Shared bicycle brand identification and density optimization method based on NSGA-III algorithm driving

Abstract

The application provides a shared bicycle brand identification and density optimization method based on an NSGA-III algorithm, which comprises the following steps: acquiring a region dynamic imaging stream containing spatial position, visual representation and environmental context information in real time; analyzing data by using a heterogeneous information perception unit, identifying a bicycle brand and generating a fusion feature vector; constructing and dynamically updating a region correlation topological atlas capable of representing bicycle density, brand distribution and service demand potential by fusing multi-modal information, and inputting the atlas into an NSGA-III algorithm with a multi-dimensional strategy collaborative deployment structure and a self-adaptive solution space exploration mechanism, and optimizing and solving by dynamically adjusting a crossover and mutation operator; and finally selecting an optimal solution meeting the multi-dimensional strategy preference from a Pareto optimal solution set, and generating a shared bicycle deployment instruction set. The application realizes accurate perception and dynamic optimization of the distribution state of multi-brand shared bicycles, and effectively improves the intelligent level and operation efficiency of vehicle scheduling.

Description

Technical Field

[0001] This invention relates to the field of intelligent scheduling technology, specifically to a shared bicycle brand recognition and density optimization method driven by the NSGA-III algorithm. Background Technology

[0002] As a vital component of urban transportation, shared bicycle systems play a significant role in improving short-distance travel efficiency, alleviating traffic congestion, and promoting green and low-carbon development. With the entry of multiple operating brands and the continuous growth in user demand, achieving rational allocation and efficient scheduling of bicycle resources has become crucial to the system's service quality and operational efficiency. Currently, this field generally employs data-driven scheduling methods, relying on various sensors and location data to monitor and analyze bicycle distribution. However, most scheduling strategies fail to fully consider the differentiated distribution needs in a multi-brand environment, lacking system support for brand identification and regional density optimization, resulting in difficulties in achieving effective coordination and resource integration among brands in actual operation. Summary of the Invention

[0003] To achieve the above objectives, this invention provides the following technical solution: a shared bicycle brand recognition and density optimization method based on the NSGA-III algorithm, comprising: S1. By deploying a multi-source data acquisition system within the target service area, real-time regional dynamic imaging streams containing spatial location information, visual representation data, and environmental context information are acquired. S2. Utilize the heterogeneous information sensing unit to perform deep analysis on the dynamic imaging stream of the region, identify and extract the geometric and semantic features of the shared bicycles, generate a fused feature vector, and extract unique brand identification information; S3. Integrate the spatial location information, visual representation data, brand identification information, and environmental context information to generate a multimodal perception dataset; based on the multimodal perception dataset, construct and update a regional association topology map, which represents the vehicle density, brand distribution concentration, and regional service demand potential in different geographical areas, and has the ability to update in response to environmental changes. S4. The regional association topology map is used as input and loaded into the NSGA-III multi-dimensional strategy collaborative allocation structure. The multi-dimensional strategy collaborative allocation structure adjusts the crossover and mutation operators of the NSGA-III algorithm in real time through an adaptive solution space exploration mechanism. S5. From the Pareto optimal solution set obtained by the NSGA-III algorithm iteration, select one or more optimal solutions that satisfy the preset strategy preference; the preset strategy preference is determined based on multi-dimensional constraints such as regional service priority, brand market strategy and operating cost; and generate a shared bicycle dispatch instruction set according to the selected optimal solution.

[0004] As a further technical solution, the heterogeneous information sensing unit processes the regional dynamic imaging stream through a multi-scale feature fusion strategy during image parsing and feature extraction. This multi-scale feature fusion strategy includes: extracting basic geometric features from the regional dynamic imaging stream, including the outline shape, spatial arrangement structure, and cluster density features of shared bicycles, to identify the physical outline and cluster distribution of the shared bicycles; identifying the precise location and boundary of each shared bicycle in the image by performing semantic segmentation and target detection processing on the regional dynamic imaging stream, and resolving its brand identifier based on the appearance features in the visual representation data; and performing feature-level fusion of the extracted geometric features with the semantic features of the shared bicycles. Geometric features primarily describe the spatial distribution characteristics of bicycles, while semantic features primarily describe the brand identity characteristics of bicycles. A fusion feature vector is generated using a weighted coefficient-based fusion method. The weights of geometric features are dynamically adjusted based on their contour clarity and morphological completeness, with higher weights assigned to geometric features that have clearer contours and more complete morphologies. The weights of semantic features are dynamically adjusted based on brand identification confidence and feature discriminative power, with higher weights assigned to semantic features that have higher identification confidence and stronger feature discriminative power. Through this dynamic weight adjustment mechanism, the final fusion feature vector can simultaneously retain the spatial distribution information and brand identity information of bicycles, and features with high importance dominate the fusion process.

[0005] As a further technical solution, the regional association topology map is constructed in the following way: the target service area is divided into multiple interconnected geographical sub-regions, each with a unique identifier and geographical coordinate range; the number of various types of shared bicycles in each geographical sub-region is calculated to form an initial density distribution with brand as the classification dimension; the bicycle flow trend between adjacent geographical sub-regions is analyzed, and the flow trend weight is calculated by statistically analyzing the frequency and number of bicycles moving from one sub-region to an adjacent sub-region in historical data. The higher the flow frequency and the larger the number of moves, the higher the flow trend weight between regions; at the same time, the brand cross-association strength is calculated. By analyzing the probability and quantity relationship of different brands of bicycles appearing simultaneously in adjacent regions, the degree of association between brands is determined. The higher the probability of simultaneous appearance and the more balanced the quantity ratio, the stronger the association between brands; finally, a dynamic link relationship between regions is established, where the link weight is jointly determined by the bicycle flow trend weight and the brand cross-association strength. The flow trend weight reflects the activity level of bicycle flow between regions, and the brand cross-association strength reflects the complementary needs of brand distribution between regions. The two are combined by weighting to form a comprehensive weight, reflecting the potential willingness of bicycles to move from high-density areas to low-density areas and the demand for brand mixing.

[0006] As a further technical solution, the dynamic construction and updating process of the regional association topology map further includes an environmental disturbance perception mechanism. This mechanism is configured to: monitor changes in external environmental factors within the target service area in real time, including weather conditions, traffic conditions, and changes in population activity; dynamically adjust the weights of bicycle flow trends and brand cross-association between the geographical sub-regions based on the importance and scope of these changes. Specifically, when environmental factors unfavorable to bicycle use are detected, such as heavy rainfall, severe traffic congestion, or sparse population activity, the outflow weight of the affected area is reduced accordingly; when environmental factors promoting bicycle use are detected, such as good weather, smooth traffic, or large-scale events, the inflow weight of the relevant area is increased. The weight adjustment magnitude changes linearly with the intensity of the environmental factor's influence; the stronger the influence, the greater the adjustment magnitude. For the region... In the domain-related topology map, the nodes, i.e., each geographical sub-region, not only record bicycle density and brand distribution concentration, but also add real-time regional service saturation. This indicator is calculated by comparing the current number of bicycles with the demand based on historical data and real-time demand forecasts. When the current number is lower than the forecast demand, it indicates high service saturation, requiring the addition of bicycles; conversely, it indicates low service saturation, requiring the redeployment of bicycles. The dynamic update also includes correlation analysis of shared bicycle usage data within the target area. By analyzing the bicycle usage frequency and idle time in different areas at different times, it identifies high-frequency usage hotspots and idle areas. Areas with usage frequency above a threshold and idle time below a threshold are identified as hotspot areas, while areas with usage frequency below a threshold and idle time above a threshold are identified as idle areas. These usage pattern characteristics are integrated as additional attributes into the regional-related topology map.

[0007] As a further technical solution, the update frequency of the regional association topology map is adaptively adjusted based on the sensing intensity of the environmental disturbance sensing mechanism. The adaptive adjustment includes: when a sudden, high-intensity change in external environmental factors is detected, such as a rainstorm, traffic accident, or large-scale event, a mandatory full update of the regional association topology map is immediately triggered, that is, the attributes and edge weights of all nodes are recalculated to quickly respond to extreme changes; when a periodic, low-intensity change in external environmental factors is detected, such as normal traffic congestion during morning and evening rush hours, light rain, or daily commuting demand, only the affected geographical sub-regions and their associated links are locally updated, that is, only the data of relevant nodes and edges are updated, thereby improving efficiency and reducing computational resource consumption; the update priority is determined based on the assessment results of the degree of environmental change and the scope of impact. The degree of change is measured by the magnitude of the deviation of environmental factors from normal values, and the scope of impact is assessed by the number and importance of affected geographical sub-regions. Environmental factors with greater degree of change and wider scope of impact correspond to higher update priorities.

[0008] As a further technical solution, the multi-dimensional strategy collaborative allocation structure, after loading the regional association topology map, further includes: in the initial population generation stage of the NSGA-III algorithm, a preference-guided initialization strategy is used. This strategy generates initial individuals representing a certain proportion of the total population based on the regions with extremely uneven density and regions with severe brand concentration identified in the regional association topology map. These individuals encode preliminary scheduling schemes for these regional issues. Specifically, for regions with excessively high density, a scheme to reduce the number of vehicles in that region is generated. The reduction amount should be proportional to the excess density of the vehicles; that is, the greater the excess density, the greater the reduction. The larger the quantity, the better. For regions with high brand concentration, solutions with other brands of bicycles are generated, and the number of such solutions is determined based on the degree of brand absence and regional needs. The remaining initial solutions are generated randomly to ensure the breadth of the search space. The preference-guided initialization strategy uses a sampling method based on regional priority scores. The regional priority scores are calculated based on the degree of regional density deviation, the degree of abnormal brand concentration, and the urgency of service needs. The higher the density deviation, the more abnormal the brand concentration, and the more urgent the service needs, the higher the regional priority score. Initial solutions are generated near geographical sub-regions with high priority scores, thereby accelerating the convergence of the NSGA-III algorithm to a meaningful solution space.

[0009] As a further technical solution, the adaptive solution space exploration mechanism further includes: adjusting the granularity of crossover operations and the intensity of mutation operations in real time based on the convergence trend and population diversity level of the NSGA-III algorithm during the evolution process; when the population convergence speed is detected to be too fast or the diversity is reduced, increasing the crossover probability and mutation probability to improve population diversity, the increase in crossover probability and mutation probability is determined according to the degree of diversity decline, the more severe the decline, the greater the increase; when the population is detected to be oscillating near the Pareto front or converging slowly, reducing the crossover probability and mutation probability to promote refined search, the decrease in crossover probability and mutation probability is determined according to the convergence stagnation time, the longer the stagnation time, the greater the decrease; the adjustment strategy is based on the population evolution state evaluation results, which are achieved by analyzing the quality changes and distribution of the solution set in each generation of the population, the quality change is measured by the degree of improvement of the objective function of the solution set, and the distribution is measured by the uniformity of the coverage of the solution set in the target space.

[0010] As a further technical solution, the adaptive solution space exploration mechanism, which optimizes the non-dominated ranking and congestion distance calculation method for the newly generated Pareto front, further includes: an integrated multi-objective dynamic weighting method. This method is configured to assign dynamic weights to the shared bicycle regional bicycle density equalization objective, the brand diversity distribution objective, and the resource idle rate minimization objective before each non-dominated ranking, based on the real-time operational needs of the current target service area. The weight allocation principle is based on operational scenario characteristics: during the morning peak hours, priority is given to ensuring density equalization, assigning a higher weight to the density equalization objective; during the evening peak hours, priority is given to reducing the idle rate, assigning a higher weight to the resource idle rate minimization objective. Higher weights are assigned to brand diversity during off-peak hours, giving it higher weight. Based on these dynamic weights, the relative positions of individuals in the multi-objective space are reassessed, and in the non-dominated ranking process, individuals that show better overall performance after weight adjustment are given priority. When calculating the crowding distance, the proximity measure based on the regional association topology map features is used, so that the crowding distance not only reflects the distribution in the target space, but also reflects the uniformity of the distribution of individuals in the geographic space and brand dimension. Specifically, this is achieved by evaluating the distribution balance and brand portfolio diversity among the geographic sub-regions in the solution. The more balanced the distribution and the more diverse the brand portfolio, the greater the crowding distance of the solution.

[0011] As a further technical solution, the generation and execution process of the shared bicycle dispatch instruction set further includes: converting the optimal solution into operable dispatch instructions, wherein the dispatch instructions include specifying the number of shared bicycles to be dispatched and brand identification information, wherein the number and type are obtained by matching the dispatch volume in the optimal solution with the target state in the regional association topology map; planning the optimal path for shared bicycles from their current parking location to the target area, taking into account real-time traffic conditions and transportation efficiency, generating a dispatch task list, and distributing it to the on-site staff of the shared bicycle operation team to ensure that the dispatch plan can be accurately implemented; the task list contains detailed dispatch instructions, including key information such as the bicycle identification to be dispatched, target location information, and expected completion time, and the priority of the dispatch instructions is determined according to the urgency of regional service and the urgency of brand demand, with higher priority instructions for more urgent services and more urgent demands.

[0012] This invention provides a shared bicycle brand recognition and density optimization method based on the NSGA-III algorithm, which has the following beneficial effects: 1. This invention integrates multi-source heterogeneous sensing data and constructs a dynamically updated regional association topology map, enabling it to accurately identify shared bicycle brand logos and more deeply perceive the dynamic associations and environmental disturbances between regions, thereby significantly improving the real-world accuracy of the system's perception and the comprehensiveness of its decision support.

[0013] 2. This invention employs the NSGA-III algorithm enhanced by a multi-dimensional strategy collaborative allocation structure. Through an adaptive solution space exploration mechanism and a preference-guided initialization strategy, it effectively solves the balance problem of convergence speed, population diversity, and strategy preference integration in multi-objective optimization. This enables the algorithm to quickly lock and continuously optimize the Pareto solution set that meets complex operational constraints. The intelligence and targeting of its optimization process far exceed that of traditional static parameter optimization algorithms. The final output scheduling scheme has better comprehensive performance in multiple objectives such as density balance, brand ratio, and cost control. Attached Figure Description

[0014] Figure 1 This is a schematic diagram of the process of the present invention. Detailed Implementation

[0015] The technical solutions in the embodiments of the present invention will be clearly and completely described below. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0016] refer to Figure 1 In this embodiment, a shared bicycle brand recognition and density optimization method based on the NSGA-III algorithm is provided. This method achieves accurate identification and dynamic optimization of the distribution status of shared bicycles by combining multi-source information perception, brand recognition, regional topology modeling, environmental disturbance perception, adaptive evolutionary optimization, and task instruction generation through a complete process from data collection to scheduling execution. This embodiment will take the operation of shared bicycles in the core urban area as the application scenario and elaborate on the specific implementation steps in detail to ensure the completeness and feasibility of the technical solution.

[0017] First, during the data acquisition phase, a multi-source data acquisition system is deployed within the target service area to acquire real-time dynamic imaging streams that characterize the distribution of shared bicycles. These dynamic imaging streams contain three types of information: spatial location information, visual representation data, and environmental context information. Spatial location information comes from the bicycle's built-in positioning device and area monitoring equipment, achieving meter-level accuracy. Visual representation data is collected by cameras deployed at road intersections, public entrances and exits, subway entrances, and high-traffic areas. These cameras collect continuous images at fixed angles and timed frequencies. Environmental context information is provided by an external sensor network, covering weather conditions, road traffic conditions, and crowd flow. Taking a one-square-kilometer urban area as an example, deploying ten fixed acquisition points can cover major traffic arteries and residential areas, thus forming a continuously updated regional dynamic imaging stream input.

[0018] During the data analysis phase, a heterogeneous information sensing unit is used to process the dynamic imaging stream of the aforementioned region. The heterogeneous information sensing unit then employs a multi-scale feature fusion strategy to identify and extract features from the bicycles. Specifically, it first analyzes basic geometric features, including the bicycle's outline boundary, size ratio, and arrangement, to characterize the bicycle's appearance and clustering. Secondly, it performs semantic segmentation and target detection on the image to obtain the individual boundaries and precise locations of the bicycles, and extracts feature regions on the bicycle body, such as color combinations, stripe structures, and text markings, thereby acquiring brand identity features. During the fusion process, geometric and semantic features are weighted and integrated to generate a fused feature vector. The weights are determined as follows: when the image clarity is high and the outline integrity exceeds a set threshold, the geometric feature weight is increased to 60%; when the image is blurry or the lighting is insufficient, the semantic feature weight is increased to over 70%, thus ensuring the stability of brand recognition under different acquisition conditions. If the confidence level of a brand recognition exceeds 85%, it can be directly confirmed as a valid recognition result; if the confidence level is between 60% and 85%, it is compared with historical recognition results to reduce misjudgments.

[0019] Subsequently, based on spatial location information, visual representation data, brand identification information, and environmental context information, a regional association topology map is constructed and dynamically updated. The specific method is as follows: First, the target area is divided into several geographical sub-regions, each with a unique number and clear geographical boundaries. For each sub-region, the number of bicycles of each brand is counted to form an initial density distribution. For example, in a sub-region with two hundred bicycles, brand A accounts for one hundred and twenty, brand B for fifty, and brand C for thirty. Therefore, the brand distribution concentration in this region is relatively high, with brand A accounting for sixty percent. Second, adjacent sub-regions are analyzed. The flow trend of bicycles between sub-regions is determined by the number and frequency of migrations in historical data. If there are an average of 30 bicycles flowing between sub-region 1 and the adjacent sub-region 2 every day, and this value accounts for 20% of the total number of bicycles in sub-region 1, then the flow trend weight is 0.2. At the same time, the cross-correlation degree of different brands in adjacent sub-regions is calculated. For example, if the probability of brand A and brand B appearing in two adjacent regions at the same time exceeds 70%, then the correlation between the two is relatively high. Finally, the dynamic link relationship between regions is determined by the flow trend weight and the brand cross-correlation degree, forming a comprehensive weight that can reflect the real flow potential and brand complementarity.

[0020] Building upon this foundation, the construction of the regional topology map also includes an environmental disturbance perception mechanism. This mechanism monitors changes in the external environment of the target area in real time. When weather factors such as heavy rainfall or high temperatures are detected, the outflow weight of the relevant area is reduced while the inflow weight of adjacent rain shelter or heat shelter areas is increased. For example, when rainfall exceeds 20 millimeters per hour, the outflow weight of the affected area is reduced by 20%, while the inflow weight of nearby subway entrances and indoor parking lots is increased by 15%. When increased population activity is detected, such as hosting large-scale exhibitions or concerts, the service demand potential of the area increases by more than 50%, and its service saturation is recalculated. The service saturation index is measured by the ratio of the current number of bicycles to the predicted demand. When an area has only 100 bicycles while the predicted demand reaches 180, the service saturation is only 0.56, and the area is judged to be undersupplied. The dynamic update process also includes the analysis of usage data. For example, when the bicycle usage frequency in an area exceeds 80 times per hour during peak hours, while the usage frequency in neighboring areas is less than 20 times per hour, the area is marked as a hotspot area, and the other area is marked as an idle area. The corresponding characteristics are written into the regional topology map.

[0021] The update frequency of the regional topology map is adaptively adjusted according to the intensity of environmental disturbances. When a sudden and high-intensity external event is detected, such as a rainstorm or a traffic accident, a full update is immediately triggered, recalculating all node and link attributes. The full update process is completed within ten seconds. When a periodic and low-intensity change is detected, such as light rain or normal traffic congestion during morning and evening rush hours, only the affected sub-regions are locally updated, with each update cycle lasting five minutes, thereby improving computational efficiency.

[0022] During the optimization and solution phase, the constructed regional association topology map is input into the NSGA-III algorithm, which has a multi-dimensional strategy collaborative allocation structure. This structure adopts a preference-guided strategy during the initial population generation. For regions with extremely uneven density, initial individuals are generated to reduce or increase the number of bicycles, with the reduction or increase determined proportionally based on the actual excess or shortage value. For example, when a region has fifty more bicycles than the reasonable density, the initial solution will generate a scheduling plan to transfer fifty bicycles. For regions with excessively high brand concentration, such as 90% of the bicycles in a region being from the same brand, an initial solution will be generated to transfer bicycles from other brands, with the number of introductions calculated based on the degree of brand absence. Regional priority scores are determined comprehensively based on density deviation, abnormal brand concentration, and urgency of demand, with high-priority regions being given priority consideration in the initial solution. The remaining initial individuals are generated randomly to ensure the breadth and diversity of the search space.

[0023] During the evolutionary process, an adaptive solution space exploration mechanism is adopted. This mechanism monitors the population convergence status and diversity level in real time. When it detects that the convergence is too fast or the diversity is declining, it increases the probability of crossover and mutation operations to increase the ability to generate new solutions. When it detects that the population is oscillating in the Pareto front for a long time or improving slowly, it reduces the probability of crossover and mutation to conduct a refined search. The population evaluation criteria include the improvement magnitude of the objective function of the solution set in each generation and the uniformity of the solution coverage in the objective space. In the process of non-dominated sorting and crowding distance calculation, a multi-objective dimension dynamic weighting method is adopted to allocate objective weights according to real-time operational needs. For example, during the morning rush hour, the weight of the single-vehicle density equalization target is set at 50%, the weight of the brand diversity target is 30%, and the weight of the resource idle rate minimization target is 20%. During the evening rush hour, the weight of the resource idle rate minimization target is increased to 50%, while during off-peak hours, the weight of the brand diversity target is increased accordingly. This ensures that the solution output by the algorithm meets the actual needs under different operating scenarios. The congestion distance calculation also introduces the balance assessment of geographic space and brand dimensions, so that the selected solution not only meets the numerical optimization target, but also ensures the rationality of geographic distribution and brand combination.

[0024] After obtaining the Pareto optimal solution set, it is filtered according to preset strategy preferences. The preset strategies include regional service priority, brand market strategy, and operating cost constraints. For example, if the operator stipulates that core business districts and subway entrance areas are the highest priority, and requires that the proportion of a certain brand in any single area does not exceed 70%, and the total dispatch mileage does not exceed 300 kilometers per day, then the solution that meets the conditions is selected as the optimal solution from the solution set. Then, the solution is transformed into a shared bicycle dispatch instruction set. The instruction set clearly specifies the number of bicycles to be dispatched or dispatched in each sub-area and the brand type. For example, dispatching 30 brand A bicycles and dispatching 20 brand B bicycles and 10 brand C bicycles in a certain area. The instruction set further generates a task list and distributes it to the operations team. The task list includes bicycle identification, current location, target location, route planning, and expected completion time. Route planning takes into account real-time traffic conditions to ensure that dispatch tasks are completed within the specified time. Task priority is sorted according to the urgency of regional service and the urgency of brand needs, giving priority to ensuring the service level of the most critical areas.

[0025] This method not only ensures balanced density and brand ratio, but also effectively reduces idle rates and operating costs, demonstrating significant practical value and promotional significance.

[0026] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A shared bicycle brand recognition and density optimization method based on the NSGA-III algorithm, characterized in that, Includes the following steps: S1. By deploying a multi-source data acquisition system within the target service area, real-time regional dynamic imaging streams containing spatial location information, visual representation data, and environmental context information are acquired. S2. Utilize the heterogeneous information sensing unit to perform deep analysis on the dynamic imaging stream of the region, identify and extract the geometric and semantic features of the shared bicycles, generate a fused feature vector, and extract unique brand identification information; S3. Integrate the spatial location information, visual representation data, brand identification information, and environmental context information to generate a multimodal perception dataset; based on the multimodal perception dataset, construct and update a regional association topology map, which represents the vehicle density, brand distribution concentration, and regional service demand potential in different geographical areas, and has the ability to update in response to environmental changes. S4. The regional association topology map is used as input and loaded into the NSGA-III multi-dimensional strategy collaborative allocation structure. The multi-dimensional strategy collaborative allocation structure adjusts the crossover and mutation operators of the NSGA-III algorithm in real time through an adaptive solution space exploration mechanism. S5. From the Pareto optimal solution set obtained by the NSGA-III algorithm iteration, select one or more optimal solutions that satisfy the preset strategy preferences; The preset strategy preference is determined based on multiple constraints such as regional service priority, brand market strategy and operating costs; and a set of shared bicycle dispatch instructions is generated based on the selected optimal solution.

2. The shared bicycle brand recognition and density optimization method based on the NSGA-III algorithm as described in claim 1, characterized in that: In the process of image parsing and feature extraction, the heterogeneous information perception unit processes the regional dynamic imaging stream through a multi-scale feature fusion strategy. The multi-scale feature fusion strategy includes: extracting basic geometric features from the regional dynamic imaging stream to identify the physical contours and cluster distribution of shared bicycles; performing semantic segmentation and target detection on the regional dynamic imaging stream to identify each shared bicycle and its brand logo; and weightedly fusing the extracted geometric features with the semantic features of the shared bicycles to generate a fused feature vector that has a stronger descriptive ability for individual shared bicycles and their surrounding environment.

3. The shared bicycle brand recognition and density optimization method based on the NSGA-III algorithm as described in claim 1, characterized in that: The regional association topology map is constructed in the following way: the target service area is divided into multiple interconnected geographical sub-regions; the number of various types of shared bicycles in each geographical sub-region is calculated to form an initial density distribution; the bicycle flow trend and brand cross-association between adjacent geographical sub-regions are analyzed to establish dynamic links between regions, where dynamic links reflect the potential willingness of bicycles to move from high-density areas to low-density areas.

4. The shared bicycle brand recognition and density optimization method based on the NSGA-III algorithm as described in claim 3, characterized in that: The dynamic construction and updating process of the regional association topology map further includes an environmental disturbance perception mechanism. This mechanism is configured to: monitor changes in external environmental factors within the target service area in real time; and dynamically adjust the weights of bicycle flow trends and brand cross-association between the geographical sub-regions based on these changes to reflect the dynamic changes in bicycle demand and supply in the real world. For nodes in the regional association topology map, i.e., each geographical sub-region, in addition to recording bicycle density and brand distribution concentration, the real-time service saturation of the region is also added. The service saturation index comprehensively considers the current number of bicycles and the predicted demand. The dynamic update also includes correlation analysis of shared bicycle usage data within the target area to identify high-frequency usage hotspots and idle areas within a certain time period, and integrate them as additional attributes into the regional association topology map.

5. The shared bicycle brand recognition and density optimization method based on the NSGA-III algorithm as described in claim 4, characterized in that: The update frequency of the regional association topology map is adaptively adjusted based on the sensing intensity of the environmental disturbance sensing mechanism. The adaptive adjustment includes: when a sudden and high-intensity change in external environmental factors is detected, a mandatory full update of the regional association topology map is immediately triggered to quickly respond to extreme changes; when a periodic and low-intensity change in external environmental factors is detected, only the affected geographical sub-regions and their associated links are locally updated.

6. The shared bicycle brand recognition and density optimization method based on the NSGA-III algorithm as described in claim 1, characterized in that: After loading the regional association topology map, the multi-dimensional strategy collaborative allocation structure further includes: in the initial population generation stage of the NSGA-III algorithm, a preference-guided initialization strategy is used. This strategy generates a certain proportion of initial individuals based on the areas with extremely uneven density and severely concentrated brands identified in the regional association topology map. These individuals represent preliminary allocation schemes for extreme problems, ensuring the quality and diversity of the initial population. The remaining initial individuals are generated randomly to ensure the breadth of the search space. The preference-guided initialization strategy uses a sampling method based on regional priority scores to prioritize generating initial solutions near high-priority geographical sub-regions requiring urgent intervention, accelerating the convergence of the NSGA-III algorithm to a meaningful solution space.

7. The shared bicycle brand recognition and density optimization method based on the NSGA-III algorithm as described in claim 6, characterized in that: The adaptive solution space exploration mechanism further includes: adjusting the granularity of crossover operations and the intensity of mutation operations in real time based on the convergence trend and population diversity level of the NSGA-III algorithm during the evolution process; dynamically increasing the crossover and mutation probabilities when the NSGA-III algorithm detects that the population convergence speed is too fast or the diversity is reduced, so as to increase the population's ability to generate new individuals and explore new solution space regions; and appropriately reducing the crossover and mutation probabilities when the population is detected to be oscillating near the Pareto front or converging slowly, so as to promote the NSGA-III algorithm to refine its search for local optima and prevent deviation from the optimal solution caused by over-exploration.

8. The shared bicycle brand recognition and density optimization method based on the NSGA-III algorithm as described in claim 7, characterized in that: The adaptive solution space exploration mechanism further includes an optimized non-dominated ranking and congestion distance calculation method for the newly generated Pareto front, comprising: an integrated multi-objective dynamic weighting method. This method is configured such that, before each non-dominated ranking, dynamic weights are assigned to the shared bicycles' regional bicycle density equalization objective, brand diversity distribution objective, and resource idle rate minimization objective based on the real-time operational needs of the current target service area, reflecting priorities under different stages or strategies. Based on these dynamic weights, the relative positions of individuals in the multi-objective space are reassessed, and during the non-dominated ranking process, individuals exhibiting better overall performance after weight adjustment are given priority. When calculating the congestion distance, a proximity metric based on regional association topological map features is used, ensuring that the congestion distance reflects the uniformity of individual distribution across geographic space and brand dimensions.

9. The shared bicycle brand recognition and density optimization method based on the NSGA-III algorithm as described in claim 1, characterized in that: The process of generating and executing the shared bicycle dispatch instruction set further includes: converting the optimal solution into operable dispatch instructions, which include: specifying the number and brand identification information of the shared bicycles to be dispatched, wherein the number and type are precisely matched with the target density and brand ratio of each geographical sub-region; planning the optimal path of the shared bicycles from the current parking location to the target area, generating a dispatch task list, and distributing it to the on-site staff of the shared bicycle operation team.

Images