Shared unmanned vehicle scheduling optimization method considering short-time demand prediction

Abstract

The invention relates to the technical field of unmanned driving, and discloses a shared unmanned vehicle scheduling optimization method considering short-time demand prediction, and the method comprises the steps: carrying out the short-time travel demand prediction based on the existing order data, and carrying out the matching of a predicted travel order and a real-time travel order, so as to reduce the real-time carpooling detour time; and after matching is completed, idle vehicles are dispatched to areas with more predicted demands, rebalance of the idle vehicles is achieved, feasible matching schemes and driving paths are optimized periodically, and finally the purposes of improving the carpooling matching rate and reducing the detouring time are achieved.

Description

Technical Field

[0001] The present invention belongs to the field of unmanned driving technology, and specifically relates to a shared unmanned driving vehicle scheduling optimization method considering short-term demand prediction. Background Art

[0002] The operation of traditional taxis and online car-hailing systems involves drivers. On the one hand, drivers may choose routes based on their experience, and their subjective factors will affect the system's service quality and vehicle dispatch efficiency; on the other hand, operations also need to consider the benefits of each driver, which increases the costs paid by travelers. This makes it difficult for taxi and online car-hailing systems to achieve unified dispatch and global benefit optimization.

[0003] Autonomous driving technology is one of the most revolutionary applications of artificial intelligence in the field of transportation. Since scientists and engineers began to envision autonomous vehicles that do not require manual operation in the early 20th century, autonomous driving technology has gradually matured after more than a century of development. As an important application of autonomous driving technology, shared autonomous vehicles, also known as self-driving taxis, which are a combination of shared travel and autonomous driving, have gradually appeared in people's daily lives.

[0004] Shared driverless vehicles are a new generation of travel that is completely controlled by the system. Its operating model is similar to that of taxis and online ride-hailing vehicles, but it does not rely on drivers. It can automatically complete driving tasks, exchange information with the system in real time, and obey system instructions for unified dispatch. It can complete driving tasks that drivers are unwilling to accept for personal benefits, eliminate the impact of the above two aspects, and is more conducive to the realization of carpooling, making unified vehicle dispatch and overall system benefit optimization possible.

[0005] The invention patent with application number 202010548189.3 discloses a driver-passenger matching method based on an autonomous shared taxi system, predicts the driving schedule based on real-time autonomous taxi information, and updates passenger matching and vehicle path arrangements with the goal of minimizing the increase in driving time. The invention patent with application number 202311258387.6 discloses a method and system for scheduling coordinated carpooling among groups of intelligent connected vehicles. It establishes a coordinated carpooling scheduling objective function by fully considering factors such as vehicle driving paths, passenger waiting time, and carpooling success rate, and uses an improved sand cat group optimization algorithm to solve the coordinated carpooling scheduling objective function. The resulting coordinated carpooling scheduling strategy for groups of vehicles has achieved good scheduling results.

[0006] The above method is based on the existing operating vehicles and order data, and better meets the carpooling needs by improving the carpooling scheduling model. However, the utilization and mining of existing data is still not sufficient, and short-term demand forecasting technology is not used to improve the performance of the scheduling system and give full play to the advantages of unmanned vehicles obeying unified scheduling. Summary of the invention

[0007] The purpose of the present invention is to solve the above problems and to design a shared unmanned vehicle scheduling optimization method considering short-term demand forecasting.

[0008] The technical solution of the present invention to achieve the above-mentioned purpose is a method for optimizing the scheduling of shared unmanned vehicles considering short-term demand forecasting, the process includes four stages: forecast demand generation, vehicle demand matching, driving path planning and vehicle rebalancing. The method includes:

[0009] Generate predicted orders, cluster the dispatching sites based on existing data, and then generate predicted travel orders in combination with the short-term prediction process; among them, the clustering of dispatching sites must first obtain the core distance and core points through the clustering minimum sample size, and then determine the set of clustering clusters with the maximum neighborhood radius, and then complete the clustering of dispatching sites; perform short-term demand forecasting based on existing data to obtain predicted travel orders.

[0010] Scheduling and optimization of shared driverless vehicles. Considering predicted orders, scheduling and optimizing shared driverless vehicles is carried out. Orders and vehicles in the system are processed centrally at regular intervals. The optimization process includes vehicle-demand matching, route planning, and vehicle rebalancing.

[0011] The existing data includes: city point of interest data, weather data and existing travel order data.

[0012] The clustering process of the scheduling sites includes:

[0013] Based on the data of the departure time and starting and ending coordinates of existing orders, combined with the characteristics of the urban road network, the OPTICS clustering algorithm is used to cluster the starting and ending coordinates of the orders to obtain the dispatch station information;

[0014] The clustering process of the OPTICS algorithm is based on the minimum clustering sample size MinPts, which mainly calculates the core distance and reachable distance of the core points;

[0015] The core distance cd is the distance that satisfies the coordinate point x and contains at least MinPts points. The calculation formula is formula (1):

[0016]

[0017] Among them, ò is the neighborhood radius, Represents the set N òThe node that is the i-th neighbor of node x in (x), Represents the core point x and point The distance between

[0018] If x is not a core point, cd(x) is meaningless;

[0019] If x is a core object, the reachable distance rd from object y to object x is the maximum value of the distance from y to x and the core distance of x, and y is the dense direct object of x; the calculation formula for the reachable distance rd of point y relative to the core object x is formula (2),

[0020]

[0021] If x is not a core point, then rd(y,x) is meaningless.

[0022] The implementation steps of the OPTICS algorithm are as follows:

[0023] Create two initialization queues, the ordered queue O and the result queue R. The ordered queue is used to store core objects and their density-reachable objects, and arrange them in ascending order according to their reachable distances. The result queue is used to record the processing order of data points, that is, add data points to the queue in the order of the algorithm processing process;

[0024] Select an unprocessed core object x, put it into the result queue R, and delete it from the data set. Find all the densities of x that directly reach the sample point y, calculate the reachable distance from y to x, and update the ordered queue O. If the ordered queue O is not empty, take out the sample point y' with the smallest reachable distance in O, put it into R, and delete it from the data set and O. If y' is a core object, find all the densities of y' that directly reach the sample points, and update O; repeat the above process until all points have been processed, and the clustering is completed to obtain the scheduling site;

[0025] Connect adjacent cluster sites, calculate the Euclidean distance between two points as the driving distance, and obtain the dispatch system road network as the basis for path calculation during dispatch.

[0026] The process of generating the predicted travel order includes:

[0027] The Markov Chain Monte Carlo method is used to make short-term predictions on order generation in different situations, and the predicted travel orders for a period of time in the future are obtained. The predicted orders are incorporated into the scheduling process to improve the performance of the scheduling system. The implementation process is as follows:

[0028] Based on the city's point of interest data, weather data, and existing travel order data, extract features, construct a sampling distribution function P for generating travel orders, and construct a Markov chain to make its stable distribution π satisfy π=P;

[0029] Then, sampling is performed based on the Monte Carlo simulation method. An initial state x0 is set to represent a travel scenario, and a random movement is made on the constructed Markov chain starting from x0 to generate multiple new travel scenarios. The sampling sequence is x0, x1, …, x t ,…;

[0030] Finally, statistical methods are used to check whether the Markov chain has converged to the target distribution and calculate the stationary sample set;

[0031] According to the traversal theorem, we obtain the appropriate number of samples n and the number of transfers m, and then obtain a random number according to the random sampling sequence, which conforms to the uniform distribution, that is, u∈[0,1]. We determine whether to accept the randomly sampled sample x by judging the relationship between u and the acceptance rate α after each iteration. Indicates that from state x i Transition to state x * The probability density function of i To state x * Time Satisfaction Then accept this state transfer x i →x * , we get x i =x * ; If not satisfied, then the state transfer is not accepted, that is, x i+1 =x i ; Thus, we can obtain the predicted order set {x m+1 ,x m+2 ,…,x n}.

[0032] The predicted order includes: departure time, arrival time, departure coordinates, and destination coordinates. The predicted order will serve as the basis for scheduling optimization of shared driverless vehicles.

[0033] The scheduling optimization process of the shared driverless car includes: vehicle-demand matching, path planning and vehicle rebalancing.

[0034] The vehicle-demand matching is that after the travel demand is submitted to the system and the predicted orders are calculated, the system searches for feasible travel plans for passengers traveling alone and those who accept carpooling; the vehicles searched by the system include idle vehicles and vehicles with empty seats, and the orders matched include real-time orders and predicted orders of unassigned vehicles; among them, the predicted orders are added to the scheduling, but because they are not real orders, there is no need to transport the virtual orders from the departure point to the destination, and the vehicle only needs to travel to the departure point of the predicted order.

[0035] Vehicles and demands are matched one-to-one in real time according to matching constraints. Ride-sharing is achieved by matching demands with vehicles with passengers. Matching constraints mainly include departure time constraints, vehicle capacity constraints, and cost constraints. Then, with the goal of maximizing system benefits, feasible matching solutions are optimized.

[0036] The car needs to match as follows:

[0037] First, we need to determine whether the vehicle-demand matching can meet the departure time requirements of the new demand. The new demand must be able to board the vehicle before the latest departure time. If the vehicle departs from station o, the time it takes to travel from the station to the starting point of the new demand r cannot exceed the latest acceptable departure time of demand r. The latest departure time is the expected departure time of the demand. Add the maximum acceptable waiting time mwait, and the calculation formula is (3):

[0038]

[0039] If a vehicle serves demand r after serving demand r', the time it takes to travel from the end point of demand r' to the starting point of the new demand r cannot exceed the latest acceptable departure time of demand r. The calculation formula is (4):

[0040]

[0041] In addition, to determine whether the vehicle-demand matching can meet the departure time requirements of the passengers on board, all passengers on board must be able to arrive at the destination before the latest arrival time, and the latest arrival time is the estimated arrival time. Add the maximum acceptable delay time mdelay of the demand, and the calculation formula is (5):

[0042]

[0043] Then, we need to judge whether the vehicle-demand matching can meet the vehicle capacity requirements, that is, the number of passengers in the vehicle cannot exceed the vehicle capacity C at any time. The calculation formula is (6):

[0044]

[0045] Based on all feasible solutions, the goal is to maximize the total system benefit, taking into account the impact of waiting time and delay, and optimizing the matching results. The calculation formula is shown in formula (7):

[0046]

[0047] To ensure that the status of vehicles and orders is unique, a series of constraints are required, see equations (8)-(12).

[0048]

[0049] Among them, F vr C is the fare of vehicle v serving demand r; vr The cost of serving demand r for vehicle v; C p C is the waiting time penalty; l Penalty for delay;

[0050] The solution optimization process can be transformed into a 0-1 programming problem for solution.

[0051] The path planning is to use the Dijkstra shortest path algorithm to calculate the driving path of a given travel plan, and calculate the time it takes for the vehicle to reach each node in the path based on a given vehicle speed;

[0052] The path distance calculation process is as follows:

[0053] First, initialize the route, select the starting point of the driving plan as the source point, set a set S to store the site set for which the shortest path has been found, and S is empty at the beginning. At the same time, set a distance array dist[] to record the shortest distance from the source point to each site. Initially, the distance of the source point is set to 0, and the distances of other vertices are set to infinity.

[0054] Then, site selection is performed. From the vertex set that has not yet been added to S, a site u with the smallest distance to dist[u] is selected to update the shortest path; for all adjacent sites v of site u, if the path length from u to v is less than the value of the current dist[v] record, then dist[v] is updated. The calculation formula is (13),

[0055] dist[v]=min(dist[v],dist[u]+w(u,v))(13)

[0056] Where w(u,v) is the length of the edge from site u to site v;

[0057] Finally, update the S set and add station u to the S set, indicating that the shortest path from the source point to u has been found; repeat the above steps until the S set contains the first waypoint on the driving plan;

[0058] After completion, the first waypoint is used as the source point, and the shortest path to the next waypoint is continued to be found until the complete path of the driving plan from the starting point to the end point is found;

[0059] The vehicle rebalancing is to dispatch idle vehicles to areas with more predicted demand after completing the order matching work, so as to achieve vehicle rebalancing;

[0060] At certain intervals, all available vehicles and unserviced orders in the system are centrally processed, feasible matching solutions are optimized, and driving routes are updated, which can maximize the order matching success rate and achieve the purpose of the optimization plan.

[0061] Beneficial Effects

[0062] Compared with the prior art, the technical solution of the present invention has the following advantages:

[0063] 1. This application proposes a short-term demand generation method for real-time shared unmanned vehicle scheduling, which can predict short-term travel demand based on urban land use, weather conditions and existing travel orders, generate predicted travel orders, and can be added to the scheduling process to fully utilize and mine existing data.

[0064] 2. This application takes into account the predicted travel situation in the process of scheduling shared driverless vehicles, and schedules short-term predicted orders together with real-time orders for carpooling. After the matching is completed, idle vehicles are dispatched to areas with more predicted demand to achieve vehicle rebalancing. This not only improves the current mismatch between vehicle supply and demand and increases the demand service rate, but also effectively reduces the impact of future demand uncertainty, reduces the detour time caused by real-time order carpooling, and improves the carpooling matching rate and scheduling system efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0065] Figure 1 It is a flow chart of a method for optimizing the scheduling of shared driverless vehicles considering short-term demand forecasting according to the present invention;

[0066] Figure 2 It is a structural diagram of a shared unmanned vehicle scheduling optimization method considering short-term demand forecasting described in the present invention. DETAILED DESCRIPTION

[0067] The present invention will be described in detail below in conjunction with the accompanying drawings. Figure 1-2 As shown;

[0068] In view of the shortcomings of the prior art, the present invention proposes a shared unmanned vehicle scheduling optimization method taking into account short-term demand forecasting, so as to make up for the shortcomings of the current shared unmanned vehicle system in real-time carpooling, such as low demand satisfaction rate and long detour time caused by unclear future demand.

[0069] The present invention solves the technical problem by adopting the following technical solution:

[0070] On the one hand, scheduling site clustering and short-term forecast order generation are performed based on existing data.

[0071] The method comprises:

[0072] The data foundation includes city point of interest data, weather data, existing travel order data, etc. The travel order includes order number, departure time, arrival time, departure coordinates, destination coordinates and other information.

[0073] Firstly, vehicle dispatching stations are clustered based on existing data.

[0074] The coordinates of the starting and ending points of travel orders are relatively scattered, and the number of orders is huge, so clustering analysis of existing travel orders is required. The density-based OPTICS clustering algorithm does not need to predetermine the number of clusters, but implements clustering based on the sorting results of the distance from the center point. The clustering results are mainly affected by the minimum sample size of the cluster and are insensitive to the maximum radius of the cluster. First, the core distance and core point are obtained based on the minimum sample size of the cluster, and then the set of cluster clusters is determined based on the maximum neighborhood radius.

[0075] Specifically, the departure time and starting and ending coordinates of existing orders in a specific city are used as the data basis, combined with the characteristics of the urban road network, and the OPTICS clustering algorithm is used to cluster the starting and ending coordinates of the orders to obtain the dispatch station information. This method mainly performs clustering based on the minimum clustering sample size MinPts, without specifying the specific number of clusters, and can better fit the actual distribution of urban transportation stations. The OPTICS algorithm mainly calculates the core distance and reachable distance of the core point.

[0076] The core distance cd is the distance that satisfies the coordinate point x and contains at least MinPts points. The calculation formula is formula (1):

[0077]

[0078] Among them, ò is the neighborhood radius, Represents the set N ò The node that is the i-th neighbor of node x in (x), Represents the core point x and point If x is not a core point, cd(x) is meaningless.

[0079] If x is a core object, the reachable distance rd from object y to object x is the maximum value of the distance from y to x and the core distance of x, and y is the dense direct object of x. The calculation formula for the reachable distance rd of point y relative to the core object x is formula (2),

[0080]

[0081] If x is not a core point, then rd(y,x) is meaningless.

[0082] The implementation steps of the OPTICS algorithm are as follows:

[0083] Create two initialization queues: ordered queue O and result queue R. The ordered queue is used to store core objects and their density-reachable objects, and is arranged in ascending order according to their reachable distances. The result queue is used to record the processing order of data points, that is, to add data points to the queue in the order of the algorithm processing.

[0084] Select an unprocessed core object x, put it into the result queue R, and delete it from the data set. Find all the densities of x that directly reach the sample point y, calculate the reachable distance from y to x, and update the ordered queue O. If the ordered queue O is not empty, take out the sample point y' with the smallest reachable distance in O, put it into R, and delete it from the data set and O. If y' is a core object, find all the densities of y' that directly reach the sample points, and update O. Repeat the above process until all points have been processed and clustering is completed to obtain the scheduling site.

[0085] Connect adjacent cluster sites, calculate the Euclidean distance between two points as the driving distance, and obtain the dispatch system road network as the basis for path calculation during dispatch.

[0086] Subsequently, travel generation predictions are performed to obtain predicted orders.

[0087] Markov Chain Monte Carlo is a Monte Carlo-based forecasting technique used to generate samples from complex probability distributions. This method uses the property of Markov chain, that is, the probability of a state transition depends only on the current state and has nothing to do with the previous state history, and is suitable for distributions that are difficult to sample directly, such as multidimensional distributions or distributions with complex structures.

[0088] The Markov Chain Monte Carlo method is used to make short-term predictions on order generation in different situations, and the predicted travel orders for a period of time in the future are obtained. The predicted orders are incorporated into the scheduling process to improve the performance of the scheduling system. The implementation process is as follows:

[0089] Based on the city's point of interest data, weather data, and existing travel order data, features are extracted, a sampling distribution function P of travel orders is constructed, and a Markov chain is constructed to make its stable distribution π satisfy π=P.

[0090] Based on the Monte Carlo simulation method, sampling is performed, and an initial state x0 is set to represent a travel scenario. Starting from x0, random movement is performed on the constructed Markov chain to generate multiple new travel scenarios, thereby obtaining a sampling sequence of x0, x1, …, x t ,….

[0091] Finally, we use statistical methods to check whether the Markov chain has converged to the target distribution and calculate the stationary sample set. According to the ergodic theorem, we obtain the appropriate number of samples n and the number of transfers m, and then obtain a random number based on the random sampling sequence, which conforms to the uniform distribution, that is, u∈[0,1]. We decide whether to accept the randomly sampled sample x by judging the relationship between u and the acceptance rate α after each iteration. Indicates that from state x i Transition to state x * The probability density function of . If x i To state x * Time Satisfaction Then accept this state transfer x i →x * , we get x i =x * ; If not satisfied, then the state transfer is not accepted, that is, x i+1 =x i . Thus, we can obtain the predicted order set {x m+1 ,x m+2 ,…,x n}.

[0092] On the other hand, the shared driverless vehicle carpooling scheduling optimization is performed by considering the short-term demand forecast. The method includes:

[0093] The scheduling optimization process of shared driverless cars includes: vehicle-demand matching, path planning, and vehicle rebalancing.

[0094] Vehicle-demand matching: After the travel demand is submitted to the system and the predicted orders are calculated, the system searches for feasible travel plans for passengers traveling alone and those who accept carpooling. The vehicles searched by the system include idle vehicles and vehicles with empty seats. The orders matched include real-time orders and predicted orders for unassigned vehicles. Among them, predicted orders are included in the scheduling, but because they are not real orders, there is no need to transport the virtual orders from the departure point to the destination. The vehicle only needs to travel to the starting point of the virtual order.

[0095] Vehicles and demands are matched one-to-one in real time according to matching constraints. Ride-sharing is achieved by matching demands with vehicles with passengers. Matching constraints mainly include departure time constraints, vehicle capacity constraints, and cost constraints, as follows:

[0096] First, we need to determine whether the vehicle-demand matching can meet the departure time requirements of the new demand. The new demand must be able to board the vehicle before the latest departure time. If the vehicle departs from station o, the time it takes to travel from the station to the starting point of the new demand r cannot exceed the latest acceptable departure time for demand r. The latest departure time is the expected departure time of the demand. Add the maximum acceptable waiting time mwait for the demand, and the calculation formula is formula (3). If the vehicle serves demand r after serving demand r', the time it takes to travel from the end point of demand r' to the starting point of the new demand r cannot exceed the latest acceptable departure time of demand r. The calculation formula is formula (4). In addition, to determine whether the vehicle-demand matching can meet the departure time requirements of the passengers on the vehicle, all passengers on the vehicle must be able to arrive at the destination before the latest arrival time, and the latest arrival time is the estimated arrival time. Add the maximum acceptable delay time mdelay of the demand, and the calculation formula is formula (5).

[0097]

[0098] Then, it is necessary to determine whether the vehicle-demand matching can meet the vehicle capacity requirements, that is, the number of passengers in the vehicle cannot exceed the vehicle capacity C at any time. The calculation formula is formula (6).

[0099]

[0100] Based on all feasible solutions, the goal is to maximize the total system benefit, taking into account the impact of waiting time and delay, and optimizing the matching results. The calculation formula is shown in formula (7). To ensure that the status of vehicles and orders is unique, a series of constraints are required, see formulas (8)-(12).

[0101]

[0102] Among them, F vr C is the fare of vehicle v serving demand r; vr The cost of serving demand r for vehicle v; C p C is the waiting time penalty; l Penalty for delay.

[0103] The solution optimization process can be transformed into a 0-1 programming problem for solution.

[0104] Path planning: The path planning uses the Dijkstra shortest path algorithm to calculate the driving path of a given travel plan, and calculates the time it takes for the vehicle to reach each node in the path based on the given vehicle speed. The path distance calculation process is as follows:

[0105] First, initialize. Select the starting point of the driving plan as the source point. Set a set S to store the set of sites for which the shortest path has been found. Initially, S is empty. At the same time, set a distance array dist[] to record the shortest distance from the source point to each site. Initially, the distance of the source point is set to 0, and the distances of other vertices are set to infinity.

[0106] Then, site selection is performed. From the set of vertices that have not yet been added to S, a site u with the smallest distance to dist[u] is selected to update the shortest path. For all adjacent sites v of site u, if the path length from u to v is less than the value of the current dist[v] record, then dist[v] is updated. The calculation formula is (13).

[0107] dist[v]=min(dist[v],dist[u]+w(u,v))(13)

[0108] Where w(u,v) is the length of the edge from site u to site v.

[0109] Finally, update the S set. Add station u to the S set, indicating that the shortest path from the source point to u has been found. Repeat the above steps until the S set contains the first waypoint on the driving plan.

[0110] After completion, the first waypoint is used as the source point to continue looking for the shortest path to the next waypoint until the complete path of the driving plan from the starting point to the end point is found.

[0111] Vehicle rebalancing: After completing the order matching work, idle vehicles are dispatched to areas with higher predicted demand to achieve vehicle rebalancing.

[0112] Dynamically optimize the vehicle dispatch plan. At certain intervals, all available vehicles and unserved orders in the system are centrally processed, feasible matching plans are optimized, and the driving routes are updated to maximize the order matching success rate and achieve the purpose of the optimization plan.

[0113] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions. The sentence "includes an element defined by ... does not exclude the existence of other identical elements in the process, method, article or device including the element".

[0114] The above technical solutions only reflect the preferred technical solutions of the technical solutions of the present invention. Some changes that may be made to certain parts thereof by technicians in this technical field all reflect the principles of the present invention and fall within the protection scope of the present invention.

Claims

1. A method for optimizing the scheduling of shared unmanned vehicles considering short-term demand forecasting, characterized in that: The method includes: Generate forecasted orders, cluster dispatching sites based on order data, and then make short-term travel demand forecasts based on real-time travel demand data to generate forecasted travel orders; Scheduling and optimization: Determine feasible scheduling plans for shared driverless vehicles based on real-time orders and predicted orders, and then optimize the scheduling plans for shared driverless vehicles.

2. The method for optimizing the scheduling of shared unmanned vehicles considering short-term demand forecasting according to claim 1 is characterized in that: The predicted order is added to the scheduling process of the shared driverless car. When serving the predicted order, the shared driverless car only needs to travel to the order starting point, and there is no need to provide travel services for the predicted order.

3. The method for optimizing the scheduling of shared unmanned vehicles considering short-term demand forecasting according to claim 2 is characterized in that: The existing data includes: city point of interest data, weather data and existing travel order data; The clustering process of the scheduling sites includes: Based on the data of the departure time and starting and ending coordinates of existing orders, combined with the characteristics of the urban road network, the OPTICS clustering algorithm is used to cluster the starting and ending coordinates of the orders to obtain the dispatch station information; The clustered sites are connected, and the Euclidean distance between the two points is calculated as the vehicle travel distance to obtain the dispatching system road network, which serves as the basis for path calculation during dispatching.

4. The method for optimizing the scheduling of shared unmanned vehicles considering short-term demand forecasting according to claim 3 is characterized in that: The process of generating predicted orders includes: Based on the city's point of interest data, weather data, and existing travel order data, extract features, construct a sampling distribution function P of travel orders, and construct a Markov chain to make its stable distribution π satisfy π=P; Then, sampling is performed based on the Monte Carlo simulation method. An initial state x0 is set to represent a travel scenario, and a random movement is made on the constructed Markov chain starting from x0 to generate multiple new travel scenarios. The sampling sequence is x0, x1, …, x t ,…; Finally, statistical methods are used to check whether the Markov chain has converged to the target distribution, calculate the stationary sample set and obtain the predicted order.

5. The method for optimizing the scheduling of shared unmanned vehicles considering short-term demand forecasting according to claim 4 is characterized in that: The predicted order includes: departure time, departure coordinates, arrival time, and destination coordinates. The predicted order will serve as the basis for optimizing the scheduling of shared driverless vehicles.

6. The method for optimizing the scheduling of shared unmanned vehicles considering short-term demand forecasting according to claim 5 is characterized in that: The scheduling optimization process of the shared driverless online car-hailing service includes: vehicle-demand matching, path planning and vehicle rebalancing.

7. The method for optimizing the scheduling of shared unmanned vehicles considering short-term demand forecasting according to claim 6 is characterized in that: The vehicle demand matching is that after the real-time travel demand is submitted to the system and the predicted order is calculated, the system finds feasible matching solutions for passengers traveling alone and those who accept carpooling; Vehicles and demands are matched one-to-one in real time according to matching constraints. When sharing a ride, demands are matched with vehicles with passengers. Matching constraints mainly include departure time constraints, vehicle capacity constraints, and cost constraints. Then, with the goal of maximizing system benefits, the feasible matching solutions are optimized.

8. The method for optimizing the scheduling of shared unmanned vehicles considering short-term demand forecasting according to claim 6 is characterized in that: The path planning is to use the Dijkstra shortest path algorithm to calculate the driving path of a given travel plan, and calculate the time it takes for the vehicle to reach each node in the path based on a given vehicle speed.

9. The method for optimizing the scheduling of shared unmanned vehicles considering short-term demand forecasting according to claim 6, characterized in that: The vehicle rebalancing is to dispatch idle vehicles to areas with more predicted demand after completing the order matching work to achieve vehicle rebalancing.

Images