Shared transportation parking facility site selection method based on multi-source travel data fusion

Abstract

This invention proposes a site selection method for shared transportation parking facilities based on multi-source travel data fusion, belonging to the field of urban transportation planning technology. The method includes: collecting spatial data within the study area; obtaining the explicit demand density of each grid based on gridded cycling endpoint data; screening off-site parking points, using clustering algorithms to identify high-frequency parking hotspots, and calculating the implicit demand density of each grid; fusing the explicit and implicit demand densities according to weights to obtain the comprehensive demand density of each grid; selecting preliminary candidate points and setting minimum spacing; conducting feasibility screening on the preliminary candidate points to obtain a candidate point set; establishing a bi-objective optimization model, using an improved NSGA-II algorithm for iterative optimization, and outputting the Pareto optimal solution set after reaching the iteration stopping condition. This invention can achieve a balance between demand coverage and construction costs, providing a scientific decision support method for urban shared transportation facility planning.

Description

Technical Field

[0001] This invention relates to the field of urban transportation planning technology, and in particular to a method for site selection of shared transportation parking facilities based on the fusion of multi-source travel data. Background Technology

[0002] Shared bicycles, shared electric bicycles, and other shared transportation tools are an important part of the urban public transportation system, and the rational layout of their parking facilities directly affects user experience and urban management efficiency. Traditional parking facility site selection methods mainly rely on manual surveys and experience-based judgments, resulting in insufficient data utilization. On the one hand, existing methods only analyze demand based on parking data within existing facilities, making it difficult to comprehensively reflect the actual distribution of parking demand across the city, especially failing to identify potential demand reflected in high-frequency parking areas outside the coverage of existing facilities. On the other hand, in the facility site selection optimization stage, traditional optimization algorithms often ignore the spatial structural characteristics of candidate points, failing to fully consider the spatial attributes of decision variables in initialization, genetic operations, and diversity preservation stages. This leads to low algorithm search efficiency, and the optimization results may show facilities being over-concentrated in high-demand areas, resulting in a poor overall spatial layout. Furthermore, existing methods lack dynamic adjustment mechanisms, making it difficult to respond promptly to changes in demand brought about by urban development, affecting the long-term utilization efficiency of facilities.

[0003] Chinese invention patent CN111582552A discloses a method for allocating shared bicycle parking spots based on a multi-objective genetic algorithm. This patent employs an improved NSGA-II algorithm to optimize the parking spot allocation problem, replacing the traditional mutation operator with a regression algorithm and introducing a Hypervolume evaluation metric to assess the algorithm's convergence and distribution. The method constructs a bi-objective optimization model with distance and density costs as objectives, uses chromosome encoding based on parking order, and achieves population evolution through tournament selection, self-crossing, and elite retention strategies, enabling the generation of parking spot recommendations for multiple user request queues in a short time. However, this patent primarily addresses the real-time parking spot allocation problem, focusing on recommending optimal parking locations for users within an existing parking spot system, without addressing how to identify new parking facility construction needs within a city and determine candidate locations. Summary of the Invention

[0004] In view of this, the present invention provides a site selection method for shared transportation parking facilities based on multi-source travel data fusion. By integrating shared transportation platform trajectory data, existing parking facility data, and urban spatial data, a comprehensive demand assessment system integrating explicit and implicit demands is established. Furthermore, an improved NSGA-II algorithm with spatial perception is used to establish a dual-objective optimization model for site selection decision-making, achieving a balance between demand coverage and construction costs, and providing a scientific decision support method for urban shared transportation facility planning.

[0005] The technical solution of this invention is implemented as follows:

[0006] This invention provides a method for site selection of shared transportation parking facilities based on multi-source travel data fusion, including:

[0007] S1. Collect shared transportation platform trajectory data, existing parking facility data and urban spatial data within the study area. After preprocessing the data, map it onto a regular grid system to obtain a gridded spatial data foundation.

[0008] S2. Based on gridded cycling destination data, the kernel density estimation method is used to calculate the parking demand density of each grid in different time periods, so as to obtain the explicit demand density of each grid in different time periods.

[0009] S3. Calculate the distance from each cycling endpoint to the nearest existing facility, filter out cycling endpoints with distances exceeding the threshold as off-site parking points, use the DBSCAN density clustering algorithm to identify high-frequency parking hotspot clusters for off-site parking points, calculate the implicit demand intensity of each hotspot cluster and allocate it to the grid it covers, and obtain the implicit demand density of each grid.

[0010] S4. The explicit demand density and implicit demand density are fused according to weight to obtain the comprehensive demand density of each grid. Grids with comprehensive demand density higher than the set quantile are selected as high demand areas. In the high demand areas, the centers of high demand grids around densely populated interest points are selected as preliminary candidate points and a minimum spacing is set. The preliminary candidate points are screened for feasibility to obtain a set of candidate points.

[0011] S5. Define a binary decision variable to represent whether a candidate point should be used to construct a facility. Establish a dual-objective optimization model and use an improved NSGA-II algorithm for iterative optimization. The improved NSGA-II algorithm includes spatial clustering of candidate points according to their geographical location, initializing the population using a hierarchical strategy, using a crossover operation based on spatial neighborhood, and calculating the distance between the dual spatial crowding of the target space and the decision space for individual selection. After the iteration stops, output the Pareto optimal solution set.

[0012] Preferably, step S1 includes:

[0013] Collect trajectory data of shared transportation platforms within the study area for a time span of no less than a preset number of months, including the starting coordinates, ending coordinates, start time, and end time of each ride;

[0014] Collect the location coordinates, service radius, and capacity information of existing parking facilities within the study area;

[0015] Collect urban spatial data, including the location and type attributes of points of interest, road network topology, and sidewalk width, intersection location, and land ownership information of candidate areas;

[0016] The collected data is preprocessed to convert all coordinates to the same coordinate system and remove abnormal records where the riding speed exceeds the preset speed threshold or the riding time is less than the preset time threshold.

[0017] The study area is divided into a regular grid with a side length of a preset range, and all data is mapped to this grid system.

[0018] Preferably, step S2 includes:

[0019] Based on the cycling destination data from the shared transportation platform's trajectory data, the kernel density estimation method is used to calculate the parking demand density for each grid. For the coordinates of any grid center point... Its parking demand density is calculated using the following formula:

[0020] ;

[0021] Where n is the total number of cycling destinations, and h is the bandwidth parameter. For Gaussian kernel function, Let k be the Euclidean distance from the k-th cycling endpoint to the center of the grid.

[0022] Kernel density analysis was performed on the data for four time periods: weekday morning peak, weekday evening peak, weekday off-peak, and all day on weekends, to obtain the explicit demand density of each grid in each time period. , where g is the grid number and t is the time period number.

[0023] Preferably, step S3 includes:

[0024] Calculate the distance from each cycling destination to the nearest existing facility, and filter out those that are more than a threshold away from the existing facility. The cycling finish line will serve as an off-site parking area.

[0025] The DBSCAN density clustering algorithm is used to identify high-frequency parking hotspots at the selected off-site parking locations, and a neighborhood radius parameter is set. and minimum sample size parameter ;

[0026] For each identified high-frequency parking hotspot cluster Calculate its implicit demand intensity:

[0027] ;

[0028] Where m is the cluster number. For clusters The total number of parking records within the area. For clusters Spatial density, i.e., the number of times a vehicle is parked per unit area. and These are the weighting coefficients;

[0029] The implicit demand intensity of each high-frequency parking hotspot cluster is allocated to the grid it covers, thus obtaining the implicit demand density of each grid. .

[0030] Preferably, in step S4, the formula for calculating the comprehensive demand density for each grid g is as follows:

[0031] ;

[0032] in This is the weighted average of the explicit demand density of the grid in each time period, i.e. T represents the total number of time periods, and the weight of each time period is... Morning and evening peak hours are assigned the first time period weight value, while off-peak hours and weekends are assigned the second time period weight value. The implicit demand density of this grid. and For fusion weighting coefficients.

[0033] Preferably, the preliminary candidate sites are screened for feasibility, specifically including:

[0034] Initial candidate points with sidewalk widths less than a width threshold are eliminated using sidewalk width data; initial candidate points with distances less than a preset length from intersections are eliminated using intersection location data; initial candidate points with property rights disputes are eliminated using land use property rights information; and initial candidate points located in inaccessible locations such as the central median of main roads are eliminated using road network data.

[0035] After iterating through and filtering all the initial candidate points, the final set of candidate points is obtained.

[0036] Preferably, the establishment of a dual-objective optimization model in step S5 is as follows:

[0037] Define decision variables as Where p is the candidate point index, This indicates that parking facilities will be built at the candidate sites. This indicates no construction; each individual corresponds to a complete decision vector. , which represents a site selection scheme, where N is the total number of candidate sites;

[0038] Establish a dual-objective optimization model, with the first objective being to maximize demand coverage:

[0039] ;

[0040] Where T represents the total number of time periods. Let G be the weighting coefficient for time period t, and G be the set of all grids. Let g be the explicit demand density in time period t. The service capacity of site selection scheme x for grid g in time period t is determined by the sum of the capacities of all selected facility points within the service radius;

[0041] The second objective is to minimize construction costs:

[0042] ;

[0043] in, Cost of constructing facilities at candidate sites;

[0044] The constraints include: the total cost shall not exceed the budget limit B. And the demand coverage rate is not lower than the minimum requirement. Right now .

[0045] Preferably, the improved NSGA-II algorithm is used to solve the bi-objective optimization model, and the algorithm flow includes:

[0046] Step 1: Spatial cluster all candidate points according to their geographical location into K spatial clusters to generate the initial population;

[0047] Step 2: Perform non-dominated ranking on the current population and divide individuals into several non-dominated layers according to Pareto dominance relationships;

[0048] Step 3: Calculate the spatial crowding distance for each individual;

[0049] Step 4: Select a preset number of parent individuals for breeding by selecting individuals with higher non-dominant levels, and within the same level, prioritize individuals with greater crowding distance.

[0050] Step 5: Perform a crossover operation based on spatial neighborhood on the selected parent generation to generate offspring; perform a mutation operation on the offspring; merge the parent and offspring generations and recalculate the non-dominated sorting and crowding degree, and select the individuals with higher sorting to enter the next generation.

[0051] Step 6: Periodically perform local searches to identify and replace the facility points with the smallest marginal contribution and the unselected candidate points with the largest marginal contribution.

[0052] Step 7: Repeat steps 2-6 until the iteration stopping condition is met, and output the Pareto optimal solution set.

[0053] Preferably, the distance between the two spatial crowding levels of each individual is calculated as follows:

[0054] For individuals s within the same non-dominated layer, their crowding distance is:

[0055] ;

[0056] ;

[0057] ;

[0058] Where s is the individual index; For traditional target space congestion distance; Distance for decision-making space congestion; σ represents the weighting coefficients; σ is the index of the objective function. and Let be the objective function value of the individual adjacent to individual s on target o; and Let S be the maximum and minimum objective function values ​​of the current population on target o; S is the set of individuals at the same level as individual s. Indexes other individuals within the same level; The Hamming distance between two decision vectors represents the number of distinct gene loci.

[0059] Preferably, a spatial neighborhood-based crossover operation is performed on the selected parent generation to generate offspring. Specifically, the operation involves randomly selecting a facility point from the first parent generation and recording it as a candidate point. ,satisfy Determine its spatial neighborhood Distance to candidate points The set of indexes of all candidate points not exceeding a preset distance, finding the one that matches the second parent individual. The selected facility point with the closest centroid distance is recorded as the candidate point. and its spatial neighborhood Swap the two parent generations in the neighborhood. and Gene loci within the genome generate offspring individuals.

[0060] The present invention has the following advantages over the prior art:

[0061] This invention integrates shared transportation platform trajectory data, existing parking facility data, and urban spatial data. It employs kernel density estimation to identify explicit demand and utilizes the DBSCAN density clustering algorithm to mine high-frequency parking hotspots outside facilities to quantify implicit demand, establishing a comprehensive demand assessment system that fully reflects the actual distribution of parking demand within the city. Based on this, feasibility screening using urban spatial data generates a reasonable set of candidate points, and a dual-objective optimization model is established to maximize demand coverage and minimize construction costs. By introducing a spatial clustering initialization mechanism, spatial neighborhood-based cross-operations, and a dual-spatial congestion distance calculation method that integrates the target space and decision space into the improved NSGA-II algorithm, the algorithm fully considers the spatial structural characteristics of candidate points during optimization, improving search efficiency and the rationality of the spatial distribution of solutions. The Pareto optimal solution set output by this method can provide decision-makers with multiple site selection schemes under different budget constraints, balancing demand coverage and cost control, and providing effective technical support for the scientific planning of urban shared transportation facilities. Attached Figure Description

[0062] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0063] Figure 1 This is a flowchart of the method of the present invention;

[0064] Figure 2 This is a diagram illustrating the technical implementation of the present invention;

[0065] Figure 3 This is a flowchart of the algorithm of the present invention. Detailed Implementation

[0066] The technical solutions of the present invention will be clearly and completely described below with reference to the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0067] like Figure 1As shown, this invention provides a method for selecting parking facilities for shared transportation vehicles based on multi-source travel data fusion, including: S1, collecting trajectory data of shared transportation platforms, existing parking facility data, and urban spatial data within the study area; mapping the preprocessed data onto a regular grid system to obtain a gridded spatial data foundation; S2, based on the gridded cycling endpoint data, using a kernel density estimation method to calculate the parking demand density of each grid in different time periods, obtaining the explicit demand density of each grid in different time periods; S3, calculating the distance from each cycling endpoint to the nearest existing facility, selecting cycling endpoints with distances exceeding a threshold as off-site parking points, using the DBSCAN density clustering algorithm to identify high-frequency parking hotspot clusters for off-site parking points, calculating the implicit demand intensity of each hotspot cluster and allocating it to the grids it covers, obtaining the implicit demand density of each grid; S4 S1. The explicit and implicit demand densities are fused according to weights to obtain the comprehensive demand density of each grid. Grids with comprehensive demand densities higher than a set quantile are selected as high-demand areas. Within the high-demand areas, the centers of high-demand grids around densely populated interest points are selected as preliminary candidate points with a minimum spacing. Feasibility screening is performed on the preliminary candidate points to obtain a set of candidate points. S2. A binary decision variable is defined to represent whether to build facilities at the candidate points. A dual-objective optimization model is established, and an improved NSGA-II algorithm is used for iterative optimization. The improved NSGA-II algorithm includes spatial clustering of candidate points according to geographical location, initialization of the population using a hierarchical strategy, crossover operation based on spatial neighborhood, and calculation of the distance between the dual spatial congestion of the target space and the decision space for individual selection. After the iteration stops, the Pareto optimal solution set is output.

[0068] like Figure 2As shown, to ensure the interpretability and reproducibility of site selection decisions, this invention is organized around a complete process of "demand identification—candidate generation—multi-objective optimization—solution output" in its technical implementation: First, an alignable spatial analysis base is established through unified coordinates, anomaly cleaning, and gridded mapping, so that trajectory points, facility points, and urban spatial elements are associated under the same spatial reference; Second, in the demand identification stage, explicit and implicit demands are distinguished. Explicit demands are used to characterize the intensity of observable parking behavior of users under the combined effect of existing service conditions and travel habits, and can reflect the characteristics of demand changes at different times. However, its results are usually affected by factors such as existing facility coverage, platform operation strategies, and user compliance with parking regulations, and it is difficult to reveal potential gaps outside facility coverage on its own; Implicit demands are reflected by the spatial aggregation characteristics of parking points outside facilities, and can reveal the real parking tendency generated under the conditions of insufficient existing facility services, inconvenient access, or mismatched management constraints, making up for the bias caused by relying solely on the density of the destination. Traditional site selection practices often involve directly determining sites based on endpoint thermal data or manual experience, or simply recommending and allocating within existing facility systems. These methods often fail to adequately consider the potential demand reflected by high-frequency parking outside facilities, and lack a systematic incorporation of spatial feasibility constraints and spatial structural characteristics during candidate generation and optimization stages. This results in redundant candidate point sets, unstable spatial distribution of the solution set, and difficulty in forming a complete process applicable to engineering implementation. This invention constructs explicit and implicit demand densities for different time periods within the same grid system, merges them to generate candidate points, and then iteratively solves the problem using a bi-objective model and an improved NSGA-II. This creates a closed-loop connection between input data, demand characterization, candidate constraints, and optimization output, thereby obtaining a Pareto solution set that can be used for decision comparison.

[0069] In one embodiment of the present invention, step S1 includes:

[0070] Collect trajectory data of shared transportation platforms within the study area for a time span of no less than three months, including the starting coordinates, ending coordinates, start time and end time of each ride. The data is collected through the API interface or data open platform of the shared transportation platform.

[0071] The location coordinates, service radius, and capacity information of existing parking facilities within the study area were collected. The service radius refers to the effective coverage area of ​​the parking facilities, which is set to 300 to 500 meters. The capacity refers to the number of vehicles that can be parked at the same time. This information was obtained from urban management departments or shared transportation platforms.

[0072] Collect urban spatial data, including the location and type attributes of points of interest such as subway stations, bus stops, shopping malls, office buildings, hospitals, and schools, the road network topology, and the sidewalk width, intersection location, and land ownership information of candidate areas. This data comes from the geographic information system of the urban planning department or open street map databases such as OpenStreetMap.

[0073] The collected data is preprocessed to convert all coordinates to the WGS84 coordinate system or a projection coordinate system suitable for regional analysis, such as the UTM projection. Abnormal records with cycling speeds exceeding 30 kilometers per hour or cycling durations less than 1 minute are removed. These abnormal records may be caused by data transmission errors, equipment failures, or abnormal user use.

[0074] The study area is divided into regular grids with side lengths of 50 to 100 meters. In this embodiment, the preferred side length is 75 meters. The choice of grid size needs to balance computational efficiency and spatial resolution. If it is too small, the computational load will be large and it will be easily affected by data sparsity. If it is too large, it will be unable to accurately depict the distribution of demand. All data is mapped to this grid system, that is, each cycling endpoint, facility location and point of interest is assigned to its respective grid according to its coordinates, thus obtaining the gridded spatial data foundation.

[0075] To ensure mapping consistency, the grid number g is determined by numbering each grid cell in the grid system, and all grid-related indicators are stored and calculated using this number as an index.

[0076] In one embodiment of the present invention, step S2 includes:

[0077] The data is divided into four time periods: weekday morning peak hours from 7:00 to 9:00, weekday evening peak hours from 17:00 to 19:00, weekday off-peak hours including 9:00 to 17:00 and 19:00 to 7:00 the next day, and the entire day of the weekend. The determination of weekdays and weekends is based on the date information of the cycling records, and statutory holidays are treated as weekends.

[0078] For the cycling destination data within each time period, the kernel density estimation method is used to calculate the parking demand density for each grid. For any grid center point coordinates... Its parking demand density is calculated using the following formula:

[0079] ;

[0080] Where n is the total number of cycling destinations within the time period, and h is the bandwidth parameter, which ranges from 300 meters to 500 meters. In this embodiment, a value of 400 meters is preferred. The choice of bandwidth parameter affects the smoothness of density estimation. A smaller bandwidth can more accurately identify local hotspots but is easily affected by data fluctuations, while a larger bandwidth generates a smoother density distribution but may miss small areas of high demand points. Gaussian kernel function , Let k be the Euclidean distance from the k-th cycling endpoint to the center of the grid.

[0081] Obtain the explicit demand density of each grid at different time periods. Where g is the grid number, with values ​​ranging from 1 to the total number of grids. t is the time period number, with values ​​from 1 to 4 corresponding to the morning peak, evening peak, off-peak, and weekend, respectively.

[0082] The purpose of time-segmented analysis is to capture the characteristics of parking demand at different times. For example, during weekday morning rush hour, demand is high near subway stations in residential areas; during evening rush hour, demand is high near commercial and office areas; and on weekends, demand is high in parks, shopping districts, and other leisure venues. Kernel density calculation uses a grid scanning method. For each grid center point, a weighted sum of all cycling destinations within a 3-hour radius is calculated. Points outside this 3-hour radius contribute negligibly to the value of the grid.

[0083] In one embodiment of the present invention, step S3 includes:

[0084] Calculate the distance from each cycling destination to the nearest existing facility using the Euclidean distance calculation method, and filter out those distances exceeding a threshold. The end point of the ride is used as the parking point outside the facility. The threshold distance is set to 400 meters in this embodiment. This distance represents the maximum tolerable distance that users are willing to walk to the standardized parking facility. Based on a walking speed of about 5 kilometers per hour, 400 meters corresponds to about 5 minutes of walking time.

[0085] The DBSCAN density clustering algorithm is used to identify high-frequency parking hotspots at the selected off-site parking locations, and a neighborhood radius parameter is set. For 150 meters, minimum sample size parameter The value is 15, meaning that only clusters containing at least 15 parking points are considered valid hotspots. The DBSCAN algorithm can identify clusters of arbitrary shapes and automatically filter out noise points, making it more suitable for spatial point data clustering compared to algorithms such as K-means.

[0086] The specific execution flow of the DBSCAN algorithm is as follows: First, mark all parking points outside the facility as unvisited. Then, traverse each unvisited point. For the current point, search its... All points within the neighborhood, if the number of points in the neighborhood is not less than If a point is found to be a core point, a new cluster is created. All points in the neighborhood are added to this cluster, and the process is recursively expanded. The neighborhood search and expansion process is repeated for each newly added point until the cluster can no longer be expanded. Then, the next unvisited point is processed. Eventually, the number of points in the neighborhood is less than [a certain value]. Points that do not belong to any cluster are marked as noise points;

[0087] For each identified high-frequency parking hotspot cluster Calculate its implicit demand intensity:

[0088] ;

[0089] Where m is the cluster number. For clusters The total number of parking records within the area. For clusters The spatial density, i.e., the number of parking points per unit area, is calculated by dividing the number of parking records within the cluster by the area of ​​the cluster's convex hull. The convex hull is calculated using the Graham scan algorithm to determine the minimum convex polygon of all parking points within the cluster. and The weighting coefficients are 0.6 and 0.4 in this embodiment, respectively, which means that more emphasis is placed on the number of parking times while taking into account the degree of spatial concentration.

[0090] The implicit demand intensity of each high-frequency parking hotspot cluster is allocated to the grid it covers. If a cluster covers multiple grids, the allocation is based on area ratio. Specifically, the intersection area of ​​the cluster's convex hull with each grid is calculated, and the cluster's implicit demand contribution to grid g is... ,in For clusters The intersection area of ​​the convex hull and the mesh g. For clusters The total convex hull area is used to sum the contributions of all clusters covering mesh g to obtain the implicit demand density of mesh g. The implicit needs identified in this step reflect the inadequacy of the existing facility layout and reveal the real need for users to choose nearby illegal parking to avoid taking detours.

[0091] In one embodiment of the present invention, step S4 includes:

[0092] For each grid g, its explicit and implicit requirements are combined to calculate the overall demand density:

[0093] ;

[0094] in This is the weighted average of the explicit demand density of the grid in each time period. T represents the total number of time periods, with a value of 4, and the time period weight. The weighting is set at 0.35 for morning and evening peak hours and 0.15 for off-peak hours and weekends. This weighting reflects the greater importance of peak hours for facility planning, because the supply and demand imbalance is more pronounced during peak hours. The implicit demand density of this grid. and To incorporate the weighting coefficients, the values ​​in this embodiment are 0.6 and 0.4, respectively, which means that more emphasis is placed on organized parking needs, but potential needs not covered by existing facilities are also considered.

[0095] Based on the comprehensive demand density, grids with demand density higher than the 75th percentile are selected as high-demand areas. That is, all grids are sorted from smallest to largest according to comprehensive demand density, and the bottom 25% of grids are selected. The 75th percentile is an empirical value and can be adjusted according to the actual planning and construction capabilities in different cities and study areas. Too high a value will miss some demands, while too low a value will lead to too many candidate points and increase the computational burden.

[0096] In high-demand areas, high-demand grid centers within 500 meters of densely populated interest points such as subway station exits, bus stops, shopping mall entrances, office building entrances, hospital outpatient buildings, and main school entrances are selected as initial candidate points. These locations have natural characteristics of concentrated parking demand, and users have a high awareness of the facility's location. The minimum distance between candidate points is set to 200 meters to ensure coverage while avoiding excessive density. In practice, candidate points are first sorted in descending order according to comprehensive demand density, and candidate points are selected in turn. If the distance between a candidate point and any point in the selected candidate point set is not less than 200 meters, it is added to the candidate point set; otherwise, the point is skipped.

[0097] The initial candidate sites underwent feasibility screening, specifically including: eliminating sites with sidewalk widths less than 3 meters using sidewalk width data, as parking facilities require approximately 1.5 to 2 meters of width and cannot obstruct normal pedestrian passage; pedestrian passage must be reserved at least 1 meter; eliminating sites less than 20 meters from intersections using intersection location data to avoid impacting traffic safety and efficiency; eliminating sites with ownership disputes using land use information to mitigate construction risks and subsequent legal disputes; and eliminating inaccessible or unsuitable locations such as central medians of main roads, under overpasses, and beside rivers using road network data, resulting in the final set of candidate sites. , where N is the total number of candidate points and p is the index of the candidate point.

[0098] In one embodiment of the present invention, the establishment of the dual-objective optimization model in step S5 specifically includes:

[0099] Define decision variables as , This indicates that parking facilities will be constructed at candidate point p. This indicates no construction; each individual corresponds to a complete decision vector. , which represents a site selection scheme, where N is the total number of candidate sites;

[0100] Establish a dual-objective optimization model, with the first objective being to maximize demand coverage:

[0101] ;

[0102] Where T represents the total number of time periods, with a value of 4. The weighting coefficient for time period t is 0.4 for morning and evening peak hours and 0.1 for off-peak hours and weekends. Compared to the time period weighting in step S4, this further emphasizes the importance of peak hours. G is the set of all grids. Let g be the explicit demand density in time period t. The service capacity of location scheme x for grid g in time period t is calculated as follows:

[0103] First, determine which selected facility points can cover grid g, i.e., facility points p satisfy the following conditions: Furthermore, the distance from the facility point to the center of grid g does not exceed the service radius r. Then, the capacities of all facility points covering grid g are summed to obtain:

[0104] ;

[0105] in Let p be the Euclidean distance from facility point p to the center of grid g. The design capacity of facility point p is used here. This is because the actual demand that is met will not exceed the actual demand of the grid. The objective function is essentially to calculate the proportion of demand met in each time period to the total demand and then take a weighted average.

[0106] The second objective is to minimize construction costs:

[0107] ;

[0108] in, The cost of constructing facilities at the candidate sites; costs include site rental fees, facility construction costs, and maintenance costs, etc. The cost may vary depending on the land price and construction difficulty, with higher costs in core business districts and lower costs in urban fringe areas;

[0109] The constraints include total cost not exceeding the budget limit B. And the demand coverage rate is not lower than the minimum requirement. Right now These two constraints ensure that the site selection scheme meets the basic requirements of economy and service level, B and The budget is set by the decision-maker based on the actual budget and service standards. In this example, it is assumed that the budget limit is 5 million yuan and the minimum coverage requirement is 70%.

[0110] like Figure 3 As shown, the improved NSGA-II algorithm is used to solve the bi-objective optimization model. The algorithm flow includes:

[0111] First, a population initialization based on spatial clustering is performed. All candidate points are divided into K spatial clusters according to their geographical location using K-means clustering, where K is 1.5 to 2 times the expected number of facilities to be built. For example, if 30 facilities are expected to be built, K is 45 to 60. In this embodiment, K is 50. K-means clustering uses Euclidean distance as a similarity measure. Iterative optimization is performed until the cluster center positions are stable. Then, an initial population of size 100 is generated. 40% of the individuals (40 individuals) are selected using a strategy of randomly selecting one point from each cluster to ensure spatial dispersion. Specifically, one candidate point is randomly selected from each cluster, and its corresponding decision variable is set to 1. A total of K points are selected, but some clusters are allowed not to be selected. 30% of the individuals (30 individuals) are selected using a strategy of prioritizing selection from high-demand clusters to ensure coverage. High-demand clusters refer to clusters with higher average comprehensive demand density of the covered grid. All clusters are arranged in descending order of average demand density, starting from the top 50%. Points are randomly selected from high-demand clusters, with each high-demand cluster having a higher probability of being selected. 20% of the individuals (20 individuals) adopt a strategy of prioritizing selection from low-cost clusters to ensure economic efficiency. Low-cost clusters refer to clusters with lower average construction costs among candidate points. All clusters are sorted in ascending order of average cost, and points are randomly selected from the top 50% of low-cost clusters. 10% of the individuals (10 individuals) adopt a completely random strategy to maintain diversity, randomly selecting the expected number of points from all candidate points. For each initial individual, the number of selected facility points fluctuates randomly between 80% and 120% of the expected construction number. For example, if 30 facilities are expected to be built, each individual selects 24 to 36 points. Individuals that do not meet the constraints are repaired. If the total cost exceeds the budget, the highest-cost point is randomly removed from the selected points until the budget constraint is met. If the coverage is lower than the minimum requirement, the point with the largest marginal contribution is randomly added from the unselected points until the coverage constraint is met.

[0112] Perform a non-dominated ranking of the current population. Based on Pareto dominance, individuals are divided into several non-dominated strata. If individual A is not inferior to individual B in all objectives and is superior to individual B in at least one objective, then A is said to dominate B. For the maximization objective... A is not inferior to B. For minimizing the objective A is not inferior to B. The specific process of non-dominated sorting is as follows: First, identify the set of individuals in the population that are not dominated by any other individual as the first non-dominated layer, i.e., the Pareto front. Then, remove the individuals from the first non-dominated layer from the population. Repeat the above process for the remaining individuals to identify the second non-dominated layer, and so on until all individuals are assigned to a certain non-dominated layer. The lower the non-dominated layer, the better the individual.

[0113] Calculate the two-space crowding distance for each individual. For an individual s within the same non-dominated layer, the crowding distance is:

[0114] ;

[0115] ;

[0116] ;

[0117] Where s is the individual index; For traditional target space congestion distance; Distance for decision-making space congestion; represents the weighting coefficient; o is the index of the objective function, taking a value of 1 or 2; and Let be the objective function value of the individual adjacent to individual s on target o. Adjacent means the individuals that are adjacent to each other after sorting all individuals in the same layer according to target o from smallest to largest. and Let S be the maximum and minimum objective function values ​​of the current population on target o. The crowding distance of boundary individuals, i.e., the individuals at the beginning and end after sorting, is set to infinity to ensure they are preserved. The crowding of the target space reflects the distribution density of individuals on the Pareto front. Individuals with high crowding are surrounded by more open space, which is conducive to maintaining the diversity of solutions. S is the set of individuals at the same level as individual s. Indexes other individuals within the same level; The Hamming distance between two decision vectors, i.e., the number of distinct gene loci, is calculated using the following formula:

[0118] ;

[0119] in Let represent the value of the p-th gene locus in the decision vector of individual s. The crowding of the decision space reflects the differences among individuals in their site selection schemes, encouraging the selection of schemes with different location combinations. The weighting coefficient is 0.6 in this embodiment, which means that more importance is attached to the diversity of the target space, but the diversity of the decision space is also taken into account.

[0120] Parental individuals are selected for breeding through a tournament selection process. Five individuals are randomly selected from the population each time for the tournament, and their quality is compared. Individuals with higher non-dominant hierarchies (i.e., smaller hierarchical numbers) are prioritized. Within the same hierarchical level, individuals with higher crowding distance are given priority. Larger individuals are added to the parent set as winners, and the tournament selection process is repeated until 100 parent individuals are selected. Parent individuals are allowed to be repeated, meaning that excellent individuals may be selected multiple times.

[0121] The selected parent generation is used to perform a spatial neighborhood-based crossover operation to generate offspring. Specifically, two different parent individuals are randomly selected from the parent set as crossover objects, and a selected facility point is randomly chosen from the first parent individual and recorded as a candidate point. That is, satisfy If no facility point is selected in the first parent generation, skip the intersection and determine the candidate points. spatial neighborhood Distance to candidate points The set of indexes of all candidate points not exceeding 1 kilometer Find the individual from the second parent generation that is related to The selected facility point with the closest centroid distance is recorded as the candidate point. and its spatial neighborhood The centroid refers to the average coordinates of all candidate points in its neighborhood.

[0122] ;

[0123] If the second parent is in the neighborhood If no facility point in the vicinity is selected, then any selected facility point from the second parent generation will be randomly selected as the facility point. Swap the two parent generations in the neighborhood. and The gene locus within the parent produces two offspring individuals. Specifically, offspring 1 inherits from parent 1 in the neighborhood. External genes and parent 2 in the neighborhood Within the gene, offspring 2 inherits the genes from parent 2 in the neighborhood. External genes and parent 1 in the neighborhood Within the gene pool, the crossover probability is set to 0.9, meaning that 90% of parent pairs undergo crossover, while 10% of parent pairs directly replicate into offspring without modification. The design philosophy of this crossover operation is to exchange facility point selection schemes within spatially proximate areas, allowing offspring to inherit the spatial layout pattern of their parents while simultaneously generating local variations.

[0124] The offspring are mutated with a probability of 0.1, randomly flipping a gene locus in the decision vector (changing 0 to 1 or 1 to 0). The mutation location is uniformly and randomly selected from all gene loci. If an individual does not meet the constraints after mutation, it is repaired using the same method as in the initialization phase. The purpose of the mutation operation is to increase population diversity and prevent the algorithm from converging prematurely to a local optimum.

[0125] The parent and offspring populations are merged to form a mixed population of 200. Non-dominated sorting and crowding calculations are recalculated. Individuals are selected for the next generation based on their non-dominated hierarchy from lowest to highest. Within the same non-dominated hierarchy, individuals are prioritized based on their crowding distance from highest to lowest, with higher crowding being selected first. This selection continues until the next generation population reaches 100, ensuring a constant population size. This elite preservation strategy ensures that superior individuals are not lost during evolution.

[0126] A local search is performed every 10 generations to accelerate convergence. For each individual in the current Pareto front (i.e., the first non-dominated layer), the facility point with the smallest marginal contribution is identified. The specific method is as follows:

[0127] For each selected facility point p in this individual, i.e. Calculate the decrease in coverage after deleting this point. ,in Indicates will The decision vector after changing 1 to 0, select The facility point with the smallest marginal contribution is taken as the point with the smallest contribution.

[0128] Simultaneously, identify the point with the largest marginal contribution among the unselected candidate points, specifically using the following method:

[0129] For each unselected candidate point q, i.e. Calculate the increase in coverage after adding this point. ,in Indicates will The decision vector after changing from 0 to 1, select The largest candidate point is selected as the point with the largest marginal contribution;

[0130] Try replacing the facility point with the smallest marginal contribution with the unselected candidate point with the largest marginal contribution, calculate the objective function value after the replacement, and if at least one objective function is improved after the replacement, then... or And other objective functions do not deteriorate. and If the replacement is not met, the replacement individual is added to the population and the non-dominated sorting is performed again. If the improvement conditions are not met, the replacement is abandoned. Local search is performed on all individuals in the Pareto front, and each individual attempts replacement at most once.

[0131] The above process is repeated, including non-dominated sorting, crowding calculation, selection, crossover, mutation, elite retention, and periodic local search, until 200 iterations are completed or the Pareto front has not changed significantly for 50 consecutive generations (i.e., the change in the objective function value of individuals in the front is less than a set threshold of 0.1%). After the iteration stops, the final Pareto optimal solution set is output. The Pareto optimal solution set contains multiple non-dominated solutions, each of which corresponds to a location scheme and its objective function value, allowing decision-makers to select the most suitable location scheme based on actual budget and coverage requirements.

[0132] The improved NSGA-II algorithm introduces a spatial clustering mechanism during the initialization phase, ensuring a reasonable spatial distribution of the initial population and avoiding the facility clustering phenomenon caused by traditional completely random initialization, thus improving the quality of the initial population. Through spatial neighborhood-based crossover operations, genetic operations exchange genes within spatially proximate regions, maintaining the spatial coherence of offspring schemes and aligning with the correlation of facility layouts in adjacent areas in practical planning, thereby improving search efficiency. By fusing the crowding distance between the target space and the decision space, it ensures both the uniformity of the Pareto front distribution and enhances the spatial diversity of site selection schemes, resulting in multiple output solutions with significantly different spatial distribution patterns, providing decision-makers with richer choices. Experiments show that compared to the traditional NSGA-II algorithm, the improved algorithm achieves higher demand coverage and a better Pareto front distribution with the same number of iterations, improving convergence speed by approximately 30%.

[0133] Furthermore, a dynamic adjustment mechanism is established to address changes in demand brought about by urban development. This mechanism comprises three stages: regular data updates, existing facility assessment, and site selection optimization. In the regular data update stage, the latest shared transportation platform trajectory data is collected every three to six months. The time span and method of data collection are the same as during the initial site selection. The newly collected data is merged with historical data to form a rolling time window. For example, data from the most recent six months is used for analysis to reflect recent demand trends. The data preprocessing, explicit demand identification, implicit demand identification, and demand fusion processes described in steps S1 to S4 are repeated to generate an updated comprehensive demand heatmap. By comparing comprehensive demand heatmaps from different periods, areas of demand growth and demand decline can be identified. Areas of demand growth refer to regions where the comprehensive demand density has increased by more than 30% compared to the previous period, while areas of demand decline refer to regions where the comprehensive demand density has decreased by more than 30% compared to the previous period.

[0134] In the existing facilities assessment phase, the effectiveness of existing facilities is evaluated based on updated demand data and actual facility usage data;

[0135] Facility usage data includes the actual number of vehicles parked at each facility during different time periods and the turnover rate, which is obtained through the operational data of the shared transportation platform;

[0136] Calculate the average utilization rate for each facility. Identify inefficient facilities with an average utilization rate of less than 30% for two consecutive statistical periods, i.e., six consecutive months. These facilities may have low utilization rates due to reasons such as decreased demand in the surrounding area, unreasonable facility location, or low user awareness. It is recommended to remove or relocate them. Removal is applicable when there is no demand in the surrounding area, while relocation is applicable when there is demand in the surrounding area but the facility location is not good. The relocation target location should be selected from areas with high demand but not covered in the surrounding area.

[0137] Identify overloaded facilities with an average monthly usage rate exceeding 95%. These facilities are operating at full capacity for extended periods, potentially forcing users to park illegally due to lack of parking. It is recommended to expand their capacity, either by adding parking spaces at the original site or by adding new facilities within a 300-meter radius to divert demand. The total capacity after expansion should be able to meet 110% to 120% of the peak-hour demand in the area to allow for redundancy. Identify newly emerging high-demand, uncovered areas—areas with high overall demand density but located more than 500 meters from the nearest existing facility. It is recommended to add new facilities to these areas to improve coverage. Candidate locations for new facilities should be selected from the high-demand grids within the area that meet feasibility criteria.

[0138] In the site selection optimization phase, based on the updated demand data, existing facility assessment results, and a new set of candidate sites, the multi-objective optimization site selection algorithm in step S5 is re-run to generate adjustment schemes. Specifically, non-inefficient facilities among the existing facilities are included in the optimization model as built-up facilities. The corresponding gene loci for these facilities are fixed at 1 in the decision variables and do not participate in the optimization. Only the decisions to remove and relocate inefficient facilities, expand overloaded facilities, and construct new candidate sites are optimized. The objective function is modified to consider the contribution of built-up facilities when calculating demand coverage. The cost function is modified to include the construction cost of new facilities, the expansion cost of expanded facilities, and the demolition cost of removed facilities. Demolition costs are typically significantly lower than construction costs, while expansion costs fall between the two. The improved NSGA-II algorithm is run with the same parameter settings as the initial site selection, outputting a Pareto optimal adjustment scheme set. Each adjustment scheme includes a list of proposed new facilities, a list of proposed removed facilities, a list of proposed expanded facilities, and the expected demand coverage and total cost. Decision-makers can select appropriate adjustment options from the Pareto frontier based on budget constraints and service objectives, and update the existing facilities database after the adjustment is completed to prepare for dynamic adjustments in the next cycle;

[0139] The establishment of a dynamic adjustment mechanism enables parking facility layout to continuously track changes in urban development and user demand, avoiding the problem of static planning schemes gradually becoming ineffective in the face of rapid urban development. Through periodic evaluation and optimization, issues such as low facility utilization and supply-demand imbalances can be identified and addressed promptly, improving the long-term operational efficiency of facilities and reducing resource waste. In practical applications, the dynamic adjustment cycle can be flexibly set according to the speed of urban development and the intensity of shared transportation use; it can be shortened to three months for rapidly developing new areas and extended to one year for mature and stable areas. Decisions regarding dynamic adjustments should be comprehensively determined by combining the development plans of urban planning departments, the operational strategies of shared transportation platforms, and public participation opinions to ensure the scientific validity and feasibility of the adjustment plan.

[0140] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for site selection of shared transportation parking facilities based on multi-source travel data fusion, characterized in that, include: S1. Collect shared transportation platform trajectory data, existing parking facility data and urban spatial data within the study area. After preprocessing the data, map it onto a regular grid system to obtain a gridded spatial data foundation. S2. Based on gridded cycling destination data, the kernel density estimation method is used to calculate the parking demand density of each grid in different time periods, so as to obtain the explicit demand density of each grid in different time periods. S3. Calculate the distance from each cycling endpoint to the nearest existing facility, filter out cycling endpoints with distances exceeding the threshold as off-site parking points, use the DBSCAN density clustering algorithm to identify high-frequency parking hotspot clusters for off-site parking points, calculate the implicit demand intensity of each hotspot cluster and allocate it to the grid it covers, and obtain the implicit demand density of each grid. S4. The explicit demand density and implicit demand density are fused according to weight to obtain the comprehensive demand density of each grid. Grids with comprehensive demand density higher than the set quantile are selected as high demand areas. In the high demand areas, the centers of high demand grids around densely populated interest points are selected as preliminary candidate points and a minimum spacing is set. The preliminary candidate points are screened for feasibility to obtain a set of candidate points. S5. Define a binary decision variable to represent whether a candidate point should be constructed. Establish a dual-objective optimization model and use an improved NSGA-II algorithm for iterative optimization. The improved NSGA-II algorithm includes spatial clustering of candidate points according to their geographical location, initializing the population using a hierarchical strategy, using a crossover operation based on spatial neighborhood, and calculating the distance between the dual spatial crowding of the target space and the decision space for individual selection. After the iteration stops, output the Pareto optimal solution set. Step S5 establishes a bi-objective optimization model, specifically as follows: Define decision variables as Where p is the candidate point index, This indicates that parking facilities will be built at the candidate sites. This indicates no construction; each individual corresponds to a complete decision vector. , which represents a site selection scheme, where N is the total number of candidate sites; Establish a dual-objective optimization model, with the first objective being to maximize demand coverage: ； Where T represents the total number of time periods. Let G be the weighting coefficient for time period t, and G be the set of all grids. Let g be the explicit demand density of grid in time period t. The service capacity of site selection scheme x for grid g in time period t is determined by the sum of the capacities of all selected facility points within the service radius; The second objective is to minimize construction costs: ； in, Cost of constructing facilities at candidate sites; The constraints include: the total cost shall not exceed the budget limit B. And the demand coverage rate is not lower than the minimum requirement. Right now .

2. The method for site selection of shared transportation parking facilities based on multi-source travel data fusion according to claim 1, characterized in that, Step S1 includes: Collect trajectory data of shared transportation platforms within the study area for a time span of no less than a preset number of months, including the starting coordinates, ending coordinates, start time, and end time of each ride; Collect the location coordinates, service radius, and capacity information of existing parking facilities within the study area; Collect urban spatial data, including the location and type attributes of points of interest, road network topology, and sidewalk width, intersection location, and land ownership information of candidate areas; The collected data is preprocessed to convert all coordinates to the same coordinate system and remove abnormal records where the riding speed exceeds the preset speed threshold or the riding time is less than the preset time threshold. The study area is divided into a regular grid with a side length of a preset range, and all data is mapped to this grid system.

3. The method for site selection of shared transportation parking facilities based on multi-source travel data fusion according to claim 1, characterized in that, Step S2 includes: Based on the cycling destination data from the shared transportation platform's trajectory data, the kernel density estimation method is used to calculate the parking demand density for each grid. For the coordinates of any grid center point... Its parking demand density is calculated using the following formula: ； Where n is the total number of cycling destinations, and h is the bandwidth parameter. For Gaussian kernel function, Let k be the Euclidean distance from the k-th cycling endpoint to the center of the grid. Kernel density analysis was performed on the data for four time periods: weekday morning peak, weekday evening peak, weekday off-peak, and all day on weekends, to obtain the explicit demand density of each grid in each time period. , where g is the grid number and t is the time period number.

4. The method for site selection of shared transportation parking facilities based on multi-source travel data fusion according to claim 3, characterized in that, Step S3 includes: Calculate the distance from each cycling destination to the nearest existing facility, and filter out those that are more than a threshold away from the existing facility. The cycling finish line will serve as an off-site parking area. The DBSCAN density clustering algorithm is used to identify high-frequency parking hotspots at the selected off-site parking locations, and a neighborhood radius parameter is set. and minimum sample size parameter ; For each identified high-frequency parking hotspot cluster Calculate its implicit demand intensity: ； Where m is the cluster number. For clusters The total number of parking records within the area. For clusters Spatial density, i.e., the number of times a vehicle is parked per unit area. and These are the weighting coefficients; The implicit demand intensity of each high-frequency parking hotspot cluster is allocated to the grid it covers, thus obtaining the implicit demand density of each grid. .

5. The method for site selection of shared transportation parking facilities based on multi-source travel data fusion according to claim 4, characterized in that, In step S4, the formula for calculating the comprehensive demand density for each grid g is as follows: ； in This is the weighted average of the explicit demand density of the grid in each time period, i.e. T represents the total number of time periods, and the time period weights are... Morning and evening peak hours are assigned the first time period weight value, while off-peak hours and weekends are assigned the second time period weight value. The implicit demand density of this grid. and For fusion weighting coefficients.

6. The method for site selection of shared transportation parking facilities based on multi-source travel data fusion according to claim 5, characterized in that, Feasibility screening of preliminary candidate sites includes: Initial candidate points with sidewalk widths less than a width threshold are eliminated using sidewalk width data; initial candidate points with distances less than a preset length from intersections are eliminated using intersection location data; initial candidate points with property rights disputes are eliminated using land use property rights information; and initial candidate points located in inaccessible locations such as the central median of main roads are eliminated using road network data. After iterating through and filtering all the initial candidate points, the final set of candidate points is obtained.

7. The method for site selection of shared transportation parking facilities based on multi-source travel data fusion according to claim 1, characterized in that, The improved NSGA-II algorithm is used to solve the bi-objective optimization model. The algorithm process includes: Step 1: Spatial cluster all candidate points according to their geographical location into K spatial clusters to generate the initial population; Step 2: Perform non-dominated ranking on the current population and divide individuals into several non-dominated layers according to Pareto dominance relationships; Step 3: Calculate the spatial crowding distance for each individual; Step 4: Select a preset number of parent individuals for breeding by selecting individuals with higher non-dominant levels, and within the same level, prioritize individuals with greater crowding distance. Step 5: Perform a crossover operation based on spatial neighborhood on the selected parent generation to generate offspring; perform a mutation operation on the offspring; merge the parent and offspring generations and recalculate the non-dominated sorting and crowding degree, and select the individuals with higher sorting to enter the next generation. Step 6: Periodically perform local searches to identify and replace the facility points with the smallest marginal contribution and the unselected candidate points with the largest marginal contribution. Step 7: Repeat steps 2-6 until the iteration stopping condition is met, and output the Pareto optimal solution set.

8. The method for site selection of shared transportation parking facilities based on multi-source travel data fusion according to claim 7, characterized in that, The spatial crowding distance for each individual is calculated as follows: For individuals s within the same non-dominated layer, their crowding distance is: ； ； ； Where s is the individual index; For traditional target space congestion distance; Distance for decision-making space congestion; σ represents the weighting coefficients; σ is the index of the objective function. and Let be the objective function value of the individual adjacent to individual s on target o; and Let S be the maximum and minimum objective function values ​​of the current population on target o; S is the set of individuals at the same level as individual s. Indexes other individuals within the same level; The Hamming distance between two decision vectors represents the number of distinct gene loci.

9. The method for site selection of shared transportation parking facilities based on multi-source travel data fusion according to claim 7, characterized in that, The selected parent generation is used to perform a spatial neighborhood-based crossover operation to generate offspring. Specifically, a selected facility point is randomly chosen from the first parent generation and recorded as a candidate point. ,satisfy Determine its spatial neighborhood Distance to candidate points The set of indexes of all candidate points not exceeding a preset distance, finding the one that matches the second parent individual. The selected facility point with the closest centroid distance is recorded as the candidate point. and its spatial neighborhood Swap the two parent generations in the neighborhood. and Gene loci within the genome generate offspring individuals.

Images