Hierarchical queuing matching method based on driver position and waiting duration

Abstract

The invention discloses a grading queuing matching method based on driver positions and waiting duration, and aims to solve the problem that ranking of low-grade drivers is reversed due to the fact that high-grade drivers are inserted into low-grade queues by starting a downward order listening function after finishing queuing in a current station queue. The situation not only destroys the queuing order of the low-grade drivers, but also seriously affects the order receiving rights and interests of the low-grade drivers. According to the invention, a queue management mechanism is reasonably designed, so that the high-level drivers are sorted according to the time of entering the low-level queue when selecting to join the low-level queue, and the original queuing sequence of the low-level drivers is not interfered, so that the order receiving opportunity of the low-level drivers is ensured. In addition, the ranking of each grade queue is only related to the time when the driver enters the corresponding vehicle type grade queue and the condition meeting the queuing requirement, and the unfair phenomenon caused by external factors or behaviors of other grade drivers is avoided.

Description

Technical Field

[0001] The present invention belongs to the field of hierarchical queuing matching, and specifically relates to a hierarchical queuing matching method based on driver location and waiting duration. Background Art

[0002] With the development of the online car-hailing business, the number of online car-hailing drivers has gradually increased. At present, when drivers on each platform receive orders at the station, there are situations where they wait for a long time without order assignment or are assigned short orders after a long wait, resulting in problems such as drivers' income not meeting expectations (long time without receiving orders or short orders) and poor experience (expected stability and fairness of order assignment time). To solve the above problems, this article provides a hierarchical queuing matching scheme based on driver location and waiting time.

[0003] The existing queuing matching ability has the following problems: Regarding the calculation of driver rankings, the time when a driver enters the area and the waiting duration are recorded for each driver individually. Different levels of drivers belong to different queues. Therefore, high-level drivers who enter the area first will insert into the low-level driver queue after turning on the downward order listening function, rather than entering the end of the queue, resulting in chaos in the queuing order of low-level drivers and affecting drivers' queuing experience and income.

[0004] In view of this, the present invention is specifically proposed. Summary of the Invention

[0005] The technical problem to be solved by the present invention is to overcome the deficiencies of the prior art and provide a hierarchical queuing matching method based on driver location and waiting duration, solving the problems proposed in the above background art.

[0006] To solve the above technical problems, the basic concept of the technical solution adopted by the present invention is:

[0007] A hierarchical queuing matching method based on driver location and waiting duration includes the following steps:

[0008] Obtain the geographical location of the driver and the time of entering the queuing area through the real-time location information uploaded by the driver side, and continuously detect the driver's status to confirm whether the driver meets the conditions for entering the queuing queue.

[0009] Classify the drivers entering the queuing area according to their service levels and assign them to the corresponding level queues; within the queue, sort the drivers in descending order of waiting duration.

[0010] Based on historical order data and the real-time driver distribution, use the kernel density estimation method to perform non-parametric modeling on passenger demand and predict the demand probability distribution in each area within different time windows.

[0011] Optimize the driver allocation decision using a two-stage stochastic programming model. In the first stage, calculate the number of drivers allocated to each region according to the predicted demand distribution. And in the second stage, after the actual demand is revealed, dynamically adjust the number and route of drivers for reallocation.

[0012] Allocate orders according to the dynamically adjusted queue order. For long-distance orders (such as those with a distance exceeding 20 kilometers), give priority to allocating them to the driver at the head of the queue with the longest waiting time; for short-distance orders (such as those with a distance less than 20 kilometers), give priority to allocating them to the driver who has just entered the queue, and at the same time ensure that the fairness of the queue for low-level drivers is not affected when high-level drivers join the low-level queue.

[0013] Optionally, the steps to obtain the geographical location of the driver and the time of entering the queuing area through the real-time location information uploaded by the driver side, and to detect the driver's status in real time, including the status of being idle, taking an order, or pausing to take an order, to confirm whether the driver meets the conditions for entering the queuing queue are as follows:

[0014] Through the positioning data uploaded by the driver-side device, the system obtains the geographical location information of each driver in real time and determines whether it enters the queuing area, which is defined by an electronic fence. Among them, for the drivers who enter the queuing area, the system automatically records the time when they enter the area and uses this time as the initial basis for subsequent queuing sorting.

[0015] The system detects the driver's current status in real time. Among them, if the driver is not currently taking an order, it meets the queuing conditions; if the driver is currently performing an order task, it does not meet the queuing conditions; if the driver manually closes the order-taking function or pauses to take an order, it does not meet the queuing conditions.

[0016] According to the driver's real-time status, only allow the drivers in the idle state to enter the queuing queue, and allocate the eligible drivers to the corresponding queues according to their service levels and the time of entering the queuing area. After being allocated to the queue, detect and update the positions and statuses of all drivers who have entered the area at fixed intervals. If the driver's status changes, reconfirm whether it meets the conditions for entering the queuing queue.

[0017] Optionally, classify the drivers who enter the queuing area according to their service levels and allocate them to the queues of corresponding levels; the steps to sort them in descending order of the driver's waiting time within the queue are as follows:

[0018] Based on the driver's service level information, classify the drivers into high-level drivers and low-level drivers.

[0019] Drivers entering the queuing area are assigned to corresponding ranked queues according to their service levels; the queues are managed independently by rank, with the high-ranked queue and the low-ranked queue being independent of each other. Then, for each driver entering the queuing area, the system records the time when the driver enters the queuing area in real time and continuously tracks the waiting duration of the driver within the area, which serves as the basis for queue sorting.

[0020] Within the same-rank queue, the drivers are sorted from highest to lowest according to their waiting durations. The driver with a longer waiting duration ranks higher, and the driver with a shorter waiting duration ranks lower; the sorting rule is dynamically adjusted each time a driver's status is updated or a new driver enters the queue to ensure the real-time nature of the sorting.

[0021] Finally, continuously monitor the queuing status of the drivers. If a driver accepts an order, exits the queue, or changes status, the system will immediately update the queue sorting and feedback the latest sorting results to other drivers in the queue.

[0022] Optionally, based on historical order data and real-time driver distribution, the steps for non-parametrically modeling passenger demand using the kernel density estimation method and predicting the demand probability distribution in different time windows for each region are as follows:

[0023] Extract historical order data and real-time driver distribution information from the platform database. The data includes the order occurrence time, order location, order completion time, and the geographical location distribution of real-time drivers; clean the abnormal data to ensure the integrity and accuracy of the data.

[0024] Based on the geographical information of the city, divide the service area into multiple small regions, and generate demand samples for time windows within each region. The samples include the number of orders and the order location distribution within each time window.

[0025] For the demand samples within each region, use the kernel density estimation method to construct the probability distribution function of demand, and non-parametrically model the historical data. Its expression is: Among them, represents the probability density function of demand within the time window, N is the number of historical demand samples, d is the demand value of the current time window,

[0026] d i is the demand sample in the historical data, h is the bandwidth parameter that controls the smoothness of the kernel function, and K(u) is the kernel function. Among them, a linear adjustment factor is added to enhance the estimation ability for boundary demands;

[0027] Combined with the real-time driver distribution data, combine the kernel density estimation results with the real-time distribution information, and dynamically adjust the weights of demand prediction so that the demand prediction can reflect the changes in the supply-demand relationship in real time. Its expression is: Among them, is the corrected demand probability density function; ΔS represents the real-time newly added driver supply; ΔD represents the real-time demand change;

[0028] According to the constructed corrected demand distribution function Predict the demand in each region within the future time window, and calculate the expected value of the demand in each region as: Finally, output the demand probability distribution and demand expected value of each region as the input basis for subsequent driver allocation and rebalancing.

[0029] Optionally, when performing step S4, first, based on historical order data and real-time driver distribution, use the improved kernel density estimation method to predict the demand distribution in each region within the current time window t where i represents the region index and s represents the demand scenario.

[0030] Optionally, when allocating drivers in the first stage, first, use the stochastic programming model to calculate the number of drivers allocated to each region to meet the optimization goal of predicted demand;

[0031] The objective function is to minimize the supply-demand imbalance cost, defined as: where: Q represents the penalty cost for supply falling short of demand; W represents the penalty cost for supply exceeding demand; R represents the set of all regions;

[0032] The allocation constraint condition is: where M t is the total number of available drivers within the current time window. After the demand actually occurs, the system reveals the real demand in each region Then, calculate the supply-demand difference in each region and adjust the driver reallocation decision based on this.

[0033] Optionally, when making the rebalancing decision in the second stage, first, dynamically optimize the reallocation quantity and path of drivers according to the allocation results in the first stage and the revealed actual demand The optimization goal is to minimize the rebalancing cost and the remaining supply-demand imbalance cost: where, C i，j represents the rebalancing cost from region i to region j, which is proportional to the distance between the two places; represents the number of drivers reallocated from region i to region j.

[0034] The total number of drivers reallocated in the second stage cannot exceed the number of drivers allocated in the first stage: where all variables must be non-negative integers:

[0035] The number of drivers assigned in the first stage and the driver routes and quantities reallocated in the second stage are output to the dispatching system. Finally, according to the reallocation result, idle drivers are guided to the target area to complete the optimization of supply-demand balance.

[0036] After adopting the above technical solution, the present invention has the following beneficial effects compared with the prior art. Of course, any product implementing the present invention does not necessarily need to achieve all the advantages described below:

[0037] The present invention aims to solve the problem that after high-level drivers complete the queuing in their own level queue during the current yard queuing, they insert into the low-level queue by turning on the function of listening for orders downward, which causes the ranking of low-level drivers to regress. This situation not only disrupts the queuing order of low-level drivers but also seriously affects the order-taking rights and interests of low-level drivers. By reasonably designing the queue management mechanism, when high-level drivers choose to join the low-level queue, they are sorted according to the time of entering the low-level queue, without interfering with the original queuing order of low-level drivers, thus ensuring the order-taking opportunities of low-level drivers. In addition, the present invention ensures that the ranking of each level queue is only related to the time when the driver enters the corresponding vehicle type level queue and the conditions for meeting the queuing requirements, and there will be no unfair phenomena due to external factors or the behavior of drivers in other levels. Through this design, the system realizes the fairness of order dispatching, improves the queuing experience and order-taking efficiency of drivers at different levels, and finally optimizes the overall operation order of the platform.

[0038] The following further describes in detail the specific implementation manners of the present invention with reference to the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] The accompanying drawings in the following description are only some embodiments. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts. In the

[0040] drawings:

[0041] Figure 1 is a schematic structural diagram.

[0042] It should be noted that these drawings and the text description are not intended to limit the scope of the concept of the present invention in any way, but to illustrate the concept of the present invention to those skilled in the art by referring to specific embodiments. DETAILED DESCRIPTION OF THE INVENTION

[0043] Now, the present invention will be further described in detail with reference to the accompanying drawings.

[0044] Please refer to Figure 1As shown in the figure, in this embodiment, a hierarchical queuing matching method based on the driver's location and waiting time is provided, including the following steps:

[0045] Through the real-time location information uploaded by the driver side, obtain the geographical location of the driver and the time of entering the queuing area, and detect the driver's status in real time, including the status of being idle, taking an order, or pausing to take an order, so as to confirm whether the driver meets the conditions for entering the queuing queue.

[0046] Classify the drivers entering the queuing area according to the driver's service level (driver evaluation), and assign them to the queues of corresponding levels; within the queue, sort the drivers according to the waiting time from high to low, and at the same time, high-level drivers can choose whether to join the low-level queue (listening to orders downward).

[0047] Based on historical order data and the real-time driver distribution, use the kernel density estimation method to non-parametrically model the passenger demand, and predict the demand probability distribution in each area within different time windows;

[0048] Use a two-stage stochastic programming model to optimize the driver allocation decision. In the first stage, calculate the number of drivers allocated to each area according to the predicted demand distribution And in the second stage, after the actual demand is revealed, dynamically adjust the number and path of the driver reallocation

[0049] Allocate orders according to the dynamically adjusted queue order. For long-distance orders (such as those with a distance exceeding 20 kilometers), give priority to allocating them to the driver at the head of the queue with the longest waiting time; for short-distance orders (such as those with a distance less than 20 kilometers), give priority to allocating them to the driver who has just entered the queue, and at the same time ensure that the fairness of the low-level drivers' queuing is not affected after high-level drivers join the low-level queue.

[0050] In this embodiment, the step of obtaining the geographical location of the driver and the time of entering the queuing area through the real-time location information uploaded by the driver side, and detecting the driver's status in real time, including the status of being idle, taking an order, or pausing to take an order, so as to confirm whether the driver meets the conditions for entering the queuing queue is as follows:

[0051] Through the positioning data uploaded by the driver-side device, the system can obtain the geographical location information of each driver in real time and determine whether it enters the queuing area, which is defined by an electronic fence. Among them, for the drivers entering the queuing area, the system automatically records the time when they enter the area, and uses this time as the initial basis for subsequent queuing sorting;

[0052] The system detects the driver's current status in real time. Among them, if the driver has not taken an order currently, it meets the queuing conditions; if the driver is currently performing an order task, it does not meet the queuing conditions; if the driver manually closes the order-taking function or pauses to take an order, it does not meet the queuing conditions

[0053] According to the real-time status of the drivers, only the drivers in the idle state are allowed to enter the queuing queue, and the eligible drivers are assigned to the corresponding queues according to their service levels and the time of entering the queuing area. After being assigned to the queue, the positions and statuses of all drivers who have entered the area are detected and updated at fixed intervals (such as every 3 seconds). If the driver's status changes (such as switching from the suspended state to the idle state), it is reconfirmed whether they meet the conditions for entering the queuing queue.

[0054] In this embodiment, the drivers entering the queuing area are classified according to their service levels (driver evaluations) and assigned to the queues of corresponding levels. The steps of sorting the drivers in the queue from the highest to the lowest waiting duration are as follows:

[0055] Based on the service level information of the drivers, the drivers are classified into high-level drivers and low-level drivers; the service level is dynamically adjusted by the platform according to the historical service performance of the drivers, and can specifically be sorted by star rating;

[0056] The drivers entering the queuing area are assigned to the corresponding level queues according to their service levels; the queues are managed independently by level, and the high-level queue and the low-level queue are independent of each other. Then, for each driver entering the queuing area, the system records the time when they enter the queuing area in real time and continuously tracks their waiting duration in the area as the basis for queue sorting;

[0057] In the same-level queue, the drivers are sorted from the highest to the lowest waiting duration. The drivers with longer waiting durations are ranked higher, and the drivers with shorter waiting durations are ranked lower; the sorting rules are dynamically adjusted every time the driver's status is updated or a new driver enters the queue to ensure the real-time nature of the sorting;

[0058] Finally, continuously monitor the queuing status of the drivers. If a driver accepts an order, exits the queue, or changes the status, the system will immediately update the queue sorting and feedback the latest sorting results to the other drivers in the queue.

[0059] In this embodiment, based on the historical order data and the real-time driver distribution, the steps of using the kernel density estimation method to perform non-parametric modeling on the passenger demand and predicting the demand probability distribution in different time windows for each region are as follows:

[0060] Extract the historical order data and the real-time driver distribution information from the platform database. The data includes the order occurrence time, the order location, the order completion time, and the geographical location distribution of the real-time drivers; clean the abnormal data to ensure the integrity and accuracy of the data;

[0061] Based on the geographical information of the city, the service area is divided into multiple small regions, and demand samples for time windows are generated within each region. The samples include the order quantity and the distribution of order locations within each time window.

[0062] For the demand samples within each region, the kernel density estimation method is used to construct the probability distribution function of demand, which models the historical data in a non-parametric way. Its expression is: Among them, represents the probability density function of demand within the time window, N is the number of historical demand samples, d is the demand value of the current time window,

[0063] d i is the demand sample in the historical data, h is the bandwidth parameter that controls the smoothness of the kernel function, and K(u) is the kernel function. Among them, a linear adjustment factor is added to enhance the estimation ability for boundary demands

[0064] Combined with the real-time driver distribution data, the kernel density estimation result is combined with the real-time distribution information to dynamically adjust the weight of demand prediction, so that the demand prediction can reflect the changes in the supply-demand relationship in real time. Its expression is:

[0065] Among them, is the corrected demand probability density function; ΔS represents the real-time newly added driver supply; ΔD represents the real-time demand change;

[0066] According to the constructed corrected demand distribution function Predict the demand quantity of each region within the future time window, and calculate the expected value of the demand for each region as: Finally, output the demand probability distribution and the expected demand value of each region as the input basis for subsequent driver allocation and rebalancing.

[0067] In this embodiment, when executing step S4, first, based on the historical order data and the real-time driver distribution, the improved kernel density estimation method is used to predict the demand distribution of each region within the current time window t where i represents the region index and s represents the demand scenario.

[0068] In this embodiment, when allocating drivers in the first stage, first, use the stochastic programming model to calculate the number of drivers allocated to each region to meet the optimization goal of predicted demand;

[0069] The objective function is to minimize the supply-demand imbalance cost, which is defined as: Among them: Q represents the penalty cost for supply falling short of demand; W represents the penalty cost for supply exceeding demand; R represents the set of all regions;

[0070] The allocation constraint is as follows: where M t is the total number of available drivers within the current time window. After the demand actually occurs, the system reveals the true demand in each region Next, calculate the supply-demand difference in each region and based on this, adjust the driver reallocation decision.

[0071] In the second stage of rebalancing decision-making in this embodiment, first, according to the allocation results of the first stage and the revealed actual demand, dynamically optimize the reallocation quantity and path of the drivers The optimization objective is to minimize the rebalancing cost and the remaining supply-demand imbalance cost: where C i，j represents the rebalancing cost from region i to region j, which is proportional to the distance between the two places; represents the number of drivers reallocated from region i to region j.

[0072] The total number of drivers reallocated in the second stage cannot exceed the number of drivers allocated in the first stage: where all variables must be non-negative integers:

[0073] The number of drivers allocated in the first stage and the paths and quantities of the drivers reallocated in the second stage are output to the dispatching system. Finally, according to the reallocation results, guide the idle drivers to the target regions to complete the supply-demand balance optimization.

[0074] Explanation of related terms

[0075] Driver location: After the driver starts driving, the order-taking status is enabled. At this time, the system reports the longitude and latitude information of the driver's location, which is used to match and determine the driver's location by the system.

[0076] Waiting duration: After the driver enters a certain region while in the driving state, the duration of waiting for orders within that region.

[0077] Hierarchical queuing: Based on parameters such as the driver's service level and vehicle brand, classify the drivers' transport capacities. Different levels have the qualification to receive different orders. In the queuing strategy, drivers of different levels belong to different queues.

[0078] Listening to lower-level orders: High-level drivers control whether to be downward compatible with the order-taking capabilities of lower-level drivers, which is the function of listening to lower-level orders.

[0079] Queue matching: In a certain area (usually airports or railway stations), to ensure the order of order assignment after drivers gather, queue sorting is carried out based on the order of drivers entering the area and their waiting time, and order assignment matching is performed according to the queue ranking (long orders are assigned first + short orders are assigned last).

[0080] Long order assignment first: Orders with a mileage above a certain number of kilometers, that is, long orders, are preferentially assigned to the driver at the head of the queue, and orders are assigned in the order of ranking from the head to the tail of the queue.

[0081] Short order assignment last: Orders with a mileage below a certain number of kilometers, that is, short orders, are preferentially assigned to the driver at the end of the queue, and orders are assigned in the order of ranking from the end to the head of the queue.

[0082] Short orders and long orders: Orders are classified based on the order kilometers. The corresponding kilometers for different airports and railway stations are different, generally 20 km.

[0083] Vehicle shutdown: The driver manually closes the order receiving ability, and the platform will no longer assign orders to such drivers

[0084] To verify the beneficial effects of the present invention in protecting the order receiving rights and interests of low-level drivers, improving the fairness of order assignment, and optimizing the operation order of the platform, the following experiments were designed and conducted. The experimental data proved the effectiveness and practical application value of the present invention.

[0085] Experimental background:

[0086] Test location: High-density demand areas such as airports and railway stations in the city are selected.

[0087] Data source: Simulate historical real order data, involving a total of about 1,000 high-level drivers and low-level drivers.

[0088] Traditional mode: High-level drivers are allowed to insert into the low-level queue by enabling the downward order listening function.

[0089] (The present invention): After high-level drivers enable the downward order listening function, they are sorted according to the time of entering the low-level queue without disturbing the original order of low-level drivers.

[0090] 2. Experimental indicators:

[0091] Order receiving rate of low-level drivers: The proportion of low-level drivers who successfully receive orders.

[0092] Average waiting time of low-level drivers: The average time (in minutes) for low-level drivers from entering the queue to receiving an order.

[0093] Number of queue ranking changes: The number of times the ranking of low-level drivers changes due to being cut in by high-level drivers during the queuing process.

[0094] Overall platform efficiency: average passenger waiting time (in minutes) and empty running rate (the proportion of the driver driving empty).

[0095] Experimental results

[0096] 1. Order acceptance rate of low - level drivers:

[0097] Mode Order acceptance rate of low-level drivers (%) Traditional mode 68.4 (This invention) 85.7

[0098] The present invention increases the order acceptance rate of low - level drivers by 17.3 percentage points, effectively protecting the rights and interests of low - level drivers.

[0099] 2. Average waiting time of low - level drivers:

[0100]

[0101]

[0102] Through improvement, the downward order - listening behavior of high - level drivers will not disrupt the original ranking of low - level drivers, significantly shortening the waiting time of low - level drivers by about 6.7 minutes.

[0103] 3. Number of changes in queue ranking:

[0104] Mode Number of ranking changes (times) Traditional mode 327 (This invention) 0

[0105] In the traditional mode, due to high - level drivers jumping the queue, the rankings of low - level drivers were disrupted many times. Under the present invention, high - level drivers join the low - level queue and are sorted according to the entry time, completely eliminating the problem of ranking changes for low - level drivers and maintaining the stability and fairness of the queue.

[0106] 4. Overall platform efficiency (average passenger waiting time and empty running rate):

[0107] Mode Average passenger waiting time (minutes) Driver empty running rate (%) Traditional mode 9.8 21.3 (This invention) 8.4 18.2

[0108] The present invention, by optimizing the driver queue management and order acceptance sequence, reduces the ineffective scheduling in the order allocation process, significantly reducing the passenger waiting time (by 1.4 minutes) and the empty running rate of drivers (by 3.1 percentage points).

[0109] The present invention is not limited to the above - mentioned embodiments. Anyone should know that structural changes made under the inspiration of the present invention, as long as they have the same or similar technical solutions as the present invention, fall within the protection scope of the present invention. The technologies, shapes, and structures not described in detail in the present invention are all well - known technologies.

Claims

1. A hierarchical queue matching method based on driver position and waiting time, characterized in that: The following steps are involved: The real-time location information uploaded by the driver is used to obtain the driver’s geographical location and the time when the driver entered the queue area, and the driver’s status is detected in real time to confirm whether the driver meets the conditions for entering the queue; Drivers entering the queue area are graded according to their service levels and assigned to queues of corresponding levels; within the queue, drivers are sorted from high to low according to their waiting time; Based on historical order data and real-time driver distribution, the kernel density estimation method is used to non-parametrically model passenger demand and predict the probability distribution of demand in different time windows in each area. A two-stage stochastic programming model is used to optimize the driver allocation decision. In the first stage, the number of drivers to be allocated to each area is calculated based on the predicted demand distribution. In the second phase, the number and routes of driver reallocation will be dynamically adjusted based on actual demand. Orders are allocated according to the dynamically adjusted queue order. For long-distance orders, priority is given to the driver at the head of the queue who has waited the longest. For short-distance orders, priority is given to the driver who has just entered the queue, while ensuring that the fairness of the queue for low-level drivers is not affected when high-level drivers join a low-level queue.

2. A hierarchical queue matching method based on driver position and waiting time according to claim 1, characterized in that: The real-time location information uploaded by the driver is used to obtain the driver's geographic location and the time when the driver entered the queue area. The driver's status is detected in real time, including whether the driver is idle, accepting orders, or suspended from accepting orders. The steps to confirm whether the driver meets the conditions for entering the queue are as follows: Through the positioning data uploaded by the driver's device, the system obtains the geographic location information of each driver in real time and determines whether he has entered the queue area, which is defined by the electronic fence. Among them, for drivers who enter the queue area, the system automatically records the time they entered the area and uses this time as the initial basis for subsequent queue sorting; The system detects the current status of the driver in real time. The driver may not be currently accepting orders, which means he meets the queuing conditions; the driver may be currently executing an order task, which means he does not meet the queuing conditions; the driver may manually turn off the order-taking function or suspend accepting orders, which means he does not meet the queuing conditions. Based on the real-time status of the driver, only idle drivers are allowed to enter the queue, and eligible drivers are assigned to the corresponding queue according to their service level and the time they enter the queue area. After being assigned to the queue, the location and status of all drivers who have entered the area are checked and updated at fixed intervals. If the driver's status changes, it is reconfirmed whether he meets the conditions for entering the queue.

3. The hierarchical queue matching method based on driver position and waiting time according to claim 1, characterized in that: Drivers entering the queue area are graded according to their service levels and assigned to queues of corresponding levels; within the queue, the steps for sorting drivers from high to low according to their waiting time are as follows: Based on the driver's service level information, the drivers are divided into high-level drivers and low-level drivers; Allocate drivers entering the queuing area to corresponding level queues according to their service levels; The queues are managed independently according to levels, and high-level queues and low-level queues are independent of each other. Then, for each driver entering the queue area, the system records the time when he enters the queue area in real time, and continuously tracks his waiting time in the area as the basis for queue sorting; In the same level queue, drivers are ranked from high to low according to their waiting time, with drivers with longer waiting time ranking higher and drivers with shorter waiting time ranking lower; The sorting rules are dynamically adjusted every time a driver's status is updated or a new driver enters the queue to ensure real-time sorting; Finally, the driver's queue status is continuously monitored. If a driver accepts an order, exits the queue, or changes status, the system will immediately update the queue order and feedback the latest order results to other drivers in the queue.

4. The hierarchical queue matching method based on driver position and waiting time according to claim 1, characterized in that: Based on historical order data and real-time driver distribution, the kernel density estimation method is used to non-parametrically model passenger demand and predict the probability distribution of demand in each area in different time windows. The steps are as follows: Extract historical order data and real-time driver distribution information from the platform database, including order occurrence time, order location, order completion time, and real-time driver geographic location distribution; clean up abnormal data to ensure data integrity and accuracy; Based on the city's geographic information, the service area is divided into multiple small areas, and demand samples of time windows are generated in each area. The samples include the number of orders and order location distribution within each time window. For the demand samples in each region, the kernel density estimation method is used to construct the probability distribution function of the demand, and the historical data is modeled in a non-parametric way. The expression is: in, represents the probability density function of the demand in the time window, N is the number of historical demand samples, d is the demand value in the current time window, d i is the demand sample in the historical data, h is the bandwidth parameter, which controls the smoothness of the kernel function, K(u) is the kernel function, Among them, the linear adjustment factor is added To enhance the ability to estimate boundary demand; Combined with the real-time driver distribution data, the kernel density estimation results are combined with the real-time distribution information to dynamically adjust the weight of demand forecasting so that the demand forecasting can reflect the changes in the supply and demand relationship in real time. The expression is: in, is the modified demand probability density function; ΔS represents the real-time additional driver supply; ΔD represents the real-time demand change; According to the modified demand distribution function constructed Predict the demand of each region in the future time window, and calculate the expected value of the demand of each region as: Finally, the demand probability distribution and expected value of each area are output as input basis for subsequent driver allocation and rebalancing.

5. The hierarchical queue matching method based on driver position and waiting time according to claim 1 is characterized in that: When executing step S4, first, based on the historical order data and the real-time driver distribution, the improved kernel density estimation method is used to predict the demand distribution of each region in the current time window. Where i represents the region index and s represents the demand scenario.

6. The hierarchical queue matching method based on driver position and waiting time according to claim 1, characterized in that: In the first stage of driver allocation, first, the number of drivers to be allocated to each area is calculated using the stochastic programming model. To meet the optimization goal of forecast demand; The objective function is to minimize the supply-demand imbalance cost, which is defined as: Where: Q represents the penalty cost of insufficient supply; W represents the penalty cost of oversupply; R represents the set of all regions; The allocation constraints are: Among them, M t is the total number of available drivers in the current time window. After the demand actually occurs, the system reveals the actual demand in each area. Next, calculate the difference between supply and demand in each region And adjust driver reallocation decisions based on this.

7. The hierarchical queue matching method based on driver position and waiting time according to claim 1, characterized in that: When making rebalancing decisions in the second phase, first, dynamically optimize the number and paths of drivers to be redistributed based on the allocation results in the first phase and the actual demand revealed. The optimization goal is to minimize the rebalancing cost and the remaining supply-demand imbalance cost: Among them, C i，j represents the rebalancing cost from region i to region j, which is proportional to the distance between the two places; represents the number of drivers reallocated from area i to area j. The total number of drivers reallocated in the second phase cannot exceed the number of drivers allocated in the first phase: All variables must be non-negative integers: Number of drivers allocated in the first phase The routes and numbers of drivers reallocated in the second phase The system outputs the results to the dispatching system, and finally, based on the reallocation results, idle drivers are guided to the target area to optimize the supply-demand balance.

Images