of the distribution of vector building elements in the target spatial area are as follows: Figure 11 As shown in the figure, Areas A and B have relatively regular building layouts and are classified as relatively regular areas; while Area C has a somewhat irregular building distribution and is classified as a relatively irregular area. DQN deep reinforcement learning calculations were performed on these two areas with different characteristics to fully verify the versatility and effectiveness of the DQN model in solving problems in different spatial forms.

[0126] For areas A and B with relatively regular building distribution and area C with irregular building distribution, this embodiment constructs constrained Delaunary triangulation and skeleton line segments for each sub-area respectively, thereby detecting spatial overlapping conflicts between vector building elements in the three sub-areas. The overlapping conflict detection results are as follows: Figure 12 shown.

[0127] In the entire study area, due to the unique architectural patterns of individual vector building elements, they are always overlapping each other during map zooming. This embodiment regards this situation as a reasonable existence and does not calculate it as an overlapping conflict. It needs to be manually deleted. The specific statistical results of the number of abnormal vector building elements with overlapping conflicts are shown in Table 1.

[0128] Table 1 Statistics of overlapping conflicts

[0129]

[0130] The overlap conflicts detected in each sub-area are screened and highlighted by numerical labels. The results can be seen in Figure 13 It is not difficult to find that the conflicts detected in the regular areas A and B can be seen in Table 1. For the irregular area C, after excluding reasonable overlapping conflicts, the number of elements with overlapping conflicts is 8, and the total number of overlapping conflicts detected in the three sub-areas A, B, and C is 23.

[0131] The selected abnormal vector building elements with overlapping conflicts are sorted by area to construct relevant parameters. For the grouping of regular areas and irregular areas, please refer to Table 2.

[0132] Table 2 Ranking and grouping of overlapping conflict elements in the study area

[0133]

[0134] Combined with the statistics of the area size of the overlapping conflict elements in the three sub-regions in Table 2, the relevant parameters within each group can be found in Table 3.

[0135] Table 3 Buffer constraints and movement distance constraints for regular and irregular areas

[0136]

[0137] The two intelligent entities with the largest area among the six groups in the entire study area are specifically displayed to provide a basis for judging the subsequent experimental results. For details, see Figure 14 .

[0138] For the regular areas of Area A and Area B and the irregular area of ​​Area C, DQN deep reinforcement learning calculations are performed in combination with the movement distance constraint and buffer range constraint parameters in Table 3. The specific parameters are shown in Table 4.

[0139] Table 4. DQN model parameter settings for regular and irregular regions

[0140]

[0141] For the agents in the study area, the maximum number of steps for each agent in each round is set to 15. Each agent with overlapping conflicts is moved in a fixed order, and the reward experience gained at each step is stored in the experience pool until all agents reach the upper limit of the number of steps or meet the constraint conditions as the action end point. Then, the initial layout is reset and the calculation of the next round is carried out to calculate the spatial overlapping conflicts. The DQN calculation results of some agents in the study area are as follows: Figure 15 As shown, the selected agent and Figure 14 The agents in the same arrangement and distribution are used to compare the effects.

[0142] By collecting the running result data of the DQN model, the number of feasible solutions output by the model and the running time loss are shown in Table 5.

[0143] Table 5 DQN calculation results for regular and irregular regions

[0144]

[0145] The reward and loss values ​​obtained from the DQN model calculation process are statistically analyzed. The changes in the reward and loss functions are shown in the following figure. Figure 16 and Figure 17 shown.

[0146] Through Figure 15 The analysis shows that for regular areas, the previous overlapping conflicts have been successfully eliminated. As shown in Table 5, by grouping the data to maximize the adaptation of the constraints, the model output of all groups of data is maintained at more than 1900 valid solutions, and the single round running time is always kept at a low level; for irregular areas, Figure 15 The calculation results still show that reinforcement learning is still feasible in dealing with spatial overlap conflicts in irregular areas. From the analysis in Table 5, we can see that the success rates of the two sets of data in irregular areas are maintained at a high level and have high computational efficiency.

[0147] Combine Figure 16 and Figure 17 As shown in the figure, from the changes in the reward function and the loss function, for the regular area, the reward functions of the four sets of data all gradually converge before 500 rounds, and the fluctuation of the reward value tends to be flat afterwards. Figure 17 From the changes in the loss function, we can see that the changes in the loss functions of the four groups of data are closely related to the changes in the reward function. The loss values ​​of the loss functions gradually decrease and fluctuate in a large range during the early action process, but then the loss values ​​of the four groups of data begin to decrease and gradually stabilize. Although the loss values ​​of Groups 1 and 3 in the regular area, and Group 1 in the irregular area did not eventually stabilize at a low level, according to the number of output valid solutions and the peak value of the reward function convergence in Table 5, it can be seen that the model maintains a high level of output of valid solutions.

[0148] Depend on Figure 17 It can be seen that in the processing of these three data sets, more actions were executed. Although the loss difference ultimately remained at a high level, combined with the number of reasonable output solutions in Table 5, this did not affect the output of the actual reasonable solution, and had little impact on the overall computational efficiency and performance. For the irregular region, analysis of the reward and loss function graphs reveals that the function changes for the two data sets are similar to those for the regular region. Similarly, the reward values ​​for both sets gradually converged after the first 500 rounds of the experiment, and then began to converge and level off. The difference is that the reward function of the first data set in the irregular region fluctuated more frequently, and its loss function converged more slowly. This corresponds to the amount of data it had to process. As shown in Table 5, this group processed the largest amount of data in the entire study area. This is because the larger the data volume, the more difficult it is for the agent to explore the correct position in the environment, the longer it takes, and the more frequent the reward value fluctuations. However, overall, this size of data did not have a substantial impact on the time loss of the model.

[0149] Overall, the DQN model shows a high success rate in resolving spatial overlapping conflicts, and its reward function converges quickly, proving that the application of the DQN model in this embodiment enables the intelligent agent to quickly select the target position through interaction with the environment in a single batch of small-scale iterations, demonstrating efficient processing capabilities, reflecting the high applicability of the DQN model in solving such problems, and reflecting a high level of technical application and practical value.

[0150] Regarding the traditional solution to spatial overlapping conflicts, this embodiment uses the ArcGIS 10.8 software platform to extract the surface features in the study area and perform topological analysis on the spatial overlapping conflicts in the study area. The platform detects a total of 1372 errors in the study area. The results are as follows: Figure 18 shown.

[0151] pass Figure 19 The irregular area of ​​the study area shown, i.e. Figure 18 The selected part in the process of spatial overlap conflict is processed, and the platform's built-in shift, deletion and other algorithms are used to process the irregular area. Figure 20 As shown. For Figure 19 Overlapping conflicts between multiple vector building features such as No. 3 and No. 5, as well as similar Figure 19 The conflict between the single buildings No. 1, No. 2 and No. 4 is handled, such as Figure 21 As shown in the three subgraphs i, j, and k, the conflict can be resolved manually by using the shift and deletion algorithm. For details, please refer to Figure 21 In the figure, there are three sub-maps: l, m, and n. However, the use of the deletion algorithm cannot guarantee the accuracy of the original map information. In addition, due to the inaccurate extraction of planar buildings by the platform, a large amount of redundant planar data is generated, as shown in Table 6.

[0152] Table 6 Comparison of detection and processing results

[0153]

[0154] Experimental comparison and data analysis show that this embodiment uses reinforcement learning to automatically and individually process each abnormal vector building element involved in the conflict, thereby improving the ability to adapt to complex scenarios and avoiding the occurrence of secondary conflicts. This greatly preserves the authenticity of the map while improving the success rate of conflict resolution. In addition, in terms of algorithm timeliness optimization, this embodiment uses the DQN model to uniformly group all conflict elements in the target space area for conflict resolution, avoiding the huge time loss caused by the traditional method of manually resolving each conflict one by one, greatly improving the timeliness of conflict resolution.

[0155] In summary, the method for handling overlapping conflicts of vector building elements based on deep reinforcement learning provided by the embodiments of the present invention has at least the following advantages:

[0156] First, it improves the accuracy of detecting spatial overlapping conflicts in a large area. By constructing constrained Delaunary triangulation and skeleton line segments, the relationship strength between adjacent surface elements is strengthened. Compared with traditional methods, the accuracy of detecting overlapping conflicts is improved.

[0157] Secondly, it improves the timeliness of resolving spatial overlapping conflicts. With the help of the DQN model's autonomous interaction with the environment and the setting of constraint conditions, it enables abnormal vector building elements involved in overlapping conflicts to autonomously find the optimal position to resolve the conflict problem, greatly improving the work efficiency of the overlapping conflict processing link.

[0158] Third, the DQN model has wide applicability and diverse output results. Based on the distribution characteristics of planar buildings in different research areas, differentiated constraint rules are formulated, which can scientifically determine the reasonable layout of conflicting elements, enable the DQN model to fully adapt to the characteristics of each region, and enable the model to output multiple reasonable layout plans within the constraint framework, providing sample data for reference of the final results.

[0159] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the various embodiments of the present invention.

Claims

1. A method for handling overlapping conflicts of vector building elements based on deep reinforcement learning, characterized in that: include: Determine and construct the constraint Delaunary triangulation and skeleton segments between each vector building element in the target spatial area to generate a spatial conflict detection field model; Connect each vector building element with the center points of other surrounding vector building elements to obtain connecting line segments, and combine the spatial conflict detection field model to determine abnormal vector building elements with overlapping conflicts; Taking abnormal vector building elements as intelligent agents, the current state information of the abnormal vector building elements is determined, and data is grouped according to the area size of the abnormal vector building elements to set constraint parameters and action reward information for each group. Combined with the pre-built DQN model, the abnormal vector building elements are enabled to autonomously interact with the environment to obtain action prediction information; According to the motion prediction information, controlling the abnormal vector building element to perform a position movement action until it reaches an optimal position; Set the constraint parameters for each group, including: Determine the map boundary constraints based on the map boundary corresponding to the target spatial area; If the distribution pattern of the surrounding building complex is a regular pattern, the minimum distance between the abnormal vector building element and the adjacent building elements in the surrounding building complex is determined, and a first moving buffer with a side length of three times the minimum distance and accommodating twice the size of the abnormal vector building element itself is constructed, thereby obtaining a moving buffer constraint condition under a regular distribution; if the distribution pattern of the surrounding building complex is an irregular pattern, the maximum distance between the abnormal vector building element and the adjacent building elements in the surrounding building complex is determined, and a second moving buffer with a side length of four times the maximum distance and accommodating twice the size of the abnormal vector building element itself is constructed, thereby obtaining a moving buffer constraint condition under an irregular distribution; According to the provisions of the mapping specifications, determine the standard distance range between abnormal vector building elements and other buildings, and determine the movement distance constraint conditions; According to the state of overlap conflict between abnormal vector building elements and other vector building elements during movement, the secondary conflict constraint condition is determined; The map boundary constraint, the moving buffer constraint, the moving distance constraint, and the secondary conflict constraint are used as constraint parameters for each group; Setting the action reward information of each group includes: summing the boundary constraint reward value, the buffer constraint reward value, the distance constraint reward value, and the secondary constraint reward value to obtain the action reward information of each group.

2. The method for handling overlapping conflicts of vector building elements based on deep reinforcement learning according to claim 1, characterized in that: Determine the constraints between each vector building element in the target spatial area, including: All edges existing in the target spatial area are encrypted, and a plane Delaunary triangulation is constructed between each vector building element to obtain a constrained Delaunary triangulation.

3. The method for handling overlapping conflicts of vector building elements based on deep reinforcement learning according to claim 1, wherein: Determine the skeleton segments between each vector building feature in the target spatial area, including: Determine the number of adjacent triangles of each triangle in the constrained delaunary triangulation network, and divide all triangles in the constrained delaunary triangulation network into three types based on the number of adjacent triangles; All triangles are traversed, and according to the type corresponding to each triangle, a segmentation curve of the blank area between each vector building element is determined, and the segmentation curve is used as a skeleton line segment in the target space area.

4. The method for handling overlapping conflicts of vector building elements based on deep reinforcement learning according to claim 1, wherein: Identify anomalous vector building features with overlapping conflicts, including: Determine whether the connecting line segment intersects with a skeleton line segment in the spatial conflict detection field model, or completely intersects with any triangle in a constrained Delaunary triangulation in the spatial conflict detection field model, to obtain a first determination result; According to the first judgment result, abnormal vector building elements with overlapping conflicts are determined.

5. The method for handling overlapping conflicts of vector building elements based on deep reinforcement learning according to claim 4 is characterized in that: Determining abnormal vector building elements with overlapping conflicts based on the first judgment result includes: If the first judgment result is not completely yes, it is determined that there is an overlapping conflict between the vector building elements, and an abnormal vector building element with an overlapping conflict is obtained.

6. The method for handling overlapping conflicts of vector building elements based on deep reinforcement learning according to claim 1, characterized in that: The DQN model includes: An online network is used to predict the expected reward value corresponding to each alternative action using the original fixed parameters, and output action prediction information based on the expected reward value; The target network is used to generate a target reward value using current fixed parameters, wherein the parameters of the target network are copied from the online network according to a set period.

7. The method for handling overlapping conflicts of vector building elements based on deep reinforcement learning according to claim 1, characterized in that: Set the action reward information for each group, including: Determining whether the abnormal vector building element satisfies the map boundary constraint condition during movement, obtaining a second determination result, and calculating a boundary constraint reward value in a current state according to the second determination result; Determining whether the abnormal vector building element satisfies the movement buffer constraint condition during movement, obtaining a third determination result, and calculating a buffer constraint reward value in a current state according to the third determination result; Determining whether the abnormal vector building element satisfies the movement distance constraint condition during the movement process, obtaining a fourth determination result, and calculating a distance constraint reward value in the current state according to the fourth determination result; It is determined whether the abnormal vector building element satisfies the secondary conflict constraint condition during the movement process to obtain a fifth determination result, and the secondary constraint reward value in the current state is calculated according to the fifth determination result.

8. The method for handling overlapping conflicts of vector building elements based on deep reinforcement learning according to claim 1, characterized in that: According to the motion prediction information, controlling the abnormal vector building element to perform a position movement action until it reaches an optimal position includes: Determining a target action of the abnormal vector building element based on the action prediction information; wherein the target action is one of a plurality of alternative movement plans uniformly divided in advance; According to the target action, the abnormal vector building element is controlled to move at equal distances until it reaches the optimal position.

Images