Two-stage car-sharing matching method and system for transferring taxis in large transportation hub

Abstract

The invention discloses a two-stage car-sharing matching method and system for taxi connection in a large transportation junction, and the method comprises the steps: taking the shortest total driving mileage of vehicles and the minimum number of service vehicles as optimization targets, considering the characteristics of passengers in the junction, and taking parameters including large luggage, departure time and arrival time as constraints; constructing a hub taxi sharing model; passenger requests are obtained, the passenger requests are clustered through an adaptive hierarchical clustering algorithm with the path vector similarity of the fusion direction and the spatial-temporal characteristics as the reference, and a passenger request cluster is obtained; and based on the hub taxi carpooling model and the passenger request cluster, solving a carpooling path planning problem by using an improved adaptive genetic algorithm to obtain a passenger matching scheme and a carpooling driving path. According to the two-stage ride-sharing matching method provided by the invention, similar passenger requests can be effectively quantified and clustered, the calculation efficiency can be improved by 85%, and meanwhile, the waiting time and the detour distance of passengers can be shortened.

Description

Technical Field

[0001] The present invention relates to the field of transportation technology, and in particular to a two-stage ride-sharing matching method and system for connecting taxis at a large transportation hub. Background Art

[0002] In recent years, the rapid rise of my country's high-speed railways has brought about many problems in the connection between urban and rural transportation. Among them, the long waiting time for passengers arriving at hubs has become a hot issue. As the key node of the entire transportation network, the rapid evacuation of hub-arrived passengers is of great value in strengthening integrated connection and improving the travel experience of passengers. Taxis or online ride-hailing cars can often only carry 1 to 2 passengers. During peak hours, it is easy to cause passengers to wait for taxis for a long time and waste resources. Taxi sharing can effectively improve the efficiency of taxi use and is a mode of transportation that takes into account both transportation capacity and passenger waiting time. In addition, the time and place of hub passengers are highly concentrated, which greatly improves the feasibility of sharing. Therefore, under the condition of concentrated passenger flow at hubs, studying hub sharing has a strong practical value for achieving rapid relief of hub passengers.

[0003] Existing taxi-pooling matching methods mainly focus on solving the problems of road network-pooling and taxi-pooling matching for connecting to public transportation systems such as subways and buses. They mostly focus on the global optimization perspective, taking into account factors such as the boarding location and time to obtain passenger matching solutions and shared ride routes. However, passengers at large transportation hubs have the characteristic of concentrated arrival at stations, which requires rapid and efficient matching of large-scale requests concentrated at the boarding point (point O), and the optimization complexity is higher.

[0004] The technical problem to be solved by the present invention is how to consider the factor of passengers' large luggage and construct a hub taxi sharing model; based on the high optimization complexity of the hub sharing problem, a two-stage sharing matching method is designed to improve the passenger matching efficiency and thus evacuate the high-intensity arriving passengers at the hub more quickly. Summary of the invention

[0005] The technical problem to be solved by the present invention is to provide a two-stage ride-sharing matching method and system for connecting taxis at a large transportation hub in view of the defects in the prior art.

[0006] The technical solution adopted by the present invention to solve its technical problem is:

[0007] The present invention provides a two-stage ride-sharing matching method for connecting taxis at a large transportation hub, the method comprising the following steps:

[0008] Step 1: Taking the shortest total vehicle mileage and the minimum number of service vehicles as the optimization objectives, and considering the characteristics of hub passengers, and taking parameters including large luggage, departure and arrival time as constraints, a hub taxi sharing model is constructed;

[0009] Step 2: Obtain passenger requests, and cluster passenger requests using an adaptive hierarchical clustering algorithm based on the path vector similarity of the fusion direction and spatiotemporal features to obtain passenger request clusters;

[0010] Step 3: Based on the hub taxi carpooling model and passenger request clusters, the improved adaptive genetic algorithm is used to solve the carpooling path planning problem and obtain the passenger matching solution and carpooling driving path.

[0011] Furthermore, the method of step 1 of the present invention comprises:

[0012] Step 11: Establish model assumptions for the hub taxi-pooling model, including:

[0013] The taxi arrives at the waiting area before the passengers and leaves the hub after all passengers have arrived;

[0014] Passengers can arrive at the designated location before the earliest scheduled departure time;

[0015] Each taxi has the same driving speed and travels at a constant speed throughout the journey;

[0016] The road network is smooth, and the impact of random disturbances including traffic congestion is not considered;

[0017] Step 12: Set the objective function and constraints of the hub taxi sharing model, including:

[0018] The objective function is set to minimize the overall consumption of the system, that is, the total mileage of all vehicles and the number of required service vehicles are minimized; ensure that each passenger can be served by only one taxi; ensure that all service taxis depart from the hub and cannot return to the hub during the service process; ensure flow balance; capacity constraints, that is, the number of passengers in each request and the total amount of luggage carried cannot exceed the carrying capacity of the car; set time window constraints; the latest arrival time of passengers in the same car is the departure time of the taxi, which must meet the departure time window of each passenger; ensure that the arrival time of each passenger must meet the arrival time window; decision variables use 0-1 integer constraints.

[0019] Step 13: Build a hub taxi sharing model based on model assumptions, objective functions and constraints.

[0020] Furthermore, the hub taxi sharing model constructed in step 13 of the present invention is specifically:

[0021]

[0022]

[0023] In this case, a directed network G = (N, E) is introduced, where N is the set of points, E is the set of edges, i, j, l are the indexes of the passenger's requested destination, i, j, l∈N, 0 represents the hub, N0 represents the set of all points including the hub (0), k is the taxi index, k∈K; the case where the taxi returns to the hub is not considered, d ij is the actual distance from i to j; c f is the fixed usage fee for each taxi; c d is the unit distance operating cost of each taxi; Q represents the total capacity of taxis; p i is the number of passengers requesting i; b i is the number of large luggage requested by passenger i; B is the number of luggage that can be placed in the trunk; W is the number of luggage that can be placed on a seat; v is the vehicle speed; s is the time for each passenger to get off the bus, i.e., the service time; [e i ,l i ] and [ee i ,ll i ] represent the departure time window and arrival time window of passenger i respectively; x ijk is a decision variable that equals 1 if passenger i is served by taxi k from point j to point l, and 0 otherwise.

[0024] Further, the method of step 2 of the present invention comprises:

[0025] Step 21: Based on the fact that the starting points of hub passengers are the same, the direction of the passenger route plays a decisive role in the driving path, and the direction similarity is calculated;

[0026] Step 22: Calculate the spatiotemporal distance similarity based on the impact of the detour degree between the passenger's destinations on the passenger's ride-sharing experience;

[0027] Step 23: According to the influence of the driving path and the ride time on the passenger matching, the direction similarity and the time-space distance similarity are combined, and the request time is integrated to calculate the path vector similarity;

[0028] Step 24: Calculate the path vector similarity distance of the two path vectors according to the path vector similarity of the two path vectors;

[0029] Step 25: Propose a method for automatically determining the optimal clustering parameters based on the clustering feature ratio, and automatically determine the number of clusters k by analyzing the change characteristics of the path similarity distance in the hierarchical clustering process;

[0030] Step 26: Clustering to obtain passenger request clusters: Based on the path vector similarity distance and the automatically determined k value, the large-scale passenger requests are divided into k request clusters.

[0031] Furthermore, the method for calculating the direction similarity in step 21 of the present invention is:

[0032] Calculate the cosine value cos<λ1,λ2> of the intersection angle between the vectors λ1 and λ2 obtained from the starting and ending points of the two paths. If μ≤cos<λ1,λ2>≤1, the two vectors are considered to be similar in direction, which can be expressed as:

[0033]

[0034] Among them, VD(R1, R2) represents the directional similarity of two paths, and M is a constant greater than a certain threshold;

[0035] The method for calculating the spatiotemporal distance similarity in step 22 is:

[0036] For any two passenger requests, with time attributes t1, t2 and location attributes x1, x2, the spatiotemporal distance similarity measure of the two paths R1 and R2 is obtained:

[0037]

[0038] Among them, VS(x1,x2) is the degree of detour between the two nodes, d(x1,x2) is the actual distance between the two nodes, d(0,x1) and d(0,x2) refer to the actual distances from hub 0 to the requested end points x1 and x2, respectively, VT(t1,t2) represents the degree of overlap of the passenger request time windows, and α and β are normalization factors that balance the importance of space-time distances.

[0039] Furthermore, the method for calculating the path vector similarity in step 23 of the present invention is:

[0040] Adding direction similarity as an influencing factor to the spatiotemporal distance similarity index, we get the expression of the path vector similarity measurement method:

[0041] VDTS(R1,R2)=VD(R1,R2)□VTS(R1,R2)

[0042] Among them, VDTS(R1, R2) represents the path vector similarity of paths R1 and R2, VD(R1, R2) represents the direction similarity of the two path vectors, and VTS(R1, R2) represents the space-time distance similarity of the two paths;

[0043] The method for calculating the path vector similarity distance in step 24 is:

[0044] The path vector similarity distance is defined as follows:

[0045] DVDTS(r1,r2)=N-VDTS(R1,R2)

[0046] Among them, DVDTS(R1, R2) represents the path vector similarity distance of requests r1 and r2, N is a constant, and the driving paths corresponding to requests r1 and r2 are R1 and R2 respectively.

[0047] Furthermore, in step 25 of the present invention, the method for automatically determining the number of clusters k by analyzing the change characteristics of the path similarity distance in the hierarchical clustering process includes:

[0048] Define the number of clusters after each merger as s, D s is the merge distance corresponding to the number of clusters s, R s It indicates the growth rate of the merged distance when the number of clusters is s; when the growth rate of the merged distance is less than a certain threshold, it means that the merged distance grows slowly, indicating that the hierarchical structure tends to be stable. Therefore, the final number of clusters k satisfies the stable ratio. The maximum number of clusters s is:

[0049]

[0050]

[0051] Among them, the stable ratio It is the last inflection point of the ratio curve with significant growth.

[0052] Further, the method of step 3 of the present invention comprises:

[0053] Step 31: For each clustered passenger request cluster, the passenger sharing plan of a car is used as a chromosome, the chromosome is encoded and an initial sharing plan that meets the constraints of the hub taxi sharing model is generated as the initial population of the genetic algorithm;

[0054] Step 32: Taking the optimization objective of the hub taxi sharing model as the fitness function, calculating the fitness of the individual, and calculating the three-quarter median, one-quarter median, maximum value and minimum value of the fitness of the population;

[0055] Step 33: Adaptively determine the crossover probability and mutation probability;

[0056] Step 34: Determine whether the population meets the convergence condition. If so, output the passenger matching solution and the shared ride path. If not, continue to execute downward.

[0057] Step 35: Determine whether the population concentration condition is met. If so, it means that the fitness of 50% of the individuals in the middle of the population is close to the fitness of nearly 100% of the individuals between the maximum and minimum values ​​of the population, indicating that the population is more concentrated. Therefore, the mutation operation is performed first to expand the diversity of the population, and then the crossover operation is performed to produce excellent individuals. If not, it means that the fitness of 50% of the individuals in the middle of the population is not close to the fitness of nearly 100% of the individuals between the maximum and minimum values ​​of the population, indicating that the population is more dispersed. Therefore, the crossover operation is performed first and then the mutation operation is performed to ensure the diversity of the population.

[0058] Step 36: Optimal preservation strategy; calculate the optimal fitness of the offspring individuals after crossover and mutation operations and compare them with the individuals with the best fitness in the parent generation. If the best individual in the parent generation is better than the offspring, remove one individual in the offspring and put the best individual in the parent generation into the offspring;

[0059] Step 37: Return to step 32.

[0060] Furthermore, the method for adaptively determining the crossover probability and the mutation probability in step 33 of the present invention is:

[0061] Crossover probability p c and mutation probability p m The adaptive calculation formula is:

[0062]

[0063] Among them, p c and p m Represent the crossover probability and mutation probability respectively, a is arcsin(f3-f1 / f max -f min ), which indicates the closeness of the fitness of the middle 50% of the individuals in the population to the fitness of nearly 100% of the individuals between the maximum and minimum values ​​of the population, and k1, k2, k3, k4 are constants;

[0064] The condition for judging the concentration of the population in step 35 is: arcsin(f3-f1) / (f max -f min )≥π / 6.

[0065] The present invention provides a two-stage ride-sharing matching system for connecting taxis at a large transportation hub, comprising:

[0066] a memory for storing executable computer programs;

[0067] The processor is used to implement the above-mentioned two-stage ride-sharing matching method for connecting taxis at large transportation hubs when executing the executable computer program stored in the memory.

[0068] The beneficial effects produced by the present invention are:

[0069] 1. This paper constructs a hub taxi carpooling model considering passengers' large luggage. Experiments show that the impact of large luggage must be considered in hub scenarios to ensure the feasibility of carpooling routes. In view of the high complexity of hub carpooling problems, a two-stage algorithm of PVS-AHC-IGA is designed to solve it. PVS represents path vector similarity, AHC represents adaptive hierarchical clustering algorithm, and IGA represents improved genetic algorithm.

[0070] The present invention proves through experiments that the two-stage carpooling matching method proposed in the present invention can effectively quantify and cluster similar passenger requests, improve the calculation efficiency by 85%, and reduce the passenger waiting time and detour distance.

[0071] 2. In terms of model construction, the present invention takes the shortest total vehicle mileage and the minimum number of service vehicles as optimization goals, while considering the characteristics of hub passengers, and taking large luggage, departure and arrival time as constraints to construct a hub taxi sharing model. This model comprehensively considers the characteristics of large transportation hubs and solves the problem of lack of corresponding models in the prior art.

[0072] 3. The present invention proposes a path vector similarity measurement method that takes into account direction and spatiotemporal features, integrates request time, and classifies passengers with similar paths and short time spans into the same cluster. This method greatly improves the accuracy of classifying similar passengers into the same cluster. In addition, the present invention proposes a method for automatically determining the optimal clustering parameters based on the clustering feature ratio to reduce the influence of the subjectivity of the clustering results.

[0073] 4. Traditional genetic algorithms achieve generation-by-generation evolution of a population through gene selection-crossover-mutation operations, where the crossover operator obtains excellent individuals through gene recombination, and the mutation operator can ensure the diversity of the population. However, this operation of first crossover and then mutation can easily lead to the population fitness becoming more and more concentrated in the later stage, and the mutation operator does not change the individuals in the population much, resulting in local convergence. To this end, the present invention proposes an adaptive algorithm flow, which uses an improved adaptive genetic algorithm to solve the carpooling path planning problem, obtains the passenger matching plan and the carpooling driving path, and prevents the algorithm from falling into the local optimum. BRIEF DESCRIPTION OF THE DRAWINGS

[0074] The present invention will be further described below with reference to the accompanying drawings and embodiments, in which:

[0075] Figure 1 A schematic diagram of the PVS-AHC-IGA algorithm flow provided by the present invention;

[0076] Figure 2 A schematic diagram of a path vector provided by the present invention;

[0077] Figure 3 A schematic diagram of the distribution of passenger request destinations is provided for example. DETAILED DESCRIPTION

[0078] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0079] Example 1

[0080] Aiming at the shortcomings of the existing technology, the present invention constructs a hub taxi carpooling model taking into account passengers' large luggage, takes minimizing the total vehicle mileage and the number of service vehicles as the optimization goal, and proposes a two-stage carpooling matching method based on PVS-AHC-IGA to solve the problem of low matching efficiency of the existing technology under large passenger flow conditions.

[0081] The embodiment of the present invention proposes a two-stage carpooling matching method for connecting taxis at large transportation hubs. The method mainly includes: constructing a carpooling model for hub taxis taking into account passengers' large luggage, clustering passenger requests using an adaptive hierarchical clustering algorithm based on the path vector similarity of fusion direction and spatiotemporal features, and obtaining a passenger matching plan and a carpooling driving path using an improved adaptive genetic algorithm. The process of the present invention is shown in the attached figure. Figure 1 shown.

[0082] Step 1: Taking the shortest total vehicle mileage and the minimum number of service vehicles as the optimization objectives, while considering the characteristics of hub passengers and taking large luggage, departure and arrival times as constraints, a hub taxi sharing model is constructed.

[0083] Step 11: Model assumptions, specifically:

[0084] 1) The taxi arrives at the waiting area before the passengers and leaves the hub after all passengers have arrived;

[0085] 2) All passengers can arrive at the designated location before the earliest scheduled departure time;

[0086] 3) Each taxi has the same driving speed and travels at a constant speed throughout the journey;

[0087] 4) The road network is unobstructed, and the impact of random disturbances such as traffic congestion is not considered.

[0088] Step 12: Set the objective function and constraints of the hub taxi sharing model, including:

[0089] Objective function (1) represents the minimum overall consumption of the system, that is, the total mileage of all vehicles and the number of service vehicles required. Constraints (2) and (3) ensure that each passenger can only be served by one taxi. Constraints (4) and (5) ensure that all service taxis depart from the hub and cannot return to the hub during the service process. Constraint (6) ensures traffic balance. Constraint (7) is a capacity constraint, that is, the number of passengers and the total amount of luggage carried by each request cannot exceed the carrying capacity of the car. Constraints (8) and (9) are time window constraints. Formula (8) stipulates that the latest arrival time of passengers riding in the same car is the departure time of the taxi, which must meet the departure time window of each passenger; Formula (9) ensures that the arrival time of each passenger must meet the arrival time window. Formula (10) is a 0-1 integer constraint on decision variables. The specific formula is shown below.

[0090] Step 13: Build a hub taxi sharing model.

[0091]

[0092]

[0093] In this case, a directed network G = (N, E) is introduced, where N is the set of points, E is the set of edges, i, j, l are the indexes of the destinations requested by passengers, i, j, l∈N, 0 represents the hub, N0 represents the set of all points including the hub (0), k is the taxi index, k∈K. The taxi returning to the hub is not considered, d ij is the actual distance from i to j; c f is the fixed usage fee for each taxi; c d is the unit distance operating cost of each taxi; Q represents the total capacity of taxis; p i is the number of passengers requesting i; b i is the number of large luggage requested by passenger i; B is the number of luggage that can be placed in the trunk; W is the number of luggage that can be placed on a seat; v is the vehicle speed; s is the time (service time) for each passenger to get off the bus; [e i ,l i ] and [ee i ,ll i ] represent the departure time window and arrival time window of passenger i respectively; x ijk is a decision variable that equals 1 if passenger i is served by taxi k from point j to point l, and 0 otherwise.

[0094] Step 2: Based on the path vector similarity integrating direction and spatiotemporal features, the passenger requests are clustered using an adaptive hierarchical clustering algorithm.

[0095] Path vector similarity: driving path and ride time are key factors affecting passenger matching. In the hub carpooling problem, the similarity of driving paths needs to consider two points: a. The starting points of hub passengers are the same, and the direction of the passenger route plays a decisive role in the driving path; b. Under the premise of similar driving directions, the distance between the passenger endpoints has little effect on the near-to-far driving mode of the present invention, but the degree of detour between the passenger endpoints will directly affect the passengers' carpooling experience. Accordingly, the present invention proposes a path vector similarity measurement method that takes into account direction and spatiotemporal characteristics, and integrates request time, aiming to classify passengers with similar paths and small time spans into the same cluster.

[0096] In the non-pooling state, requesting r1 and r2 can obtain two paths R1 and R2 starting from the hub. Take the starting points q1 and q2 and the end points z1 and z2 of the two paths, connect the starting and end points, and obtain vectors λ1 and λ2, as shown in Figure 2 shown.

[0097] Step 21: Calculate directional similarity.

[0098] Calculate the cosine value cos<λ1,λ2> of the intersection angle between the vectors λ1 and λ2 obtained from the starting and ending points of the two paths. If μ≤cos<λ1,λ2>≤1, the two vectors are considered to be similar in direction, which can be expressed as:

[0099]

[0100] Among them, VD(R1, R2) represents the directional similarity of the two paths, and M is a large constant.

[0101] Step 22: Calculate the spatiotemporal distance similarity.

[0102] The time attribute refers to the time window from the hub, which stipulates that passengers with overlapping departure time windows have matching opportunities. The longer the overlapping time, the higher the chance of successful matching. The location attribute refers to the degree of detour between the requested destinations. For any two passenger requests, with time attributes t1, t2 and location attributes x1, x2, the time-space distance similarity measurement method of the two paths R1 and R2 is obtained, as shown in formula (14):

[0103]

[0104] Among them, VS(x1,x2) is the degree of detour between the two nodes, d(x1,x2) is the actual distance between the two nodes, d(0,x1) and d(0,x2) refer to the actual distances from hub 0 to the requested end points x1 and x2, respectively, VT(t1,t2) represents the degree of overlap of the passenger request time windows, and α and β are normalization factors that balance the importance of space-time distances.

[0105] Step 23: Calculate path vector similarity.

[0106] By adding direction similarity as an influencing factor to the spatiotemporal distance similarity index, the expression of the path vector similarity measurement method is obtained as follows:

[0107]

[0108] Among them, VDTS(R1, R2) represents the path vector similarity of paths R1 and R2, VD(R1, R2) represents the directional similarity of the two path vectors, and VTS(R1, R2) represents the spatiotemporal distance similarity of the two paths.

[0109] Adaptive hierarchical clustering algorithm: Since the hierarchical clustering algorithm needs to input the number of clusters k as a parameter, the present invention proposes a method for automatically determining the optimal clustering parameters based on the cluster feature ratio to reduce the influence of the subjectivity of the clustering results.

[0110] Step 24: Define the path vector similarity distance.

[0111] The path vector similarity distance is defined as follows:

[0112] DVDTS(r1,r2)=N-VDTS(R1,R2) (16)

[0113] Among them, DVDTS(R1, R2) represents the path vector similarity distance of requests r1 and r2, N is a constant, and VDTS(R1, R2) represents the path vector similarity of the driving paths R1 and R2 of requests r1 and r2.

[0114] Step 25: Automatically determine the k value.

[0115] In the hierarchical clustering process, the path vector similarity distance has a monotonically increasing characteristic. The path similarity distance between similar entities or subclusters is generally small. When similar entities or subclusters are merged, the path vector similarity distance increases slowly; and when entities or subclusters with large differences are merged, the path vector similarity distance increases significantly. Therefore, by analyzing the changing characteristics of the path similarity distance in the hierarchical clustering process, it can provide an important reference for automatically determining the number of clusters k. As shown in formula (17), the number of clusters after each merger is defined as s, D s is the merge distance corresponding to the number of clusters s, R s It indicates the growth rate of the merged distance when the number of clusters is s. When the growth rate of the merged distance is small (the merged distance grows slowly), it means that the hierarchical structure tends to be stable. Therefore, the final number of clusters k satisfies the stable ratio. The maximum number of clusters s is shown in formula (18).

[0116]

[0117] Among them, the stable ratio It is the last inflection point of the ratio curve with significant growth.

[0118] Step 26: Clustering to obtain passenger request clusters.

[0119] The path vector similarity distance is obtained based on the path vector similarity, and the large-scale passenger requests are divided into k request clusters according to the automatically determined k value.

[0120] Step 3: Use the improved adaptive genetic algorithm to solve the carpooling path planning problem and obtain the passenger matching solution and carpooling route.

[0121] Traditional genetic algorithms achieve generational evolution of a population through gene selection-crossover-mutation operations, where the crossover operator obtains excellent individuals through gene recombination, and the mutation operator can ensure the diversity of the population. However, this operation of first crossover and then mutation can easily lead to the population fitness becoming more and more concentrated in the later stage, and the mutation operator does not change the population individuals much, resulting in local convergence. To this end, the present invention proposes an adaptive algorithm flow to prevent the algorithm from falling into a local optimum. The solution flow chart is shown in the figure below. Figure 1 shown.

[0122] Step 31: For each clustered passenger request cluster, the passenger sharing plan of a car is used as a chromosome, the chromosome is encoded and an initial sharing plan that meets the constraints of the hub taxi sharing model is generated as the initial population of the genetic algorithm.

[0123] Step 32: Taking the optimization objective of the hub taxi sharing model as the fitness function, calculate the fitness of the individual, and calculate the three-quarter median, one-quarter median, maximum value and minimum value of the population fitness.

[0124] Step 33: Adaptively determine the crossover probability and mutation probability. Since fixed control parameters are prone to affect the convergence performance of the genetic algorithm, in order to better play the role of control parameters, the crossover probability p c and mutation probability p m The adaptive calculation formula of is shown in equations (19) and (20).

[0125]

[0126] Among them, p c and p m Represent the crossover probability and mutation probability respectively, a is arcsin(f3-f1 / f max -f min), which indicates how close the fitness of the middle 50% of the individuals in the population is to the fitness of nearly 100% of the individuals between the maximum and minimum values ​​of the population. k1, k2, k3, k4 are constants.

[0127] Step 34: Determine whether the population meets the convergence condition. If so, output the passenger matching solution and the shared ride path. If not, continue to execute downward.

[0128] Step 35: Determine arcsin(f3-f1) / (f max -f min )≥π / 6. If so, it means that the closer the fitness of the middle 50% of the individuals in the population is to the fitness of nearly 100% of the individuals between the maximum and minimum values ​​of the population, the more concentrated the population is. Therefore, it is necessary to perform a mutation operation first to expand the diversity of the population and then perform a crossover operation to produce excellent individuals. If not, it means that the closer the fitness of the middle 50% of the individuals in the population is to the fitness of nearly 100% of the individuals between the maximum and minimum values ​​of the population, the more dispersed the population is. Therefore, it is necessary to perform a crossover operation first and then a mutation operation to ensure population diversity.

[0129] Step 36: Optimal preservation strategy. Calculate the optimal fitness of the offspring individuals after crossover and mutation operations and compare them with the individuals with the best fitness in the parent generation. If the optimal individual of the parent generation is better than the offspring, remove one individual of the offspring and put the optimal individual of the parent generation into the offspring. The optimal preservation strategy is also the key to ensure the convergence speed of the adaptive genetic algorithm designed by the present invention.

[0130] Step 37: Return to step 32.

[0131] Example 2

[0132] To realize the present invention, the data used is the real taxi data of Wuhan, China, with Hankou Station as the research object of the large transportation hub, and the taxi data departing from Hankou Station between 22:00 and 23:00 are screened, totaling 229. The data includes order ID, starting and ending latitude and longitude, and order start and end time.

[0133] Experimental parameter settings:

[0134] The parameters used in the model are set as follows: The fixed taxi usage fee c f is 20 yuan; the unit distance operation cost c d The cost is 1 yuan / km. The number of suitcases that can be placed in the trunk is 2; the number of pieces of luggage that can be placed on one seat is 2. The vehicle speed v is 40 km / h. The time s for getting off at each service point is 0.2 minutes.

[0135] The parameters in the solution algorithm are set as follows: In the PVS-based hierarchical clustering algorithm, the spatial weight α and the temporal weight β are 1 and 5 respectively, and the maximum cosine value of the angle between the two vectors is μ In the IGA algorithm, parameters k1 = 1.2, k2 = 1.0, k3 = 1.5, k4 = 0.5, and the maximum number of iterations is M is 100.

[0136] Regarding instance generation, the departure time window i is the order start time. Combined with the actual taxi operation data analysis in Wuhan, the passenger waiting time width is randomly generated in the range of [5,15]. The arrival time window is designed as [e i +d ij / v+s,l i +γ(d ij / v+s)], where γ=1.5 represents the maximum detour rate acceptable to passengers. Next, the number of passengers and the number of large luggage carried are randomly generated for each request. The number of passengers in each request is 1 and 2, with probabilities of 80% and 20%, respectively, and the number of large luggage carried is 0, 1, and 2, with probabilities of 50%, 40%, and 10%, respectively. The actual distances between all requests are obtained using the Python package OSMnx.

[0137] Experimental comparison:

[0138] In order to analyze the solution effect of the two-stage algorithm, the solution results of the PVS-AHC-IGA algorithm are compared with the solution results of the IGA algorithm directly. Table 1 shows the average results of the two algorithms after 10 solutions. It can be found that the two-stage algorithm of the present invention is equivalent to the model optimization target result solved under the global optimization method, but the time consumption is reduced by 85%, and the average detour ratio of passengers and the average waiting time of passengers are reduced by 8% and 30% respectively. It shows that the clustering method in this paper can effectively cluster similar passengers in the same cluster, improve the calculation efficiency of the algorithm, and thus achieve the purpose of rapid matching of large-scale passenger requests at the hub.

[0139] Table 1 Comparison of optimization results for passenger cases

[0140]

[0141] It should be understood that the size of the serial numbers of the steps in the above embodiments does not mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present application.

[0142] It should be understood that those skilled in the art can make improvements or changes based on the above description, and all these improvements and changes should fall within the scope of protection of the appended claims of the present invention.

Claims

1. A two-stage ride-sharing matching method for connecting taxis at large transportation hubs, characterized in that: The method comprises the following steps: Step 1: Taking the shortest total vehicle mileage and the minimum number of service vehicles as the optimization objectives, and considering the characteristics of hub passengers, and taking parameters including large luggage, departure and arrival time as constraints, a hub taxi sharing model is constructed; Step 2: Obtain passenger requests, and cluster passenger requests using an adaptive hierarchical clustering algorithm based on the path vector similarity of the fusion direction and spatiotemporal features to obtain passenger request clusters; Step 3: Based on the hub taxi carpooling model and passenger request clusters, the improved adaptive genetic algorithm is used to solve the carpooling path planning problem and obtain the passenger matching solution and carpooling driving path.

2. The two-stage ride-sharing matching method for connecting taxis at large transportation hubs according to claim 1 is characterized in that: The method for step 1 includes: Step 11: Establish model assumptions for the hub taxi-pooling model, including: The taxi arrives at the waiting area before the passengers and leaves the hub after all passengers have arrived; Passengers can arrive at the designated location before the earliest scheduled departure time; Each taxi has the same driving speed and travels at a constant speed throughout the journey; The road network is smooth, and the impact of random disturbances including traffic congestion is not considered; Step 12: Set the objective function and constraints of the hub taxi sharing model, including: The objective function is set to minimize the overall consumption of the system, that is, the total mileage of all vehicles and the number of service vehicles required are minimized; ensure that each passenger can be served by only one taxi; ensure that all service taxis depart from the hub and cannot return to the hub during the service process; ensure flow balance; capacity constraints, that is, the number of passengers in each request and the total amount of luggage carried cannot exceed the carrying capacity of the car; set time window constraints; the latest arrival time of passengers in the same car is the departure time of the taxi, which must meet the departure time window of each passenger; ensure that the arrival time of each passenger must meet the arrival time window; decision variables use 0-1 integer constraints; Step 13: Build a hub taxi sharing model based on model assumptions, objective functions and constraints.

3. The two-stage ride-sharing matching method for connecting taxis at large transportation hubs according to claim 2 is characterized in that: The hub taxi sharing model built in step 13 is as follows: In this case, a directed network G = (N, E) is introduced, where N is the set of points, E is the set of edges, i, j, l are the indexes of the passenger's requested destination, i, j, l∈N, 0 represents the hub, N0 represents the set of all points including the hub (0), k is the taxi index, k∈K; the case where the taxi returns to the hub is not considered, d ij is the actual distance from i to j; c f is the fixed usage fee for each taxi; c d is the unit distance operating cost of each taxi; Q represents the total capacity of taxis; p i is the number of passengers requesting i; b i is the number of large luggage requested by passenger i; B is the number of luggage that can be placed in the trunk; W is the number of luggage that can be placed on a seat; v is the vehicle speed; s is the time for each passenger to get off the bus, i.e., the service time; [e i ,l i ] and [ee i ,ll i ] represent the departure time window and arrival time window of passenger i respectively; x ijk is a decision variable that equals 1 if passenger i is served by taxi k from point j to point l, and 0 otherwise.

4. The two-stage ride-sharing matching method for connecting taxis at large transportation hubs according to claim 1 is characterized in that: The method of step 2 includes: Step 21: Based on the fact that the starting points of hub passengers are the same, the direction of the passenger route plays a decisive role in the driving path, and the direction similarity is calculated; Step 22: Calculate the spatiotemporal distance similarity based on the impact of the detour degree between the passenger's destinations on the passenger's ride-sharing experience; Step 23: According to the influence of the driving path and the ride time on the passenger matching, the direction similarity and the time-space distance similarity are combined, and the request time is integrated to calculate the path vector similarity; Step 24: Calculate the path vector similarity distance of the two path vectors according to the path vector similarity of the two path vectors; Step 25: Propose a method for automatically determining the optimal clustering parameters based on the clustering feature ratio, and automatically determine the number of clusters k by analyzing the change characteristics of the path similarity distance in the hierarchical clustering process; Step 26: Clustering to obtain passenger request clusters: Based on the path vector similarity distance and the automatically determined k value, the large-scale passenger requests are divided into k request clusters.

5. The two-stage ride-sharing matching method for connecting taxis at large transportation hubs according to claim 4 is characterized in that: The method for calculating the directional similarity in step 21 is: Calculate the cosine value cos<λ1,λ2> of the intersection angle between the vectors λ1 and λ2 obtained from the starting and ending points of the two paths. If μ≤cos<λ1,λ2>≤1, the two vectors are considered to be similar in direction, which can be expressed as: Among them, VD(R1, R2) represents the directional similarity of two paths, and M is a constant greater than a certain threshold; The method for calculating the spatiotemporal distance similarity in step 22 is: For any two passenger requests, with time attributes t1, t2 and location attributes x1, x2, the spatiotemporal distance similarity measure of the two paths R1 and R2 is obtained: Among them, VS(x1,x2) is the degree of detour between the two nodes, d(x1,x2) is the actual distance between the two nodes, d(0,x1) and d(0,x2) refer to the actual distances from hub 0 to the requested end points x1 and x2, respectively, VT(t1,t2) represents the degree of overlap of the passenger request time windows, and α and β are normalization factors that balance the importance of space-time distances.

6. The two-stage ride-sharing matching method for connecting taxis at large transportation hubs according to claim 4 is characterized in that: The method for calculating the path vector similarity in step 23 is: Adding direction similarity as an influencing factor to the spatiotemporal distance similarity index, we get the expression of the path vector similarity measurement method: Among them, VDTS(R1, R2) represents the path vector similarity of paths R1 and R2, VD(R1, R2) represents the direction similarity of the two path vectors, and VTS(R1, R2) represents the space-time distance similarity of the two paths; The method for calculating the path vector similarity distance in step 24 is: The path vector similarity distance is defined as follows: DVDTS(r1,r2)=N-VDTS(R1,R2) Among them, DVDTS(R1, R2) represents the path vector similarity distance of requests r1 and r2, N is a constant, and the driving paths corresponding to requests r1 and r2 are R1 and R2 respectively.

7. The two-stage ride-sharing matching method for connecting taxis at large transportation hubs according to claim 4 is characterized in that: In step 25, the method of automatically determining the number of clusters k by analyzing the change characteristics of the path similarity distance in the hierarchical clustering process includes: Define the number of clusters after each merger as s, D s is the merge distance corresponding to the number of clusters s, R s It indicates the growth rate of the merged distance when the number of clusters is s; when the growth rate of the merged distance is less than a certain threshold, it means that the merged distance grows slowly, indicating that the hierarchical structure tends to be stable. Therefore, the final number of clusters k satisfies the stable ratio. The maximum number of clusters s is: Among them, the stable ratio It is the last inflection point of the ratio curve with significant growth.

8. The two-stage ride-sharing matching method for connecting taxis at large transportation hubs according to claim 1 is characterized in that: The method of step 3 includes: Step 31: For each clustered passenger request cluster, the passenger sharing plan of a car is used as a chromosome, the chromosome is encoded and an initial sharing plan that meets the constraints of the hub taxi sharing model is generated as the initial population of the genetic algorithm; Step 32: Taking the optimization objective of the hub taxi sharing model as the fitness function, calculating the fitness of the individual, and calculating the three-quarter median, one-quarter median, maximum value and minimum value of the fitness of the population; Step 33: Adaptively determine the crossover probability and mutation probability; Step 34: Determine whether the population meets the convergence condition. If so, output the passenger matching solution and the shared ride path. If not, continue to execute downward. Step 35: Determine whether the population concentration condition is met. If so, it means that the fitness of 50% of the individuals in the middle of the population is close to the fitness of nearly 100% of the individuals between the maximum and minimum values ​​of the population, indicating that the population is more concentrated. Therefore, the mutation operation is performed first to expand the diversity of the population, and then the crossover operation is performed to produce excellent individuals. If not, it means that the fitness of 50% of the individuals in the middle of the population is not close to the fitness of nearly 100% of the individuals between the maximum and minimum values ​​of the population, indicating that the population is more dispersed. Therefore, the crossover operation is performed first and then the mutation operation is performed to ensure the diversity of the population. Step 36: Optimal preservation strategy; calculate the optimal fitness of the offspring individuals after crossover and mutation operations and compare them with the individuals with the best fitness in the parent generation. If the best individual in the parent generation is better than the offspring, remove one individual in the offspring and put the best individual in the parent generation into the offspring; Step 37: Return to step 32.

9. The two-stage ride-sharing matching method for connecting taxis at large transportation hubs according to claim 8, characterized in that: The method for adaptively determining the crossover probability and the mutation probability in step 33 is: Crossover probability p c and mutation probability p m The adaptive calculation formula is: Among them, p c and p m Represent the crossover probability and mutation probability respectively, a is arcsin(f3-f1 / f max -f min ), which indicates the closeness of the fitness of the middle 50% of the individuals in the population to the fitness of nearly 100% of the individuals between the maximum and minimum values ​​of the population, and k1, k2, k3, k4 are constants; The condition for judging the concentration of the population in step 35 is: arcsin(f3-f1) / (f max -f min )≥π / 6.

10. A two-stage ride-sharing matching system for connecting taxis at large transportation hubs, characterized in that: include: a memory for storing executable computer programs; The processor is used to implement the two-stage ride-sharing matching method for connecting taxis at a large transportation hub as described in any one of claims 1 to 9 when executing the executable computer program stored in the memory.

Images