Bus route planning and visual analysis method of influencing factors based on GPS data

Abstract

The present invention relates to a bus route planning and visual analysis method for influencing factors based on GPS data. First, city taxi order data and city point of interest data are preprocessed to obtain passenger travel origin-destination data and POI category data. Passenger boarding and alighting hotspot grids are extracted, and candidate bus stops are mined based on grid clustering. Furthermore, a multi-objective model for bus route planning is constructed, and an optimal bus route set is generated through a multi-objective optimization method. Attribute weight quantification and ranking results are obtained based on statistical route attribute information. Finally, a visual analysis system supports planners in exploring and analyzing routes in the optimal route set and supports changing route attribute weights. The system retrains based on the adjusted attribute weights or subjective ranking to update the ranking results. Through interactive exploratory analysis, an optimal bus route planning solution is obtained, providing a scientific decision-making basis and practical means for transportation departments to carry out bus route planning, thereby facilitating the development of smart cities.

Description

Technical Field

[0001] The present invention belongs to the technical field of public transportation route planning, and in particular relates to a method for public transportation route planning and visual analysis of influencing factors based on GPS data. Background Art

[0002] Public transportation supports and guides sustainable urban development. Prioritizing public transportation development is an effective way to address environmental pollution, traffic congestion, and improve urban transport efficiency. However, with social development and changes in travel patterns, urban bus routes are no longer adequately meeting passengers' daily travel needs, and operational efficiency still needs to be improved. Bus route planning requires a holistic approach, considering the combined impact of multiple factors, seeking a synergy between urban development and actual needs, and balancing the costs and benefits of both travelers and operators. However, accurately understanding passenger travel needs, developing bus route planning solutions that meet multiple objectives, and quantifying and analyzing the influencing factors and their weightings presents significant challenges.

[0003] Traditional bus stop discovery and route design methods rely on population censuses or travel intention surveys. These methods collect relatively subjective and small-scale samples. Bus route planning based on such limited samples fails to truly meet passenger travel needs and is time-consuming and labor-intensive. Methods that use trajectory data to perceive passenger travel needs and place bus stops lack consideration of the overall distribution of bus stops in a city, and the accuracy of candidate bus stop discovery remains to be improved. Furthermore, the vast solution space of possible bus route planning options makes improving the efficiency and accuracy of finding the optimal route set extremely challenging. It is necessary to achieve a balanced optimal bus route set that is mutually constrained by multiple objectives, while ensuring both effectiveness and efficiency. Furthermore, quantitative analysis of route influencing factors and their weights is crucial for comparing candidate routes and developing reasonable route planning solutions. This can improve the accuracy and credibility of optimal route recommendations. Summary of the Invention

[0004] To address the above issues, the present invention proposes a visual analysis framework for bus route planning that can perceive passenger travel needs. It organically integrates intelligent processing of urban traffic data, visual representation, and interactive analysis to identify passenger travel hotspots and improve the accuracy of bus stop mining and route design. It also quantitatively analyzes the impact of route attribute weights on the formulation of route planning schemes, providing scientific decision-making basis and practical means for transportation departments to carry out bus route planning, thereby promoting the development of smart cities.

[0005] The technical solution adopted in the present invention is as follows:

[0006] The present invention provides a bus route planning and influencing factor visual analysis method based on GPS data, characterized in that the method comprises the following steps:

[0007] Step 1: Collect city taxi order data and city point of interest data;

[0008] Step 2: Preprocess the collected data to obtain passenger travel origin-destination OD data and POI category data;

[0009] Step 3: Based on the data obtained in step 2, extract the city's hotspot boarding and alighting grids and cluster them. Determine the candidate bus stop locations based on the clustering results, and calculate the passenger flow transfer volume and vehicle travel time between candidate stops.

[0010] Step 4: Construct a multi-objective model for bus route planning, solve the model based on the NSGA-II algorithm, and obtain the optimal bus route set;

[0011] Step 5: Use Ranking SVM to quantify the attribute weights, sort, and calculate the contribution values ​​of the routes in the optimal bus route set to correct the output results of the bus route planning multi-objective model;

[0012] Step 6: Perform visual analysis on the routes in the optimal bus route set to obtain the optimal bus route planning solution.

[0013] In one embodiment, in step 2: data preprocessing includes at least one of data desensitization, filtering duplicate data, removing missing data in major fields, removing invalid location information data, or coordinate system conversion and timestamp calculation.

[0014] In one embodiment, in step 3, the step of extracting boarding and alighting grids at city hotspots based on the data obtained in step 2 includes:

[0015] Step S1: Input the passenger boarding and alighting location point set P = <p1,p2,...,p n >, p i =(lon i ,lat i ), p i Indicates the location of boarding and alighting, n indicates the total number of boarding and alighting, lon i with lat i Indicates the latitude and longitude of the corresponding point;

[0016] Step S2: Initialize the grid set Step size step = 0.0001, reference point longitude and latitude Lon, Lat, so that lon i ≤Lon, lat i ≥lat all hold true;

[0017] Step S3: Traverse the points p in the set P i, calculate the longitude and latitude of the grid to which it belongs, lon g ,lat g :

[0018] lon g =floor((lon i -Lon) / step)*step+Lon

[0019] lat g =floor((lat i -Lat) / step)*step+Lat

[0020] In the formula, floor represents the rounding function, that is, the grid to which it belongs is calculated according to the step size;

[0021] Step S4: Determine the grid g i =(lon g ,lat g )∈G holds true:

[0022] If not, g i Add to the set G and initialize the grid connectivity Number of boarding and alighting within the grid otherwise, And execute the next step;

[0023] Step S5: Calculate (lon g ±step,lat g ±step)、(lon g ±step,lat g )、(lon g ,lat g ±step), determine whether they are in the set G, and get the number w of existence, such that

[0024] Step S6: After all boarding and alighting points are traversed, the boarding and alighting hotspot grid set G is output.

[0025] Furthermore, the grid clustering method in step 3 includes:

[0026] Step S1: Input: G = <g1,g2,...,g n >, n is the number of grids, where each grid g i =(lon i ,lat i ,CD i ,PDR i );

[0027] Step S2: Set a walkable distance threshold Max_Raduis, which is 500m by default, and the cluster center set

[0028] Step S3: Add g1 as the cluster center to the set C;

[0029] Step S4: Traverse all grids in G, for grid g i , calculate the distance SD(g i ,c j ), find the cluster center grid c with the smallest distance; if SD(g i ,c)≤Max_Radius, then the grid belongs to this cluster, add it to the cluster, and execute the next step; otherwise, the grid g i Add to C;

[0030] Step S5: Calculate the values ​​of all grids in the cluster to which the cluster center grid c belongs by the following formula, and take the grid c with the largest value new Add the mesh as the cluster center to C and delete c:

[0031]

[0032] Where w1 is the weight of the grid connectivity, w2 is the weight of the grid passenger flow, n is the number of grids in the cluster to which the cluster center grid c belongs, is the maximum function;

[0033] Step S6: After all grids are traversed, the candidate bus station set C is output. <c1,c2,...,c m >, m is the number of bus stops.

[0034] Furthermore, the calculation method of the real space distance SD between GPS track points is:

[0035] SD=Δσ*R (3-1)

[0036]

[0037] Where R represents the average radius of the Earth, and Δlat and Δlon in Formula 3-2 represent the longitude and latitude differences between two GPS points, respectively.

[0038] In one embodiment, the multi-objective model for bus route planning in step 4 includes: minimizing the route running time function and maximizing the route cumulative passenger flow function.

[0039] The function of minimizing line running time includes:

[0040] ΔT R =h+T(sr|R| ,s n ) (4-1)

[0041]

[0042] Where: ΔT R is the total running time of the bus between each two stations, h is the waiting time at each station, |R| represents the current number of stations in route R, s i represents the bus stop at station i, T is the vehicle travel time between two stations, which can be obtained from the Traveltime matrix, R=(r1,r2,...,r k ) represents the set of stations that have been added to the line and the destination that has not been reached. d , s n For the newly added bus stops, the running time of the entire route is the sum of the travel time between two adjacent bus stops and the waiting time of all stops on the entire route;

[0043] The function for maximizing the cumulative passenger flow of a route includes:

[0044]

[0045]

[0046] Where: ΔD R is the cumulative passenger flow of all stations in the line, and F is the transfer passenger flow between two stations, which can be obtained from the Passenger matrix.

[0047] Furthermore, the function of minimizing route operating time and the function of maximizing route cumulative passenger flow also have the following constraints:

[0048] (1) Adequate site spacing;

[0049] dist(s i+1 ,s i )<δ,(i=1,2,...,n-1) (4-5)

[0050] Where: δ is the maximum distance between two stations, set δ = 1.5 km, dist is the distance between any two stations;

[0051] (2) The vehicle is always traveling in the direction of the destination;

[0052]

[0053]

[0054] Where: (x(i), y(i)) is the station s iThe difference between the longitude and latitude of the departure point s1, that is, the direction vector, (x(n), y(n)) is the destination point s n The difference between the latitude and longitude of the starting point s1 is the direction vector. This constraint can ensure that the bus always moves in the direction of the destination. The constraint formula is shown in 4-6.

[0055] (3) Far from the starting point. When selecting sites, it is necessary to ensure that each new site is far away from the starting point s1. The constraint formula is shown in 4-7:

[0056] dist(s i+1 ,s1)>dist(s i ,s1),(i=1,2,...,n-1) (4-7)

[0057] (4) Close to the destination. When selecting a station, ensure that each new station is closer to the destination. n , as shown in formula 4-8:

[0058] dist(s i+1 ,s n )<dist(s i ,s n ),(i=1,2,...,n-1) (4-8)

[0059] (5) No zigzag route. This condition is to ensure the smooth flow of the generated route and no "Z"-shaped route is generated. The constraint formula is shown in 4-9:

[0060]

[0061] Where: dist is the distance between the two sites, is the minimum function.

[0062] In one embodiment, step 4 solves the model based on the NSGA-II algorithm to obtain the optimal bus route set. The specific method process is as follows:

[0063] Step S1: Input candidate bus station set S = {s1, s2, s3...}, where s i =(lon i ,lat i );

[0064] Passenger flow matrix Passenger, travel time matrix Traveltime;

[0065] Origin o and destinations d , s o ,s d ∈S;

[0066] Step S2: Initialize the total number of customized bus routes, that is, set the maximum population size N and the maximum number of iterations g max , set the number of iterations g = 1;

[0067] Step S3: Generate an initial population, that is, generate a set of feasible bus routes of size N;

[0068] Step S4: Calculate the objective function value of each line, formula 4-2 and formula 4-4;

[0069] Step S5: Perform a genetic crossover operation on the current population. For each line, randomly select a site to exchange with another individual and the constraints must be met at the same time, thereby generating a new line.

[0070] Step S6: Perform a genetic mutation operation on the current population. For each route, randomly select two sites in one of the routes, and replace the original route with a new route between the two sites while satisfying the constraints to generate a new route.

[0071] Step S7: Calculate the objective function value of the new route and merge the old and new routes. Perform fast non-dominated sorting on all individuals according to the two objective functions, calculate the individual crowding distance, and perform elite retention strategy operations to form the next generation route set, and g = g + 1;

[0072] Step S8: When g≤g max When g>g max When , the algorithm ends, and the line set at this time is P od .

[0073] Furthermore, the roulette probability selection method is adopted in step S3, and the steps are as follows:

[0074] It is used to express the probability of each site being selected, where S * The line contains s j All possible sites to connect to

[0075] allowSize=length(P) / / Number of candidate sites;

[0076] selectP=Random(0,1) / / pregenerate a pseudo-random number uniformly distributed between [0, 1];

[0077] selectNode=0 / / Initialize the number of selected sites to 0 (site numbers start from 0);

[0078] sumP=0 / / Initialize the cumulative probability to 0;

[0079] startIndex = Random(allowedSize) + 1 / / Randomly generate the starting position of the roulette wheel;

[0080] while sumP<selectP do:

[0081] sumP=sumP+P[(startIndex-1)%allowedSize] / / Calculate cumulative probability;

[0082] startIndex=startIndex+1;

[0083] end while;

[0084] selectNode=(startIndex-2)%allowedSize / / Select the site index to be added to the line;

[0085] Starting from the starting station, selectNode is selected according to the constraints, and a bus route is generated until the destination. The iteration is carried out until the scale is N.

[0086] In one embodiment, weight quantification is implemented for five attributes of the optimal bus route set: cumulative route passenger flow, route operating time, route length, number of stations, and number of accessible POIs. Cumulative route passenger flow and route operating time are important attributes in bus route planning research. Route length and number of stations can reflect the investment cost of route operation. The number of accessible POIs measures the reach of the route and the number of POIs from the perspective of accessibility.

[0087] Due to the adoption of the above technical solution, the beneficial effects of the present invention are:

[0088] In the candidate bus stop discovery phase, this technical solution analyzes daily passenger travel data from a data-driven perspective and proposes a grid-based approach to perceive passenger travel demand. This approach clusters grids based on factors such as hotspot grid connectivity, the number of boarding and alighting within the grid, and the acceptable walking distance for passengers, thereby improving the accuracy of candidate bus stop discovery. This technical solution constructs a multi-objective model for route planning, incorporating key concepts from the non-dominated sorting genetic algorithm (NSGA-II) multi-objective optimization approach to solving the optimal bus route set. This approach prioritizes the synergistic effects of route planning and achieves an equilibrium solution between travelers and operators, maximizing cumulative passenger flow while minimizing route operating time. Integrating bus route planning design principles, this solution proposes a multi-objective optimization method based on NSGA-II to generate route planning candidates, improving the accuracy of generated routes and the efficiency of finding the optimal set. A ranking support vector machine (SVM) approach is used to quantify route attribute weights, enabling planners to modify weights and perform subjective ranking operations, thereby enhancing the accuracy and reliability of route ranking recommendations. This paper proposes a visual analysis framework for bus route planning that perceives passenger travel needs, improves the accuracy of bus stop mining and route design, quantifies the impact of route attribute weights on the formulation of route planning schemes, and provides valuable decision-making basis for transportation departments in bus route planning. BRIEF DESCRIPTION OF THE DRAWINGS

[0089] The present invention will now be described by way of example with reference to the accompanying drawings, in which:

[0090] Figure 1 This is a flow chart of the architecture of a method for bus route planning and visual analysis of influencing factors based on GPS data provided by an embodiment of the present invention;

[0091] Figure 2 Schematic diagram of data preprocessing provided by an embodiment of the present invention;

[0092] Figure 3 This is a city POI information map provided by an embodiment of the present invention;

[0093] Figure 4 This is a boarding and alighting hotspot grid map provided by an embodiment of the present invention;

[0094] Figure 5 This is a graph showing how the number of clusters changes with Max_Radius, provided by an embodiment of the present invention;

[0095] Figure 6 This is a graph showing changes in the evaluation index along with Max_Radius provided by an embodiment of the present invention;

[0096] Figure 7This is a clustering result diagram provided by an embodiment of the present invention;

[0097] Figure 8 This is a comparison chart of clustering results provided by an embodiment of the present invention;

[0098] Figure 9 This is a comparison diagram of the geographic spatial locations of candidate bus stops provided by an embodiment of the present invention;

[0099] Figure 10 This is a schematic diagram of the Ranking SVM principle provided by an embodiment of the present invention;

[0100] Figure 11 Schematic diagram of the application of Ranking SVM on bus route attributes provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0101] In order to enable those skilled in the art to better understand the solutions of the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the embodiments described are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts should fall within the scope of protection of the present invention.

[0102] The research framework of the technical solution in this invention includes five modules: data preprocessing, candidate bus station mining, optimal bus route set solution, route attribute weight quantification, and visual analysis. First, urban taxi order data and urban point of interest (POI) data are preprocessed to obtain passenger travel origin-destination (OD) data and POI category data; then, an optimal bus route set is generated, and attribute weight quantification and ranking results are obtained based on statistical route attribute information; finally, the visualization system supports planners in exploring and analyzing routes in the optimal route set and supports changing route attribute weights. The system will retrain to update the ranking results based on the adjusted attribute weights or subjective ranking. Through interactive exploration and analysis, a satisfactory bus route planning solution is obtained, providing decision support for route planners.

[0103] like Figure 1 Specifically, the method includes the following steps:

[0104] Step 1: Collect city taxi order data and city point of interest data;

[0105] Step 2: Preprocess the collected data to obtain passenger travel origin-destination OD data and POI category data;

[0106] Step 3: Based on the data obtained in step 2, extract the city's hotspot boarding and alighting grids and cluster them. Determine the candidate bus stop locations based on the clustering results, and calculate the passenger flow transfer volume and vehicle travel time between candidate stops.

[0107] Step 4: Construct a multi-objective model for bus route planning, solve the model based on the NSGA-II algorithm, and obtain the optimal bus route set;

[0108] Step 5: Use Ranking SVM to quantify the attribute weights, sort, and calculate the contribution values ​​of the routes in the optimal bus route set to correct the output results of the bus route planning multi-objective model;

[0109] Step 6: Perform visual analysis on the routes in the optimal bus route set to obtain the optimal bus route planning solution.

[0110] In a specific embodiment, data acquisition and preprocessing include:

[0111] The candidate bus stop mining work of this application is based on the urban taxi order data. The boarding and alighting records of taxi passengers reflect the travel needs of people in the city. Analyzing them from the perspective of data mining can perceive the travel needs of passengers, and thus make the locations of the generated candidate bus stops more reasonable. Based on this, the taxi order data generated by a certain city in August 2021 were collected and processed for candidate bus stop mining. As shown in Table 1, each piece of data contains seven fields, namely, order number (desensitized), boarding time, disembarkation time, longitude and latitude of the passenger boarding location, and longitude and latitude of the disembarkation location.

[0112] Table 1 Data description of taxi order dataset

[0113]

[0114]

[0115] The data preprocessing operation is mainly to remove dirty data and improve the accuracy of data analysis. The data set used in the experiment of this application is the taxi order data of a certain city from August 1, 2021 to August 26, 2021. Due to the inevitable loss and deviation problems in the data collection process, the original data cannot be directly analyzed. Therefore, this application performs the following on the above taxi order data set: Figure 2 The data preprocessing operation shown first desensitizes the data, further filters out duplicate data, removes missing data in key fields (such as the lack of latitude and longitude information that makes it impossible to locate the boarding and alighting locations), removes invalid location information data (outside the urban area), converts the coordinate system and timestamp, and finally obtains the OD data of passenger travel.

[0116] Through the above operations, we finally obtained 2,251,274 valid taxi order data, with a total of 4,502,548 passenger boarding and alighting behaviors. Combined with the investigation and research, this application assumes that the frequency of urban bus operation is half an hour. From 0:00 to 4:00 in the morning, passenger travel shows a downward trend. From 4:00 to 8:00 in the morning, with the start of the morning rush hour, the number of taxi boarding and alighting gradually increases. During the daytime, passenger travel is stable until 21:00 at night, with the end of the evening rush hour, the number of boarding and alighting shows a downward trend, which is in line with the actual travel patterns of urban residents. It is more reasonable to use the taxi boarding and alighting data of this city for data mining and analysis.

[0117] In addition, in order to increase users' understanding of route functionality and improve the route planning process, this application also processes urban POI data, which can achieve the above purpose by calculating the types and number of POIs covered by the accessible areas of candidate bus stops obtained through mining. POI refers to information labeling data of urban entities or buildings in urban areas that are closely related to the daily lives of residents, such as schools, shopping malls, parks, and various institutions. The POI dataset mainly describes the geographical location and categories of these places. The POI data within the entire city contains rich information. Combining it with urban traffic data for exploration can reflect potential activities in the city to a certain extent. The POI dataset used in this application also comes from official statistics in 2021, corresponding to the year of taxi order data. The fields of each data include latitude and longitude, address name, and category code, as shown in Table 2. The POI data finally used in this application are 10 major categories sorted out on this basis. The names and quantities of various categories are as follows. Figure 3 As shown, there are ten categories including catering services, accommodation and residence, scenic spots, science, education and culture, institutions and groups, life services, shopping malls and supermarkets, leisure and entertainment, companies and enterprises, and transportation facilities.

[0118] Table 2 POI dataset data description

[0119]

[0120] In a specific embodiment, the grid extraction and grid clustering part includes:

[0121] Pickup and drop-off hotspot grid extraction

[0122] After obtaining the passenger travel demand data (i.e. OD data), it is necessary to further explore the hot spots of urban traffic on this basis to provide a basis for mining candidate bus stops. This application proposes a method to aggregate boarding and alighting points into a grid, and mark the connectivity degree (CD) and the number of boarding and alighting records (PDR) of each grid to determine the traffic hot spots. In the boarding and alighting hotspot grid extraction algorithm, step represents the step length used to calculate the grid, that is, the side length of the divided grid, which is 0.0001°≈11m; the longitude and latitude of the reference point Lon, Lat respectively represent the longitude and latitude of the upper left vertex of the minimum rectangle containing the urban geographic space. Using the above data to extract the boarding and alighting hotspot grid, a total of 293,496 boarding and alighting hotspot grids are obtained, such as Figure 4 As shown in the figure, each grid is marked with two indicators, CD and PDR, which provide a basis for mining candidate bus stops.

[0123] The steps for extracting boarding and alighting grids at urban hotspots based on the data obtained in step 2 include:

[0124] Step S1: Input the passenger boarding and alighting location point set P = <p1,p2,...,p n >, p i =(lon i ,lat i ), p i Indicates the location of boarding and alighting, n indicates the total number of boarding and alighting, lon i with lat i Indicates the latitude and longitude of the corresponding point;

[0125] Step S2: Initialize the grid set Step size step = 0.0001, reference point longitude and latitude Lon, Lat, so that lon i ≤Lon, lat i ≥Lat all hold true;

[0126] Step S3: Traverse all points in P, for point p i , calculate the longitude and latitude of the grid:

[0127] lon g =floor((lon i -Lon) / step)*step+Lon

[0128] lat g =floor((lat i -Lat) / step)*step+Lat

[0129] Where floor represents the rounding-down function, that is, the grid to which the step size is rounded down is calculated;

[0130] Step S4: Determine the grid g i =(lon g ,lat g )∈G holds true:

[0131] If not, g i Add to the set G and initialize the grid connectivity Number of boarding and alighting within the grid otherwise, And execute the next step

[0132] Step S5: Calculate (lon g ±step,lat g ±step)、(lon g ±step,lat g )、(lon g ,lat g ±step), determine whether they are in the set G, and get the number w of existence, such that

[0133] Step S6: After all boarding and alighting points are traversed, the boarding and alighting hotspot grid set G is output.

[0134] Grid clustering and candidate bus stop generation

[0135] Data clustering can be used to group data and use cluster centers to represent certain meanings. However, current clustering algorithms are not well suited for candidate bus stop generation tasks, which requires clustering hotspot grids to obtain the cluster center of each cluster as a candidate bus stop. Therefore, this application proposes a candidate bus stop mining method (Identification of Candidate Bus Stops, ICBS) based on GPS data to perceive passenger travel needs, comprehensively considering the connectivity of passenger travel hotspot areas and the service range of bus stops, to improve the rationality and accuracy of candidate bus stop mining. The ICBS algorithm is as follows:

[0136] Step S1: Input: G = <g1,g2,...,g n >, n is the number of grids, where each grid g i =(lon i ,lat i ,CD i ,PDR i );

[0137] Step S2: Set a walkable distance threshold Max_Raduis, which is 500m by default, and the cluster center set

[0138] Step S3: Add g1 as the cluster center to the set C;

[0139] Step S4: Traverse all grids in G, for grid g i , calculate the distance SD(g i ,c j ), find the cluster center grid c with the smallest distance; if SD(g i ,c)≤Max_Radius, then the grid belongs to this cluster, add it to the cluster, and execute the next step; otherwise, the grid g i Add to C;

[0140] Step S5: Calculate the values ​​of all grids in the cluster to which the cluster center grid c belongs by the following formula, and take the grid c with the largest value new Add the mesh as the cluster center to C and delete c:

[0141]

[0142] Where w1 is the weight of the grid connectivity, w2 is the weight of the grid passenger flow, n is the number of grids in the cluster to which the cluster center grid c belongs, is the maximum value function;

[0143] Step S6: After all grids are traversed, output: candidate bus station set C = <c1,c2,...,c m >, m is the number of bus stops.

[0144] This algorithm has three parameters: Max_Radius = 500m, and weights w1 = 0.5 and w2 = 0.5. w1 and w2 do not affect the number of clusters; they are used to determine cluster centers. w1 is the grid connectivity weight, aiming to select candidate stations at locations with high passenger flow within each cluster; w2 is the grid passenger flow weight, aiming to select candidate stations at locations with high passenger flow within each cluster. Clearly, these two factors are equally important, meaning that while boarding and alighting behavior is important, the number of boarding and alighting events should also be considered. Therefore, w1 = w2 = 0.5 (CD and PDR are normalized). Considering that traditional Euclidean distance cannot represent the actual distance between two GPS track points on Earth, this application uses spatial distance for calculation, as shown in Formula 3-1. Here, R represents the average radius of the Earth, approximately 6378137 meters. Δlat and Δlon in Formula 3-2 represent the longitude and latitude differences between two GPS points, respectively. SD in Formula 3-1 can approximate the actual spatial distance between the two GPS points.

[0145] SD=Δσ*R (3-1)

[0146]

[0147] Where R represents the average radius of the Earth, Δlat and Δlon in Formula 3-2 represent the longitude and latitude differences between the two GPS points, and α1 and α2 represent the longitude angles of the two GPS points.

[0148] In the method, Max_Radius affects the number of clusters. According to the "China Urban Road Traffic Planning and Design Code", it is recommended that the service area of ​​buses be within a radius of 500 to 800. Therefore, in order to verify the impact of different values ​​of Max_Radius on the clustering results, this application conducted a clustering experiment with a step length of 20m in the interval [500, 800], and evaluated the corresponding results using four commonly used clustering indices: the sum of squares due to error (SSE), silhouette coefficient (SC), Calinski-Harabaz index (CHI), and Davidson-Bouldin index (DBI). The smaller the SSE and DBI index values, the better the clustering effect. The larger the CHI value, the better the clustering effect. The closer the SC value is to 1, the better the clustering effect.

[0149] like Figure 5 As shown in , the number of generated clusters decreases as Max_Radius increases, that is, the size of clusters in a certain area becomes larger and the number becomes smaller as the acceptable walking range increases. Figure 6 As shown in the figure, SSE increases with its value, and its value is the smallest when Max_Radius = 500, indicating the best clustering effect; CHI decreases with its value, and similarly, its value is the largest when it is 500, indicating the best clustering effect; SC and DBI fluctuate with the change of Max_Radius, with SC value showing a downward trend and DBI showing an upward trend, but both have the best clustering effect when their value is 500.

[0150] In summary, and considering that the walking range acceptable to passengers should be as small as possible, this application sets the value of Max_Radius to 500m.

[0151] The clustering results are as follows Figure 7 As shown in Figure 2, we finally get 2357 candidate bus stops, and their scatter plot is as follows: Figure 8As shown in (a), with these sites as the center and a radius of 500m, the types and numbers of POIs reachable from each candidate site are counted in combination with POI data.

[0152] In order to serve the practical application problem of generating candidate bus stops for a city, the present invention focuses on the following two issues: (1) a geographical location with large passenger flow should be selected, and (2) the entire area of ​​the city should be covered and the location should be as central as possible to the "service area". Traditional clustering methods such as K-Means are easily affected by data density during clustering, that is, multiple clusters will be generated at locations with high passenger flow. Based on the design principles of urban bus routes, setting up multiple redundant candidate bus stops within a certain range does not conform to the bus stop design principles, and the cluster center position generated by the algorithm cannot meet actual needs, that is, pedestrian-inaccessible areas, because the cluster center takes the average position of all sample data in the cluster. This application was verified by a comparative experiment with the K-Means method. The cluster center results obtained by the K-Means method are as follows: Figure 8 As shown in (b), K=2357, which is the same number of clusters as the method of this technical solution. The data input also includes the longitude, latitude, CD and PDR of the grid, and the results are quantitatively and qualitatively compared.

[0153] First, we compared the four indicators mentioned in the previous section for evaluating clustering results. As shown in Table 3, we can see that the method of this technical solution is superior to the K-Means method in terms of SSE, SC, and DBI indicators. The CH index, on the contrary, takes into account the intra-cluster distance and inter-cluster distance. The intra-cluster distance is represented by the distance between the sample point in the cluster and the cluster center point, and the inter-cluster distance is represented by the distance between the sample point and the center point of other clusters. The larger the CH value, the smaller the intra-cluster distance, the larger the inter-cluster distance, and the better the clustering effect. This is from Figure 8 (b) It can be seen that the K-Means clustering result is relatively dense, so the distance within the cluster is much smaller. However, from the practical application of this application, such a clustering result will lead to the generated candidate bus stops being too dense or even redundant.

[0154] Table 3 Comparison of indicators

[0155]

[0156] In addition, to view the results in a certain area of ​​the city, Figure 9 As shown in (a), it can be found that the cluster centers generated by the ICBS method are reasonably distributed in the urban area, while the stations generated by the K-Means method in the same area are closer and more in number, as shown in Figure 9(b) and position 3, such results are obviously redundant and do not conform to the bus route design principles. Similarly, when observing a certain area of ​​the city, the cluster center of each cluster obtained by K-Means is the average position of the samples in the cluster, so the results may fall into the area that is inaccessible to vehicles, such as Figure 9 The centers of the clusters are located in the river and buildings at positions 1 and 2, while the connectivity and passenger flow of the grid are considered in the clustering method of this application, and the center positions of the generated clusters are more reasonable.

[0157] In summary, compared with the K-Means method, the ICBS algorithm is more suitable for the practical application problems of this technical solution, and does not require the prior determination of the number of clusters, that is, the number of candidate bus stops. The method will combine the acceptable walking range of passengers, that is, the service range of the station, to achieve a reasonable layout of candidate bus stops in the urban area.

[0158] Cumulative passenger flow and travel time are important metrics for bus route planning, and the multi-objective bus route planning model constructed in this application is based on these two metrics. Therefore, after obtaining the aforementioned candidate bus stops, we use real-world travel data to calculate passenger flow transfers and travel times between stops, providing a basis for statistically analyzing relevant route information.

[0159] In a specific embodiment, the multi-objective model for bus route planning in step 4 includes: minimizing the route running time function and maximizing the route cumulative passenger flow function.

[0160] The function of minimizing line running time includes:

[0161] ΔT R =h+T(s r|R| ,s n ) (4-1)

[0162]

[0163] Where: ΔT R is the total running time of the bus between each two stations, h is the waiting time at each station, |R| represents the current number of stations in route R, s i represents the bus stop at station i, T is the vehicle travel time between two stations, which can be obtained from the Traveltime matrix, R=(r1,r2,...,r k ) represents the set of stations that have been added to the line and the destination that has not been reached. d , s n For the newly added bus stops, the running time of the entire route is the sum of the travel time between two adjacent bus stops and the waiting time of all stops on the entire route;

[0164] The function for maximizing the cumulative passenger flow of a route includes:

[0165]

[0166]

[0167] Where: ΔD R is the cumulative passenger flow of stations in the line, and F is the transfer passenger flow between two stations, which can be obtained from the Passenger matrix.

[0168] Furthermore, the function of minimizing route operating time and the function of maximizing route cumulative passenger flow also have the following constraints:

[0169] (1) Adequate site spacing;

[0170] dist(s i+1 ,s i )<δ,(i=1,2,...,n-1) (4-5)

[0171] Where: δ is the maximum distance between two sites, set δ = 1.5 km, dist is the distance between the two sites;

[0172] (2) The vehicle is always traveling in the direction of the destination;

[0173]

[0174]

[0175] Where: (x(i), y(i)) is the station s i The difference between the longitude and latitude of the departure point s1, that is, the direction vector, (x(n), y(n)) is the destination point s n The difference between the latitude and longitude of the starting point s1 is the direction vector. This constraint can ensure that the bus always moves in the direction of the destination. The constraint formula is shown in 4-6.

[0176] (3) Far from the starting point. When selecting sites, it is necessary to ensure that each new site is far away from the starting point s1. The constraint formula is shown in 4-7:

[0177] dist(s i+1 ,s1)>dist(s i ,s1),(i=1,2,...,n-1) (4-7)

[0178] (4) Close to the destination. When selecting a station, ensure that each new station is closer to the destination. n , as shown in formula 4-8:

[0179] dist(si+1 ,s n )<dist(s i ,s n ),(i=1,2,...,n-1) (4-8)

[0180] (5) No zigzag route. This condition is to ensure the smooth flow of the generated route and no "Z"-shaped route is generated. The constraint formula is shown in 4-9:

[0181]

[0182] Where: dist is the distance between the two sites, is the minimum function.

[0183] In a specific implementation, step 4 solves the model based on the NSGA-II algorithm to obtain the optimal bus route set. The specific method process is as follows:

[0184] Step S1: Input candidate bus station set S = {s1, s2, s3...}, where s i =(lon i ,lat i );

[0185] Passenger flow matrix Passenger, travel time matrix Traveltime;

[0186] Origin o and destinations d , s o ,s d ∈S;

[0187] Step S2: Initialize the total number of customized bus routes, that is, set the maximum population size N and the maximum number of iterations g max , set the number of iterations g = 1;

[0188] Step S3: Generate an initial population, that is, generate a set of feasible bus routes of size N;

[0189] Step S4: Calculate the objective function value of each line, formula 4-2 and formula 4-4;

[0190] Step S5: Perform a genetic crossover operation on the current population. For each line, randomly select a site to exchange with another individual and the constraints must be met at the same time, thereby generating a new line.

[0191] Step S6: Perform a genetic mutation operation on the current population. For each route, randomly select two sites in one of the routes, and replace the original route with a new route between the two sites while satisfying the constraints to generate a new route.

[0192] Step S7: Calculate the objective function value of the new route and merge the old and new routes. Perform fast non-dominated sorting on all individuals according to the two objective functions, calculate the individual crowding distance, and perform elite retention strategy operations to form the next generation route set, and g = g + 1;

[0193] Step S8: When g≤g max When g>g max When , the algorithm ends, and the line set at this time is P od .

[0194] Furthermore, the roulette probability selection method is adopted in step S3, and the steps are as follows:

[0195] It is used to express the probability of each site being selected, where S * The line contains s j All possible sites to connect to

[0196] allowSize=length(P) / / Number of candidate sites;

[0197] selectP=Random(0,1) / / pregenerate a pseudo-random number uniformly distributed between [0, 1];

[0198] selectNode=0 / / Initialize the number of selected sites to 0 (site numbers start from 0);

[0199] sumP=0 / / Initialize the cumulative probability to 0;

[0200] startIndex = Random(allowedSize) + 1 / / Randomly generate the starting position of the roulette wheel;

[0201] while sumP<selectP do;

[0202] sumP=sumP+P[(startIndex-1)%allowedSize] / / Calculate cumulative probability;

[0203] startIndex=startIndex+1;

[0204] end while;

[0205] selectNode=(startIndex-2)%allowedSize / / Select the site index to be added to the line;

[0206] Starting from the starting station, selectNode is selected according to the constraints, and a bus route is generated until the destination. The iteration is carried out until the scale is N.

[0207] To obtain a feasible solution to the route planning model and a feasible route with optimal overall benefits, we analyzed NSGA-II and set relevant parameters based on the previous section. This application sets the initial population size to 30, the maximum number of iterations to 500, and solves the problem according to step 3 of the algorithm. This experiment primarily verifies the applicability of the NSGA-II method's "fast non-dominated sorting" and "individual crowding distance calculation" concepts in solving this application's multi-objective model.

[0208] In a specific embodiment, the routes in the Pareto optimal route set generated in step 4 have multiple characteristic attributes for users to compare, such as the total number of stations, the types and number of POIs covered, etc. Therefore, sorting these routes is a multi-attribute sorting problem. In previous studies, we found that for multi-attribute sorting problems, quantifying attribute weights can help users better perform focused sorting, and based on the user's subjective sorting behavior, it can also understand their behavior and recommend routes that better meet their needs. Therefore, this application introduces the Ranking SVM method to achieve the quantification of attribute weights of bus routes and the ranking recommendation of routes in the optimal route set.

[0209] Ranking SVM is a machine learning method to solve the ranking problem. Its main idea is to transform the ranking problem into a binary classification problem. The standard SVM algorithm provides data items in an m-dimensional space and provides a set of corresponding labels to mark the category to which each point belongs. In the algorithm, it is usually represented as a set of tuples (d i ,y i ), and when dealing with binary classification problems, the label y i ∈{-1,1}. The output model is a hyperplane composed of The vector definition in , which cuts into the data space and is optimized so that y i = -1 is on one side of the hyperplane, and y i = 1 is on the other side, and the negative distance between the point and the hyperplane is large enough. Ranking SVM applies the idea of ​​optimizing the hyperplane to the sorting problem with pairwise constraints. In addition to obtaining a complete set of data points with corresponding labels, a limited pair of data points d is also obtained. i and d j , and a label indicating which of the two data points is better. Figure 10As shown, a1, a2, and a3 are attributes of data items d1, d2, and d3 respectively. The training samples are defined so that a1-a2, a1-a3, and a2-a3 are positive samples, and a2-a1, a3-a1, and a3-a2 are negative samples. Then a binary classifier (support vector machine) is trained to classify these training samples.

[0210] (1) Derivation constraints

[0211] The input data of the Ranking SVM model is the difference vector of paired data items. Suppose there are k data items {d1, d2, ..., d k}, where each data item has m attributes {a1, a2, ..., a m}; First, a set of all pairwise difference vector combinations is created as a model training instance and put into the Ranking SVM for training to obtain the weight results of each attribute. For each i, j∈{1,...,k}, where i≠j. The training tuple is derived according to formula 4-12, where r(d i ) represents the data item d i The ranking score is calculated as shown in 4-13, w j Indicates the weight corresponding to the data item. That is, if d i Ranked higher than d j , then the classification is θ=1, otherwise θ=-1, where Now the sorting problem is transformed into a binary classification problem, and Ranking SVM can be applied.

[0212]

[0213]

[0214] (2) Sorting

[0215] The ranking result is obtained by calculating the difference between two data items on all attributes and forming a difference vector, which is then input into the Ranking SVM classifier. The difference vector of data points i and j is represented as (d i -d j Based on the above input, the Ranking SVM model will predict which of the two data items has a higher ranking. The specific calculation process is as follows. First, the model calculates the difference vector (d i -d j ) and the dot product of the model's internal weight w. If the dot product w·(d i -d j ) is positive, then the vector difference belongs to the positive class y=1, so d i The ranking of d jBefore. If the dot product is negative, the vector difference belongs to the negative class y = -1, then d j Relatively higher than d in the ranking i By inputting the difference vector constructed by any two data items into the trained Ranking SVM model, we can finally get the ranking results for all data items.

[0216] Similarly, the weight results obtained by the Ranking SVM model can be combined with formula 4-13 to rank the scores of each data item, and the sorting operation can be achieved by comparing the scores. This calculation method is more flexible and can speed up the efficiency of the sorting calculation. The final result is the same as the result of constructing the difference vector, calculating the dot product, and using the Ranking SVM model to predict. The derivation process is as follows: w×(d i -d j )=w×d i -w×d j If w×d i >w×d j , then w×(d i -d j )>0, so d i Ranked better than d j , on the contrary d j Ranked better than d i .

[0217] (3) Contribution value calculation

[0218] The contribution value of a data item helps users understand how much of the ranking score of the data item is affected by a given attribute. The calculation formula is shown in 4-14, indicating that attribute a j For data item d i Normalized contribution to the ranking score. Where l is the attribute index with the largest attribute score for the data item.

[0219]

[0220] The above formula shows that the contribution value ranges from [0, 1]. Higher values ​​indicate a greater contribution to the ranking score of the data point (14 points). When the attribute weight is large, the attribute value contributes more to the ranking score of the data point. When the attribute weight is small, the attribute value contributes less to the ranking score of the data point. When both the attribute weight and the data value are large, the contribution is the largest.

[0221] Quantitative ranking of line attribute weights and contribution value calculation

[0222] This embodiment mainly introduces the process of implementing attribute weight quantization sorting and contribution value calculation for the Pareto optimal path set generated in the previous section.

[0223] Based on the above-mentioned Ranking SVM principle, this application implements weight quantification around the five attributes of the generated bus routes, namely: cumulative passenger flow of the route (attr1), route operating time (arrt2), route length (attr3), number of stations (attr4) and number of accessible POIs (attr5). Among them, attribute 1 and attribute 2 are important attributes that need to be considered in previous bus route planning studies, attributes 3 and 4 can reflect the input cost of route operation, and attribute 5 measures the functionality of the accessible area of ​​the route from the perspective of accessibility. These attributes can all reflect the planners' considerations on whether the route meets the needs of passengers. Therefore, this application takes them as the five main attributes to consider and conducts subsequent experiments.

[0224] First, while obtaining the Pareto line set, this application counts the five attribute values ​​of each line R, such as Figure 11 As shown in a, the training tuples are constructed by initializing and sorting according to the important attribute of cumulative passenger flow, as shown in Figure 8 As shown in (b), the attribute weight value can be obtained through Ranking SVM training.

[0225] At the same time, this application provides an interface for changing attribute weights. After the user obtains the attribute weight quantization result, if he wants to increase or decrease the attribute weight, he will add a new training tuple (x, y) to the training instance, such as Figure 11 As shown in c. This vector is similar to the difference vector above, except that the position indicating the attribute the user wishes to change is marked with a 1, and all other positions are 0. The same applies to the labels y. For attributes whose weights the user wishes to increase, a training tuple (-x, y = -1) is constructed. For attributes whose weights the user wishes to decrease, a training tuple (-x, y = 1) is constructed. These are then sorted and the contribution value calculated using Formulas 4-13 and 4-14.

[0226] The Pareto-optimal route set was then quantified and ranked using a Ranking SVM, with contribution values ​​calculated. First, the weights of the five attributes for each of the 30 bus routes in the set were calculated. Initial ranking was performed based on cumulative passenger flow, and model training was performed to quantify the attribute weights.

[0227] The experimentally obtained weights for various route attributes are shown in the initial weights in Table 4. "Accumulated passenger flow" (attr1) has the highest weight, indicating that the Ranking SVM accurately predicted the initial ranking criteria, resulting in the maximum weight for this attribute. To verify the interface for increasing and decreasing attribute weights, we input constructed training tuples into the model to decrease the weight of accumulated passenger flow and increase the weights of "travel time" (attr2) and "route length" (attr3). This instructs the model to focus more on travel time and route length and less on accumulated passenger flow.

[0228] Table 4 Line attribute weight quantification results

[0229]

[0230] The weights for each attribute obtained by retraining the model are shown in Table 4. The weight changes indicate that the weight of attr1 decreases, while the weights of attr2 and attr3 become negative. The weights of the remaining two attributes also change accordingly. For bus routes, all other conditions being equal, shorter travel times and route lengths yield the greatest benefits. Therefore, attr2 and attr3 are negatively correlated with route ranking, indicating that the experimental results meet practical requirements and verifying the accuracy of increasing and decreasing weights.

[0231] As described above, the above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the above embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the above embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A bus route planning and influencing factor visual analysis method based on GPS data, characterized by: The method comprises the following steps: Step 1: Collect city taxi order data and city point of interest data; Step 2: Preprocess the collected data to obtain passenger travel origin-destination OD data and POI category data; Step 3: Based on the data obtained in step 2, extract the city's hotspot boarding and alighting grids and cluster them. Determine the candidate bus stop locations based on the clustering results, and calculate the passenger flow transfer volume and vehicle travel time between candidate stops. Step 4: Construct a multi-objective model for bus route planning, solve the model based on the NSGA-II algorithm, and obtain the optimal bus route set; Step 5: Use Ranking SVM to quantify the attribute weights, sort, and calculate the contribution values ​​of the routes in the optimal bus route set to correct the output results of the bus route planning multi-objective model; Step 6: Perform visual analysis on the routes in the optimal bus route set to obtain the optimal bus route planning solution; The step of extracting the boarding and alighting grids in the city hotspot based on the data obtained in step 2 includes: Step S1: Input the passenger boarding and alighting location point set P = <p1,p2,...,p n >, p i =(lon i ,lat i ), p i Indicates the location of boarding and alighting, n indicates the total number of boarding and alighting, lon i with lat i Indicates the latitude and longitude of the corresponding point; Step S2: Initialize the grid set Step size step = 0.0001, reference point longitude and latitude Lon, Lat, so that lon i ≤Lon, lat i ≥Lat all hold true; Step S3: Traverse the points p in the set P i , calculate the longitude and latitude of the grid to which it belongs, lon g ,lat g : lon g =floor((lon i -Lon) / step)*step+Lon lat g =floor((lat i -Lat) / step)*step+Lat In the formula, floor represents the rounding function, that is, the grid to which it belongs is calculated according to the step size; Step S4: Determine the grid g i =(lon i ,lat i ,CD i ,PDR i )∈G holds true: If not, g i Add to the set G and initialize the grid connectivity Number of boarding and alighting within the grid otherwise, And execute the next step; Step S5: Calculate (lon g ±step,lat g ±step)、(lon g ±step,lat g )、(lon g ,lat g ±step), determine whether they are in the set G, and get the number w of existence, such that Step S6: After all boarding and alighting points are traversed, the boarding and alighting hotspot grid set G is output.

2. The bus route planning and influencing factor visual analysis method based on GPS data according to claim 1 is characterized in that: In step 2: data preprocessing includes at least one of data desensitization, filtering duplicate data, removing missing data in major fields, removing invalid location information data, or coordinate system conversion and timestamp calculation.

3. The bus route planning and influencing factor visual analysis method based on GPS data according to claim 1 is characterized in that: The grid clustering method in step 3 includes: Step S1: Input: G = <g1,g2,...,g n >, n is the number of grids, where each grid g i =(log i ,lat i ,CD i ,PDR i ); Step S2: Set a walkable distance threshold Max_Raduis, which is 500m by default, and the cluster center set Step S3: Add g1 as the cluster center to the set C; Step S4: Traverse all grids in G, for grid g i , calculate the distance SD(g i ,c j ), find the cluster center grid c with the smallest distance; if SD(g i ,c)≤Max_Radius, then the grid belongs to this cluster, add it to the cluster, and execute the next step; otherwise, the grid g i Add to C; Step S5: Calculate the values ​​of all grids in the cluster to which the cluster center grid c belongs by the following formula, and take the grid c with the largest value new Add the mesh as the cluster center to C and delete c: Where w1 is the weight of the grid connectivity, w2 is the weight of the grid passenger flow, z is the number of grids in the cluster to which the cluster center grid c belongs, is the maximum function; Step S6: After all grids are traversed, the candidate bus station set C is output. <c1,c2,...,c m >, m is the number of bus stops.

4. The bus route planning and influencing factor visual analysis method based on GPS data according to claim 3 is characterized in that: The calculation method of the real space distance SD between GPS track points is: SD=Δσ*r (3-1) Where r represents the average radius of the Earth, Δlat and Δlon in Formula 3-2 represent the longitude and latitude differences between the two GPS points, and α1 and α2 represent the longitude angles of the two GPS points.

5. The bus route planning and influencing factor visual analysis method based on GPS data according to claim 1 is characterized in that: The multi-objective model for bus route planning in step 4 includes: minimizing the route running time function and maximizing the route cumulative passenger flow function. The function of minimizing line running time includes: ΔT R =h+T(s r|R| ,s n ) (4-1) Where: ΔT R is the total running time of the bus between each two stations, h is the waiting time at each station, |R| represents the current number of stations in route R, s i represents the bus stop at station i, T is the vehicle travel time between two stations, which can be obtained from the Traveltime matrix, R=(r1,r2,...,r k ) represents the set of stations that have been added to the line and the destination that has not been reached. d , s n For the newly added bus stops, the running time of the entire route is the sum of the travel time between two adjacent bus stops and the waiting time of all stops on the entire route; The function for maximizing the cumulative passenger flow of a route includes: Where: ΔD R is the cumulative passenger flow of all stations in the line, and F is the transfer passenger flow between two stations, which can be obtained from the Passenger matrix.

6. The bus route planning and influencing factor visual analysis method based on GPS data according to claim 5 is characterized in that: The function of minimizing route running time and the function of maximizing route cumulative passenger flow also have the following constraints: (1) Adequate site spacing; dist(s i+1 ,s i )<δ,(i=1,2,...,n-1) (4-5) Where: δ is the maximum distance between two stations, set δ = 1.5 km, dist is the distance between any two stations; (2) The vehicle is always traveling in the direction of the destination; Where: (x(i), y(i)) is the station s i The difference between the longitude and latitude of the origin s1, that is, the direction vector, (x(n), y(n)) is the terminal station s n The difference between the latitude and longitude of the origin s1 is the direction vector. This constraint ensures that the bus always moves in the direction of the destination. The constraint formula is shown in 4-6. (3) Far from the starting point. When selecting sites, it is necessary to ensure that each new site is far away from the starting point s1. The constraint formula is shown in 4-7: dist(s i+1 ,s1)>dist(s i ,s1),(i=1,2,...,n-1) (4-7) Where: dist is the distance between two sites; (4) Close to the destination. When selecting a station, ensure that each new station is closer to the destination. n , as shown in formula 4-8: dist(s i+1 ,s n )<dist(s i ,s n ),(i=1,2,...,n-1) (4-8) Where: dist is the distance between two sites; (5) No zigzag route. This condition is to ensure the smooth flow of the generated route and no "Z"-shaped route is generated. The constraint formula is shown in 4-9: Where: dist is the distance between the two sites, is the minimum function.

7. The bus route planning and influencing factor visual analysis method based on GPS data according to claim 6 is characterized in that: Step 4 solves the model based on the NSGA-II algorithm to obtain the optimal bus route set. The specific method process is as follows: Step S1: Input candidate bus station set S = {s1, s2, s3...}, where s i =(lon i ,lat i ); Passenger flow matrix Passenger, travel time matrix Traveltime; Origin o and destinations d , s o ,s d ∈S; Step S2: Initialize the total number of customized bus routes, that is, set the maximum population size N and the maximum number of iterations g max , set the number of iterations g = 1; Step S3: Generate an initial population, that is, generate a set of feasible bus routes of size N; Step S4: Calculate the objective function value of each line, formula 4-2 and formula 4-4; Step S5: Perform a genetic crossover operation on the current population. For each line, randomly select a site to exchange with another individual and the constraints must be met at the same time, thereby generating a new line. Step S6: Perform a genetic mutation operation on the current population. For each route, randomly select two sites in one of the routes, and replace the original route with a new route between the two sites while satisfying the constraints to generate a new route. Step S7: Calculate the objective function value of the new route and merge the old and new routes. Perform fast non-dominated sorting on all individuals according to the two objective functions, calculate the individual crowding distance, and perform elite retention strategy operations to form the next generation route set, and g = g + 1; Step S8: When g≤g max When g>g max When , the algorithm ends, and the line set at this time is P od .

8. The bus route planning and influencing factor visual analysis method based on GPS data according to claim 7 is characterized in that: In step S3, a roulette wheel probability selection method is used, and the steps are as follows: It is used to express the probability of each site being selected, where S * The line contains s j All possible sites to connect to allowSize=length(P) / / Number of candidate sites; selectP=Random(0,1) / / pregenerate a pseudo-random number uniformly distributed between [0, 1]; selectNode=0 / / Initialize the number of selected sites to 0 (site numbers start from 0); sumP=0 / / Initialize the cumulative probability to 0; startIndex = Random(allowedSize) + 1 / / Randomly generate the starting position of the roulette wheel; while sumP<selectP do: sumP=sumP+P[(startIndex-1)%allowedSize] / / Calculate cumulative probability; startIndex=startIndex+1; end while; selectNode=(startIndex-2)%allowedSize / / Select the site index to be added to the line; Starting from the starting station, selectNode is selected according to the constraints, and a bus route is generated until the destination. The iteration is carried out until the scale is N.

9. The bus route planning and influencing factor visual analysis method based on GPS data according to claim 1 is characterized in that: The weights of the five attributes of the optimal bus route set are quantified, namely, the cumulative passenger flow, route operation time, route length, number of stations, and number of accessible POIs. The cumulative passenger flow and route operation time are important attributes in bus route planning research. The route length and number of stations can reflect the input cost of route operation. The number of accessible POIs measures the functionality of the route's accessible area from the perspective of accessibility.

Images