A shared bicycle station site selection method based on density and in-out degree balance clustering

Abstract

The application discloses a shared bicycle station site selection method based on density and in-out degree balance clustering, introduces an in-out balance coefficient, and performs clustering on a shared bicycle station candidate area through a composite clustering algorithm of the density and the balance coefficient; a GPS data set of the bicycle is taken as a sample point data set, the sample point data set is clustered by improving and combining the characteristics of a Clique grid clustering algorithm, a plurality of shared station candidate addresses are acquired, each shared station candidate address is optimized, and the best shared station address is determined by combining POI matching of the area. A Boolean expression is used to define complex attribute conditions in a formalized manner, and then a clustering result of the shared bicycle station under the complex attribute conditions is given based on the density and a threshold value. The application has high efficiency and accuracy, can perform data mining on massive user travel OD data, analyze hot spot gathering areas of bicycle user orders, provide good early data preparation for the bicycle station, and has the characteristics of lightness and flexibility.

Description

Technical Field

[0001] This invention relates to the field of data mining and analysis, specifically to a method for selecting shared bicycle stations based on density and in-degree balanced clustering. Background Technology

[0002] In the first and second stages of public bicycle development, the presence of docking stations required bicycles to be parked in fixed locations, significantly restricting user parking behavior without impacting traffic or management. Shared bicycles, a new type of public bicycle sharing service, are primarily operated by commercial companies and rely on the internet. They support smart QR code unlocking, have no fixed parking stations, offer high flexibility, and have high usage density. Adopting a "pick up and use anytime, drop off and go" model, they are highly mobile and flexible, allowing users to choose distances from 1 to 3 kilometers. As a connecting solution to traditional transportation modes, the widespread adoption of shared bicycles demonstrates their flexibility and convenience, effectively solving the "last mile" problem in cities.

[0003] China's market, with its massive user base and high daily usage frequency, boasts one of the world's most active user markets, generating high expectations for the shared bicycle market and demonstrating a generally positive trend in its application. Shared bicycle systems provide urban residents with a low-carbon and environmentally friendly mode of transportation. However, due to the inherent randomness of user usage, different areas often exhibit an imbalance in the number of bicycles. High-demand areas for bicycle rental lack bicycles, while high-demand areas for bicycle return lack sufficient space. This situation is known as the "tidal phenomenon" of shared bicycles, which is more pronounced during morning and evening rush hours. The dispersed nature of shared bicycle parking areas undoubtedly poses a challenge for shared bicycle companies, increasing their daily management costs. For users, peak usage times present challenges such as the lack of nearby bicycles or difficulty in finding a bike. Therefore, setting up parking stations can flexibly address these two issues. This method, also known as electronic fences for shared bicycles, can be used for parking and retrieval.

[0004] Currently, K-means and DBSAN are commonly used in the market and research for clustering traffic nodes. A density-based grid clustering method called Clique has also emerged. Clustering is also frequently used for site selection and planning of shared bicycle stations.

[0005] Traditional clustering methods for shared bicycle stations mostly rely solely on density, neglecting real-world application scenarios. For example, considering spatiotemporal factors, shared bicycles exhibit a tidal phenomenon with peak hours and varying spatial distributions at different times, which limits the effectiveness of clustering in such cases. The selection of shared bicycle stations should consider the inflow and outflow of the area, using this as a clustering condition to maintain a stable balance in the number of bicycles used at each station, facilitating bicycle location and management. To address these issues, this invention proposes a shared bicycle station location selection method based on density and inflow-out-degree balance clustering. Summary of the Invention

[0006] As a further extension of the emerging urban transportation network, shared bicycles have effectively solved the "last mile" problem of urban transportation systems. As a connecting solution to traditional modes of transportation, the widespread adoption of shared bicycles demonstrates their flexibility and convenience. However, precisely because of their overly flexible nature, the manpower and material costs for vehicle retrieval and maintenance are substantial, and improper parking has led to traffic congestion and disruptions to road traffic. Therefore, establishing shared bicycle stations along the roadside can facilitate management for bicycle companies and make it easier for users to find their bikes. Planning and designing recommended parking spots for shared bicycles is of significant research value in the use and management of shared bicycles. It not only helps companies to rationally plan and construct electronic fences but also improves the availability and utilization rate of shared bicycles, alleviates the "tidal phenomenon" of shared bicycles, and increases the proportion of green travel. Setting up stations helps guide and regulate users' parking behavior, avoiding phenomena such as haphazard parking and occupation of other road resources that affect the city's appearance, and helps reduce the operating costs of shared bicycle companies, improving service quality and social benefits. The redistribution of shared bicycles can help alleviate the "difficulty in finding bikes" and congestion caused by the unreasonable distribution of shared bicycles.

[0007] With the rapid development of urban roads and transportation systems, the deployment of shared bicycles has been expanding rapidly. However, the development of these systems has also been affected and limited by various factors, such as the limited number of bicycles and the unstable geographical locations of parking spots. These factors have led to a bottleneck in the development of existing shared bicycle systems, gradually weakening their market appeal. In particular, the issues of shared bicycle space optimization and the imbalance caused by peak passenger flow tidal phenomena are becoming increasingly prominent. Furthermore, the rationality of bicycle deployment point selection and the determination of deployment volume between different stations remain unclear, and an effective spatial configuration model has not yet been developed. Researching the travel characteristics and spatial optimization of shared bicycles is crucial for addressing the imbalance caused by the mismatch between residents' temporal and spatial travel behaviors and bicycle space allocation.

[0008] The technical solution adopted in this invention is to propose a shared bicycle site selection method based on density and in-degree balance clustering. It introduces an in-degree balance coefficient and uses a composite clustering algorithm of density and balance coefficients to cluster candidate areas for shared bicycle sites. By matching different priorities of POI sites, shared sites are established, solving the problem of haphazard parking of shared bicycles and their occupation of urban roads, thus alleviating the management pressure on urban traffic networks. The GPS data of the bicycles is used as a sample point dataset. The Clique grid clustering algorithm is improved and clustered to obtain multiple candidate addresses for shared sites. These candidate addresses are optimized, and the optimal shared site address is determined by combining POI matching for the area.

[0009] Clique grid clustering is only for density clustering. However, shared bicycles are greatly affected by spatiotemporal factors and the frequency of use. Therefore, density should not be used as the only clustering condition. It is necessary to consider adding an in-degree balance coefficient for the number of bicycles entering and leaving a certain area within a day to ensure that the number of bicycles entering and leaving the area is stable at a certain balance threshold, which means that the supply and demand in the area is stable within a certain range.

[0010] The clustering algorithm in this invention involves a process that proposes a community search algorithm based on complex attribute conditions of access balance and density. The complex attribute conditions are formally defined using Boolean expressions, and then clustering results for shared single-site points under these conditions are given based on density and a threshold. Based on the Clique algorithm, a density threshold is pre-defined, and a boundary grid density threshold is used to determine the boundary grid to improve cluster boundary accuracy. This grid clustering algorithm directly projects data objects into the grid space according to a given grid step size and calculates the grid density and balance value. If the grid density and balance value are greater than the given density threshold, the grid is determined to be a dense and balanced grid, and the cluster is added to the current cluster. The search continues until all adjacent grids are non-dense and non-balanced, ending the loop. This process is repeated until all clusters are found.

[0011] The proposed solution is efficient and accurate. It provides an effective solution for shared bicycle travel demand under the massive data volume in megacities. It performs data mining on massive user travel OD data to analyze and identify hotspots of bicycle user orders, providing good preliminary data preparation for bicycle stations. It is also lightweight and flexible. Attached Figure Description

[0012] Figure 1 This is a diagram of the overall solution architecture.

[0013] Figure 2 This is a schematic diagram of data preprocessing.

[0014] Figure 3 This is a diagram of the clustering algorithm structure.

[0015] Figure 4 This is a diagram of the address selection algorithm. Detailed Implementation

[0016] The technical solution of the present invention will now be described clearly and in detail with reference to the accompanying drawings in the embodiments of the present invention.

[0017] The core solution of this invention is a method for selecting shared bicycle sites based on improved CLIQUE grid clustering. This method involves: preprocessing GPS data from bicycle lock / unlock orders; determining passenger travel trajectories based on the preprocessed data to obtain a set of latitude and longitude coordinates of discrete destinations; clustering these coordinates; and using an improved BLA-CLIQUE clustering method during density clustering to identify the start and end clusters of passenger orders. Further iterative processing is then performed on these clustering results to obtain the iteratively processed clusters of passenger destinations, i.e., the clustering results. Finally, based on these clustering results, Points of Interest (POIs) are matched according to preset rules to determine the required candidate shared bicycle sites.

[0018] Predefined: 1) Mesh cells: DS = D1, D2, D3, ..., D n It is an n-dimensional dataset with subspace D i It is expanded to be equal to the largest subspace D max And D i Based on grid step size g s It is divided into m equal intervals, thus making D i Divide into m non-intersecting rectangular units, i.e., grid units g. 2) Dense units: Given a density threshold θ d When the number of data objects x > θ projected onto the range g by DS d When g is a dense element gd; for non-dense elements... Conversely, the same applies. 3) Balance unit: Given a balance threshold θ b When the number of data objects x > θ projected onto the range g by DS d When g is the equilibrium unit g b For non-equilibrium units Conversely, the same applies. 4) Boundary mesh: Given a threshold ε, g≠g b and g b ∈{adjacent grids of g} if ρ g If > θ, then g is the boundary grid.

[0019] The preprocessing method involves denoising and processing the order data of shared bicycles, removing data with obvious location offsets and errors, cropping the order data to retain only the GPS latitude and longitude data for unlocking and locking, and generating a new CSV file.

[0020] After preprocessing, a density threshold and balance coefficient are defined. A sample point dataset is obtained, and the city is divided into several grids of equal length and width according to a customizable grid granularity. Each grid cell is marked as unvisited. An unvisited sample grid is randomly selected from this set and marked as visited. It is then determined whether the selected grid cell meets both the density and balance coefficient thresholds. If so, the grid cell is marked as meeting the conditions; otherwise, no processing is performed. This process iterates until a grid meets the conditions, at which point a neighboring grid is searched. The selected grid cell is first checked to see if it belongs to another cluster. If so, it is marked as visited; otherwise, the sample data from the neighboring grid is merged into the two grids. The density and in-degree balance coefficient of this merged grid are recalculated, and the density and threshold conditions are checked again. If they meet, they are merged into a single cluster; otherwise, no processing is performed. After iterating through all the city grid cells, multiple clusters are obtained. The grid granularity, density threshold, and balance coefficient can be customized to adapt to different clustering conditions. For example, the number of shared bicycles in the clustered urban grid area should not be less than a certain threshold, and the inflow and outflow of bicycles in the grid area should be consistently greater than a certain threshold, indicating that the area is frequently used by vehicles and can be considered as a candidate area for site selection. The following is a detailed description of the BLA-CLIQUE clustering algorithm used:

[0021] 1) Divide the sample set into grid cells of custom size according to the city's latitude and longitude. Each grid cell is the same size, and mark all cells as "unvisited".

[0022] 2) Randomly select a cell as cluster number 0, and mark this cell as "visited";

[0023] 3) Traverse the surrounding grid with this cluster as the center. First, determine whether it has been visited. If it has been visited, it means that it has been gathered into other clusters and no processing is required. If it has not been visited, merge it with the previous cluster and recalculate its density and balance threshold. When its density is greater than the density threshold and its balance coefficient is greater than its balance coefficient threshold, add it to the cluster and mark it as visited. If the conditions are not met, no processing is required.

[0024] 4) Iterate through the clusters in sequence, and finally the clusters are completed to obtain a set of clusters, which is the region of the candidate site.

[0025] The specific method for matching POI sites based on the clustering results and determining the required shared bicycle candidate sites according to preset rules is as follows: 1) Set the buildings, residential areas, hotels, supermarkets and subway stations in the area as the first, second, third and fourth priorities, respectively; 2) Obtain the coordinates of the center point of each cluster based on the clustering results. The specific steps are as follows: 1) MapGis software can be used to connect the latitude and longitude points of all sample data within each cluster to construct the corresponding spatial region, and then extract and output the coordinates of the center point; 2) Using the center point coordinates corresponding to the cluster number obtained in step 2), for these coordinates, firstly, POI (Point of Interest) site matching is performed within the set range. If a suitable stop is selected, and if it is a first or second level POI, the final result is output. If no suitable site is matched, the matching range needs to be further expanded; if no first or second priority site is matched, then third or fourth priority POI sites are considered for matching; 3) By dynamically matching POIs, the candidate sites corresponding to each cluster are obtained; 4) Finally, considering the passengers' spending power and travel preferences, trial operation sites can be set up as shared bicycle sites.

[0026] The advantages of this invention compared to existing technologies are as follows: The method for selecting candidate sites for shared bicycles provided by this invention, during the clustering stage, not only uses commonly used density as a clustering condition but also considers the vehicle usage status during clustering and significantly reduces data dimensionality. This aims to achieve the goal of simultaneously measuring both conditions during the clustering process, ultimately improving the accuracy and relevance of the clustering results. This invention is based on the BLA-CLIQUE clustering algorithm, improving upon the original CLIQUE grid clustering algorithm and increasing the data volume for clustering computation. Finally, through POI site matching based on specific rules, it can meet the candidate site selection scenarios with different needs, achieving fast convergence and high accuracy. It should be understood that although this specification describes embodiments, not every embodiment contains only one independent technical solution. This descriptive method is merely for clarity. Those skilled in the art should consider the specification as a whole, and the technical solutions in various embodiments can be appropriately combined and implemented according to the understanding of those skilled in the art.

[0027] The detailed descriptions listed above are merely specific descriptions of feasible embodiments of the present invention and are not intended to limit the scope of protection of the present invention. All equivalent embodiments or modifications made without departing from the spirit of the inventive technique should be included within the scope of protection of the present invention.

Claims

1. A shared bicycle station site selection method based on density and betweenness centrality balanced clustering, characterized in that, Firstly, the in-out balance coefficient is introduced, and the candidate area of the shared bicycle station is clustered through the composite clustering algorithm of density and balance coefficient; the shared station is set up through the different priority matching of the POI station, the GPS data set of the bicycle is taken as the sample point data set, the characteristics of the Clique grid clustering algorithm are improved, and the sample point data set is clustered to obtain multiple shared station candidate addresses, the shared station candidate addresses are optimized, and the best shared station address is determined by combining the POI matching in the region; The complex attribute condition is defined formally using a Boolean expression, and then the clustering result of the shared bicycle station under the complex attribute condition is given based on density and threshold; based on the Clique algorithm, the density threshold is determined in advance, and the boundary grid density threshold is used to determine the boundary grid to improve the cluster boundary accuracy, the grid clustering algorithm will project the data object directly into the grid space according to the given grid step, and calculate the grid density and balance value, if the grid density and balance value is greater than the given density threshold, the grid is determined to be dense and balanced, and is added to the current cluster and the grid is searched, until all adjacent grids are non-dense and non-balanced grids, the loop is ended; the above operation is repeated until all clusters are found; The GPS data of the bicycle lock and unlock order is preprocessed, the passenger's travel trajectory is determined according to the preprocessed data, and the latitude and longitude set of the discrete destination of the passenger's destination is obtained; the latitude and longitude set of the discrete destination of the passenger's destination is clustered, in the density clustering process, the improved BLA-CLIQUE clustering method is used to obtain the aggregation area of the start and end of the passenger's order, and the clustering result of the aggregation area of the start and end of the passenger's order is iteratively processed to obtain the aggregation area of the passenger's destination after iteration, that is, the clustering result, and according to the preset rule, the POI station is matched according to the clustering result, and finally the required shared bicycle candidate station is determined; the specific description of the BLA-CLIQUE clustering algorithm is as follows: 1) The sample set is divided into grid cells of a user-defined size according to the city latitude and longitude, each grid cell is of the same size, and all cells are marked as unvisited "unvisited"; 2) A cell is randomly selected as a cluster with a number of 0, and the cell is marked as visited "visited"; 3) The surrounding grid is traversed with the cluster as the center, first it is judged whether it is visited, if it is visited, it means that it has been aggregated into other clusters, and no processing is done, if it is not visited, it is merged with the previous cluster and the density and balance threshold is recalculated, when the density is greater than the density threshold and the balance coefficient is greater than the balance coefficient threshold, it is added to the cluster and marked as visited "visited", if it does not meet the requirements, it is not processed; 4) Iteration is performed in turn, and finally a cluster set is obtained, which is the candidate station area; The specific method for matching the POI station according to the preset rule based on the clustering result is as follows: 1) setting a building, a residential area, a hotel, a supermarket and a subway station in a region as a first, a second, a third and a fourth priority respectively; 2) obtaining the center point coordinates of each cluster according to the clustering result; the specific steps are as follows: using MapGis software, connecting the longitude and latitude points of all sample data in each cluster to construct a corresponding space region, then extracting the coordinates of the center point and outputting; 3) obtaining the center point coordinates corresponding to the number of cluster numbers through 2), first matching the POI station in the set range, if a required parking station is selected, if it is a first or second level POI, the final result is output, if no suitable station is matched, the matching range needs to be further expanded for matching; if no first or second priority station is still matched, the third or fourth priority POI station is considered for matching; 4) obtaining the candidate station corresponding to each cluster by dynamically matching the POI; 5) finally considering the passenger's consumption ability and travel willingness preference, setting a trial operation station as a shared bicycle station. 2.The shared bicycle station site selection method based on density and balance clustering of in-out degrees according to claim 1, characterized in that, Predefined: 1) Grid cell: DS = D1, D2, D3,..., D n is an n-dimensional dataset, a subspace D i is extended to equal the largest subspace D max and D i is divided into m equal intervals according to the grid step gs, thus dividing D i into m disjoint rectangular cells, namely grid cells g; 2) Dense cell: given density threshold θ d when the number of data objects x that DS projects into range g > θ d then g is a dense cell gd; for non-dense cells and vice versa; 3) Balanced cell: given balance threshold θ b when the number of data objects x that DS projects into range g > θ b then g is a balanced cell gb; for non-balanced cells and vice versa; 4) Boundary grid: given threshold ε, g ≠ g b and g b ∈ {adjacent grids of g}; if ρ g > θ, then g is a boundary grid. 3.The shared bicycle station site selection method based on density and accessibility balanced clustering of claim 1, wherein, The preprocessing method is to denoise and process the sample of the shared bicycle order data, remove the data with obvious positioning deviation and errors, crop the order data, only keep the GPS longitude and latitude data of unlocking and locking, and generate a new csv file. 4.The shared bicycle station site selection method based on density and balance clustering of access degrees according to claim 3, characterized in that, After preprocessing, the density threshold and the balance coefficient are defined, the sample point data set is obtained, the city is divided into a plurality of equal length and width grids according to the grid granularity which can be defined, and the grid units are marked as unvisited state; a unvisited sample grid is randomly selected from the grid unit set and marked as visited state; it is judged whether the selected grid unit satisfies the density threshold and the balance coefficient threshold; if yes, the grid unit is marked as a grid satisfying the condition; if not, no processing is performed; iteration is performed, when the grid satisfies the condition, a grid around the grid is searched, it is first judged whether the selected grid unit belongs to other cluster; if yes, the grid unit is marked as visited state; if not, the sample data of the surrounding grid is merged into two grids, the density and the in-out degree balance coefficient of the merged grid are recalculated, and it is judged whether the threshold conditions are satisfied; if yes, the grid is merged into a cluster, if not, no processing is performed; after iteration of the entire city grid unit, a plurality of clusters are obtained; wherein the grid granularity, the density threshold and the balance coefficient are defined to adapt to different clustering conditions.

Images