A vehicle order dispatching method and dispatching system based on the supply-demand relationship of orders within a region

Abstract

The present invention discloses a vehicle order dispatching method based on the supply-demand relationship of orders within a region. The method mainly includes the following steps: Step 1: Divide the spatial region and predict the number of orders that will appear within the region in a future period of time; Step 2: Combine the predicted number of orders with the number of drivers, analyze the supply-demand relationship of orders in a specific region, and use queuing theory to analyze the waiting time of drivers after completing orders; Step 3: Combine the waiting time of drivers after completing orders with the order mileage to analyze the idle rate of orders, and dispatch orders for drivers according to the idle rate to ensure the maximization of the total income of all drivers. The present invention also discloses a driver order dispatching system based on the supply-demand relationship of orders within a region. When dispatching orders for drivers, the present invention takes into account the impact of future order information on current order dispatching and can maximize the total income of all drivers.

Description

Technical Field

[0001] The present invention belongs to the technical field of intelligent algorithms, and relates to a vehicle order dispatching method in crowdsourcing. More specifically, it relates to a vehicle order dispatching method and a dispatching system based on the supply-demand relationship of orders within a region. Background Art

[0002] With the large-scale popularization of the mobile Internet, people's travel patterns have gradually undergone earth-shaking changes. People can publish their order information on some taxi-hailing platforms, and then the platform dispatches drivers to the location of the passengers and sends the passengers to their destinations. Since passengers' patience is limited, they may cancel the order after waiting for a period of time and choose other means of transportation to reach their destinations. The platform needs to arrange a driver to handle the order before the passenger cancels the order. The platform needs to reasonably dispatch orders to maximize the total revenue of all drivers as much as possible.

[0003] In solving the order dispatching problem, the most common algorithm is the greedy algorithm. Each time when allocating an order, the greedy algorithm will select an optimal order under the current greedy strategy and assign it to the driver. For example, when we choose the nearest neighbor as the greedy strategy, whenever a driver finishes the current order and is about to pick up the next order or is just about to pick up an order, we assign the nearest order to him; whenever a new order appears, we select the nearest available driver to pick up the order. Similarly, when we use maximizing the reward as the greedy strategy, we always give priority to selecting the order with the highest reward each time. Batch processing is also one of the common methods for solving the taxi order dispatching problem. Different from the greedy algorithm, batch processing does not dispatch orders to drivers when a new driver or a new order appears. Instead, it divides continuous time into multiple time periods, and at the end of each time period, it matches the new orders and drivers that appear in this time period. The matching in each stage can use the bipartite graph matching method to select the order dispatching combination with the maximum revenue, and then assign the corresponding orders to each driver. This order dispatching method can achieve local optimality and ensure that the maximum revenue can be achieved in each time period.

[0004] The existing several types of order dispatching algorithms have the following deficiencies: First, these works only consider the current orders and do not combine future order information, resulting in the inability to obtain the global optimal solution; Second, the optimization goal of these works is to serve as many orders as possible. However, in fact, the revenues of different orders are different. Therefore, maximizing the number of served orders does not mean maximizing the total revenue of all drivers; Finally, these works do not consider the impact of the supply-demand relationship of orders within a region on order dispatching. When the destination of an order is a region with a supply-demand imbalance where the supply is less than the demand, after the driver completes the order, he will wait for a long time to pick up the next order, which indirectly reduces the value of the previous order. Summary of the Invention

[0005] To address the deficiencies of the prior art, the objective of the present invention is to provide a vehicle dispatching method based on the supply-demand relationship of orders within a region. By using a deep learning model to predict the number of orders in a certain region within a future time period, then analyzing the supply-demand relationship of orders in this region in combination with the calculated number of drivers, and subsequently obtaining the expected waiting time of drivers after completing an order through queuing theory analysis. Finally, analyze the idle rate of orders as an indicator of the supply-demand relationship within the region to guide the dispatching process.

[0006] The specific steps of the present invention are as follows:

[0007] Step (1): Divide the spatial region and predict the number of orders that will appear in the region within a future period of time;

[0008] Step (2): Combine the predicted number of orders with the number of drivers, analyze the supply-demand relationship of orders in a specific region, and use queuing theory to analyze the empty-load time of drivers after completing an order. The empty-load time refers to the time that a driver needs to wait before receiving the next order after completing an order;

[0009] Step (3): Combine the empty-load time of drivers after completing an order with the order mileage to analyze the idle rate of orders. The idle rate refers to the ratio of the empty-load time to the total time during the process of a driver picking up and dropping off passengers; Dispatch orders for drivers according to the idle rate to ensure the maximization of the total income of all drivers; The idle rate refers to the ratio of the empty-load time to the empty-load time + the driver's passenger-carrying time.

[0010] The present invention takes one day as a complete dispatching cycle, divides the 24 hours in a day into multiple time periods, and uniformly dispatches idle drivers at the end of each time period to handle the orders generated within the current time period.

[0011] In step (1) of the present invention, according to the pre-set grid parameter n, the space is divided into n×n rectangular grids of the same size, or n×n regions, ensuring that the side length of each grid or region is approximately 2 kilometers; Calculate the length and width of the space through the difference in longitude and latitude between the top-left vertex and the bottom-right vertex of the space, and then divide the length and width by the grid parameter n to calculate the length and width of each sub-region rectangle.

[0012] In step (1) of the present invention, in combination with the historical order data of the region, use a deep learning model to predict the number of orders in the region within a given time period; Obtain the number of drivers in this region during this period by adding the number of drivers in the current region to the number of drivers whose order destinations are the current region during the time period to be predicted.

[0013] In step (2) of the present invention, it is assumed that the driver generation rate is μ and the order generation rate is λ in a future period in the order destination area. If μ is greater than λ, it indicates that the supply of orders in this area is less than the demand, and the expected waiting time of the driver is relatively long; on the contrary, it indicates that the supply of orders is greater than the demand, and the expected waiting time of the driver is relatively short.

[0014] In step (3) of the present invention, based on the fact that maximizing the total income of drivers under the condition that the income per unit time remains unchanged when the driver picks up passengers is equivalent to minimizing the empty driving time of the driver, and is also equivalent to minimizing the idle rate of the driver, a driver order assignment method using a local search algorithm to minimize the total idle rate in each time period is proposed; compared with the existing greedy algorithm that simply faces the idle rate, the driver order assignment algorithm using the local search algorithm in the present invention can maximize the total income of all drivers.

[0015] The deep learning model used in the present invention is DeepST, and it is trained with the order distribution data of New York City; the order distribution data includes the following information: order id, order appearance time, order completion time, longitude, latitude, order income, and maximum order waiting time.

[0016] The local search algorithm proposed in the present invention will first calculate a candidate order assignment plan through a greedy algorithm facing the idle rate. Because dispatching the current order will affect the number of drivers in the area when the driver arrives at the order destination, which will ultimately lead to a deviation in the calculation of the idle rate of subsequent orders. Therefore, the local search algorithm will traverse each driver in the candidate order assignment plan and select an order with a smaller idle rate for it.

[0017] The greedy algorithm facing the idle rate will calculate the idle rate of each order, and then sort the orders in ascending order of the idle rate, and preferentially dispatch available drivers to handle orders with a small idle rate. During the process of dispatching drivers for orders, it is necessary to dispatch an order as close to it as possible.

[0018] In the present invention, for an order, an available driver represents a driver who is not serving any order at the current moment and is within a given distance from the target order; the given distance is set to be within two kilometers.

[0019] Based on the above method, the present invention also proposes a driver order assignment system based on the supply and demand relationship of orders within a region, and the system includes:

[0020] A region division module, which is used to divide the total prediction region;

[0021] A prediction module, which is used to predict the number of orders and the number of drivers in a certain future period in a region;

[0022] A demand analysis module, which is used to analyze the supply and demand relationship of orders in the area and predict the waiting time of drivers;

[0023] An order dispatching module, which uses an order dispatching algorithm to dispatch idle drivers to handle the orders generated in the current time period at the end of each time period.

[0024] The beneficial effects of the present invention compared with the prior art include: the queuing theory method used in the present invention is different from the method used in the prior art. The prior art can only consider the original number of drivers and the number of orders. In practice, for example, on rainy days, the number of orders is much larger than the number of drivers, which is not very feasible; while the present invention analyzes all possible situations and analyzes the prediction results of the waiting time, and takes into account the impact of future order information on the current order dispatching when dispatching orders for drivers; combined with the supply and demand relationship of the order destination, preferentially dispatch orders with a smaller idle rate for drivers to reduce the waiting time of drivers after completing orders. At the same time, the results show that analyzing the waiting time through queuing theory has high accuracy; the goal of minimizing the empty driving time of drivers is proposed to maximize the total income of all drivers. The results show that the local search algorithm proposed by the present invention has better effects than the existing algorithms. Description of the Drawings

[0025] Figure 1 It is a flowchart of the vehicle order dispatching method of the present invention.

[0026] Figure 2 It is a graph of the number of orders in each grid area of the present invention.

[0027] Figure 3 It is a structural diagram of the DeepST model of the present invention.

[0028] Figure 4 It is a graph of the order change rate of the present invention. Detailed Embodiments

[0029] Combined with the following specific embodiments and drawings, the present invention will be further described in detail. The processes, conditions, experimental methods, etc. for implementing the present invention, except for the specifically mentioned content below, are all common knowledge and well-known common sense in the art, and the present invention has no special restrictions.

[0030] In the present invention, "waiting time" and "empty driving time" have the same meaning.

[0031] The vehicle order dispatching method based on the supply and demand relationship of orders in the area proposed by the present invention includes the following steps:

[0032] Step (1), divide the spatial area and predict the number of orders that will appear in the area in the future for a period of time;

[0033] Step (2): Combine the predicted order quantity and the number of drivers, analyze the order supply-demand relationship in a specific area, and use queuing theory to analyze the empty-load time of drivers after completing an order. The empty-load time refers to the time required for a driver from completing an order to receiving the next order.

[0034] Step (3): Combine the empty-load time of drivers after completing an order and the order mileage to analyze the order idle rate. The idle rate refers to the ratio of the empty-load time to the total time during which a driver picks up and drops off passengers. Dispatch orders for drivers according to the idle rate to maximize the total income of all drivers.

[0035] A deep learning model is used to predict the order quantity in each divided area; according to the number of drivers and the order quantity, analyze the order supply-demand relationship in an area, and calculate the expected waiting time of drivers; combine the empty-load time of drivers to analyze the order idle rate; use a greedy algorithm oriented to the idle rate to obtain a candidate order dispatch plan; and use a local search algorithm to update the candidate order dispatch plan to obtain a better order dispatch plan.

[0036] Flow of vehicle order dispatch method in Embodiment 1

[0037] General flow of vehicle order dispatch method

[0038] As Figure 1 shown, the flow of the method is mainly divided into the following three stages: prediction stage, supply-demand relationship analysis, and order dispatch. The prediction stage needs to complete the prediction of the number of orders and drivers. In the supply-demand relationship analysis stage, the order supply-demand relationship in the future time period of this area will be analyzed by using the order quantity and the number of drivers predicted in the previous stage, and the expected waiting time of drivers in each area will be calculated. Finally, in the order dispatch stage, combine the mileage of the order itself and the expected waiting time of the driver at the order destination, analyze the idle rate of the current order, and then preferentially dispatch drivers to handle orders with a small idle rate to reduce the time required for the driver to wait for the next order after sending the passenger to the destination.

[0039] 1 Prediction stage

[0040] In the prediction stage, the present invention needs to predict the number of future orders and drivers. For order prediction, the present invention divides the time of a day into multiple time periods, then trains a deep learning model with the order data of the previous time period, and then uses the deep learning model to predict the number of orders in the next time period. Considering that human activities are uncertain, it is impossible to directly and accurately predict the location where each order appears in the next time period. Therefore, the present invention first divides the area into rectangular grids according to longitude and latitude, and then predicts the number of orders in units of areas. For driver number prediction, the present invention takes the number of currently idle drivers plus the number of drivers who can deliver orders to area j in time period i as the predicted number of drivers in area j in time period i in the future.

[0041] 2 Supply and demand relationship analysis

[0042] In the supply and demand relationship analysis stage, the present invention uses the number of drivers and orders in area j in time period i obtained in the prediction stage to analyze and obtain the driver generation rate λ and order generation rate μ in this area in this time period. According to the analysis, the waiting time of the driver after delivering the order to the destination area j in time period i can be obtained. When the supply of orders is greater than the demand in area j in time period i, the number of orders will be greater than the number of drivers, and the waiting time is less at this time, which means that the driver can find the next order after waiting for a short time. When the supply of orders is less than the demand in area j in time period i, the number of orders will be less than the number of drivers, and the waiting time is less at this time, which means that the driver can find the next order after waiting for a short time.

[0043] 3 Order assignment

[0044] In the order assignment stage, the present invention calculates the idle rate of the order by combining the mileage of the order itself and the waiting time of the driver at the order destination. The orders are sorted according to the idle rate, and the driver is first assigned to serve the order with a smaller idle rate, and a candidate order assignment plan is formed using the greedy algorithm. However, considering that the previous order assignment will affect the calculation of the subsequent order idle rate, it is necessary to traverse all the drivers in the candidate order assignment plan and select the order with the smallest idle rate within the acceptable range for each driver.

[0045] Embodiment 2 Related data of the vehicle order assignment method

[0046] The data involved in this method mainly includes: global spatial information, regional information, regional historical order data, order information, and driver information. Tables 1 to 5 detail the structural information of these 5 types of data.

[0047] Among them, the global spatial information refers to the space where the driver and the order are located. This method only processes the orders generated within this space, and at the same time, does not dispatch the driver to areas outside this space.

[0048] Table 1 Global Spatial Information

[0049]

[0050] The said area is the basic unit for predicting the order quantity and the driver quantity. Since it is impossible to accurately locate the position where each order is generated, the present invention divides the space into multiple areas, and then predicts the order quantity in units of areas.

[0051] Table 2 Area Information

[0052]

[0053] The historical order data of the said area is the data used by the present invention to train the deep learning model. Considering that it is impossible to accurately predict the order quantity generated in an area at a certain moment, the present invention divides a complete order dispatching cycle into multiple time periods as needed. The historical order data of the area contains the order data information of an area in a specific time period, and the historical information mainly includes the following parts: area number, time period, the order quantity that appears in this time period, the time period that disappears in this time period, and the weather condition at that time, etc.

[0054] Table 3 Area Historical Order Data

[0055]

[0056] The said order information is the ride information released by the passenger, including the time when the passenger releases the order, the location information of the passenger's current position, and the destination of the passenger. At the same time, considering that the patience of the passenger is limited, the passenger will cancel the order if the platform fails to process the order for a long time. Therefore, it is necessary to save the estimated cancellation time of the order. The driver must pick up the passenger before this time. The revenue is the remuneration that the driver will obtain after completing the order.

[0057] Table 4 Order Information

[0058]

[0059] The said driver information mainly includes the current position of the driver and the earliest time required for the driver to pick up the passenger. The present invention needs to maintain the earliest idle time of the driver at all times during the order dispatching process so as to dispatch orders to the idle drivers in a timely manner.

[0060] Table 5 Driver Information

[0061]

[0062]

[0063] System for Implementing Vehicle Order Assignment Method in Embodiment 3

[0064] An embodiment of the present invention describes a vehicle order assignment system, which includes: a regional division module, a prediction module, a demand analysis module, and an order assignment module.

[0065] The specific usage methods of each module are as follows:

[0066] 1. Regional Division Module

[0067] In an embodiment of the present invention, according to a preset parameter n, the space is divided into n×n rectangles of the same size. Calculate the difference between the longitude and latitude of the upper right corner and the longitude and latitude of the lower left corner, and then divide this difference by n to obtain the length and width of each rectangle. Divide the space according to the calculated length and width, and number the regions from top to bottom and from left to right. Taking each rectangle as a region and counting the number of orders in this region during a certain time period, the number of orders in a certain region in a future time period can be obtained.

[0068] When the order distribution during a certain idle time period is as Figure 2 shown (each gray dot represents an order that appears during this time period). Assuming that n is set to 4, the space can be divided into a 4×4 region as Figure 2 shown, numbered 1, 2, 3... respectively. Then count the number of orders generated in each region during this time period. As Figure 2 shown, it can be seen that the number of orders in region 5 is 3, while the number of orders in region 7 is 0.

[0069] 2. Prediction Module

[0070] In the order assignment system of the present invention, it is necessary to predict the order demand in each region within a given time period. For drivers who rejoin, it is easy to estimate when they will deliver the order to the destination based on their tasks and the time required to drop off passengers. In practice, due to the uncertainty of individual user behavior, it is difficult to predict the exact location and timestamp of a specific passenger or driver. In order to utilize the distribution of passenger demand and supply, the present invention predicts the number of passengers in a given region (i.e., the spatial range of the region, such as a square region or a hexagonal region) within a given time period (i.e., the next 5 minutes). Some deep learning models can be applied to predict the passenger demand in a specific region and time period. For example, the DeepST model can be used to predict the number of orders in region a in a future time period k Predicted number of orders The model architecture of DeepST is as Figure 3As shown, it predicts the order information for the next time period by inputting the historical order quantity in this area and the external weather information. It extracts trend information, periodic information, and proximity information from the historical order data, then integrates these three pieces of information with the external weather information to finally obtain the prediction result. The present invention uses the order data of New York City to train this model, then predicts the order quantity in each area for the next time period, and combines the set R of unassigned orders in the current time period k , to obtain the available orders for the next time period.

[0071] The number of drivers can be inferred through the expected trips of the drivers. The present invention can obtain the set D of currently idle drivers in area a by analyzing Table 5 k , and obtain the current task list of the drivers. From the task list, it can be analyzed that the set of drivers arriving in area a before time period i k , k Table 6 shows an example of a driver's current task list. From the table, it can be found that the drivers who can end the order with a time period of 5 and a destination area of 2 are 1, 2, 3, and 5. Assume that the size of the set D k of currently idle drivers is 2 at this time, then the number of idle drivers in area 2 at time period 5 can be obtained to be approximately 6.

[0072] Table 6 Driver's Current Task List

[0073] Driver ID Departure time period Latest end time period Destination area ID 1 1 4 2 2 2 5 2 3 1 5 2 4 3 4 1 5 3 5 2

[0074] 3. Demand Analysis Module

[0075] The present invention will analyze and obtain the expected waiting time of the drivers in area a k at time period i through the predicted order quantity and the number of drivers. The present invention obtains the generation rate μ of the drivers and the generation rate λ of the orders by dividing the number of drivers and orders generated in this time period by the time period length t c . The calculation method of the generation rate is as follows:

[0076]

[0077]

[0078] At the same time, considering that the patience of passengers is limited, the orders will decrease at a specific rate, and this rate is related to the current order quantity n and is related to the generation rate μ of the drivers. The rate of order decrease is β is a parameter statistically obtained from the data. The larger β is, the less patient the passengers are, and in this case, the rate of order decrease is larger. The β selected by the present invention in the experiment is 2. Figure 4 ​shows the change rate of the order quantity, and the order quantity changes from n to n + 1 at a rate of λ n and from n + 1 to n at a rate of μ n+1 . Where λ n is always equal to λ, and when n > 0, μ n = μ + r(n), when n < 0, μ n = μ. Assume that the probability of the order quantity being n is p n . In the case of long-term statistics, the number of times the order quantity changes from n to n + 1 is basically the same as the number of times it changes from n + 1 to n. Therefore, there is μ n p n = λ n-1 p n-1 , and Finally, we can get

[0079]

[0080] Next, consider the following three cases: μ > λ, μ < λ, μ = λ.

[0081] 1. When μ < λ, from the above analysis, the value of p0 can be solved as follows:

[0082]

[0083] When the current order quantity n is greater than 0, each arriving driver can be assigned an order in a very short time. Therefore, the waiting time at this time is 0, and when n is equal to 0, the driver needs to wait time to be assigned an order. And when n is less than 0, at this time |n| represents the number of drivers waiting. Therefore, the newly added driver needs to wait time. Therefore, in this embodiment, the waiting time ET(λ, μ) can be estimated by the following formula.

[0084]

[0085] 2. When μ > λ, the driver generation rate is greater than the order generation rate. However, considering that the number of drivers that a region can accommodate is limited and cannot increase indefinitely. Therefore, in this embodiment, it is assumed that the maximum number of drivers that this region can accommodate is K. Therefore, we can assume that when n < -K, p n is 0. At the same time, denote as θ.

[0086] Then it can be analyzed that

[0087]

[0088] Therefore, it can be solved that

[0089]

[0090] And the waiting time at this time can also be estimated by the same method, and the results are as follows:

[0091]

[0092] 3. When μ = λ, it is also assumed that this area can accommodate at most K drivers. Therefore, it can be analyzed that the waiting time is

[0093]

[0094] At this time,

[0095] In the experiment, according to historical data, the hyperparameters β and K are set to 2 and 256 respectively.

[0096] 4. Order Assignment Module

[0097] The order assignment module in the embodiments of the present invention includes an algorithm framework based on batch processing as shown in Algorithm 1. The method of the present invention does not assign orders to drivers in real time, but assigns orders to all idle orders generated in this batch after each time period ends. The length of a time period is set to Δ, and the value of Δ is generally 30 seconds to 1 minute. Specifically, the following processing is performed on the orders generated in time period i. First, retrieve the order set R k of each area a k and the set D k of available drivers. The order R k here includes the orders in area a k that were not assigned to drivers in the previous time period and the orders that newly appeared in this time period. In addition, the available drivers D k include the drivers who were not assigned any passengers in the previous time period, and the drivers who re-joined the platform in area a k after completing the assigned tasks before the current time period. In order to estimate the passenger arrival rate and the number of re-joined drivers in the current scheduling period [t, t + t c , the present invention also predicts the number of upcoming passengers k and the number of re-joined drivers in area a Then, the vehicle order assignment method of the present invention is used to achieve the pairing of passengers and drivers to maximize the total revenue of the platform in the current scheduling period [t, t + t c .

[0098] Algorithm 1 Algorithm Framework Based on Batch Processing

[0099]

[0100] Description of Vehicle Order Assignment Method in Embodiment 4

[0101] 4.1 Greedy Algorithm Oriented to Idle Rate

[0102] In this embodiment, it is assumed that the generation rate of drivers at destination a k is μ (k) , and the generation rate of orders is λ (k) . Then, through the analysis in Embodiment 3, the expected waiting time of drivers at a k is ET(λ (k) , μ (k) ). Subsequently, the idle rate of driver d j serving order r i is as follows:

[0103]

[0104] where cost(s i , e i ) represents the time cost required from the starting position to the end of the order. The greedy algorithm oriented to idle rate minimizes the empty-load time of drivers by selecting orders with relatively small idle rates as much as possible. Algorithm 2 shows the method of the greedy algorithm for selecting suitable orders for drivers.

[0105] Algorithm 2 Greedy Algorithm Oriented to Idle Rate

[0106]

[0107] 4.2 Local Search Algorithm

[0108] In Algorithm 2, the order generation rate λ and driver generation rate μ of all regions are first calculated. Then, the idle rate of each order is calculated. However, in fact, the assignment of the current order will affect the calculation of the subsequent idle rate. Therefore, the present invention uses a local search algorithm. First, a candidate order assignment set is generated through Algorithm 2, and then each driver in the order assignment set is traversed to select a better order for each driver. The specific algorithm process is as follows:

[0109] Algorithm 3 Local Search Algorithm

[0110]

[0111] 4.3 Demonstration of Technical Effects

[0112] The present invention realizes an order assignment method based on supply and demand analysis through a local search algorithm. By predicting the future number of orders and drivers, the waiting time of drivers in a region is obtained, thereby reflecting the supply and demand relationship of orders in the region.

[0113] In terms of order prediction, this method uses the order data from January to May 2013 in New York City as a dataset to train the model. First, the original space is divided into 256 regions, and the historical order information of each region is obtained through the region division module. The specific configuration of the dataset is shown in Table 7:

[0114] Table 7 Dataset Configuration

[0115]

[0116]

[0117] Four order prediction methods are evaluated, namely: HA (using the mean of historical orders as the prediction result for the next time period), LR (linear regression), GBRT (gradient regression tree), and DeepST-GC (a prediction method based on deep learning). As can be seen from Table 8, the deviation of using DeepST-GC to predict the order quantity is less than that of other prediction algorithms, and the prediction deviation for each region and each time period is only about 15 orders. This shows that the order dispatching method proposed by the present invention has good results in realizing order prediction by using deep learning methods.

[0118] Table 8 Prediction Effect Diagram

[0119] RMSE (%) RealRMSE (s) DeepST-GC 2.30 15.03 HA 7.46 48.21 LR 3.40 21.66 GBRT 2.74 17.67

[0120] After obtaining the predicted order quantity, the method of the present invention combines the quantity calculated by the driver to analyze the supply-demand relationship of orders within the region. The waiting time of the driver is calculated by simulating the actual waiting time of the driver during the order dispatching process. In the embodiments of the present invention, the number of drivers is set to be from 1000 to 10000 respectively, and the deviation of waiting time prediction is analyzed through experiments. As can be found from Table 9, it is effective for this method to use queuing theory to analyze the waiting time of the driver. When the number of drivers reaches 3000, the prediction deviation of the waiting time reaches the minimum. When the number of drivers is relatively small, the waiting time of each driver is very small, but they have to wait until the end of the time period before they can be assigned orders, so the waiting time is relatively large.

[0121] Table 9 Waiting Time Prediction Results

[0122] Number of drivers MAE (s) RMSE (%) RealRMSE (s) 1000 2.48 4.91 12.14 2000 2.11 4.97 9.33 3000 1.83 4.72 5.02 4000 1.97 5.11 7.04 6000 2.17 5.42 9.23 8000 2.51 5.92 12.14 10000 5.33 6.72 41.92

[0123] Finally, based on the order supply-demand relationship in the combination area, this method proposes an algorithm for matching suitable orders for drivers to ensure that the driver's idle time can be minimized and the revenue can be maximized. The comparison algorithms used in the embodiments of the present invention are a random algorithm for allocating orders to drivers at the end of each time period and a greedy algorithm for the idle rate. In the experiment, the maximum waiting time for orders is set between two minutes and five minutes, and then the time period length is set to 5 seconds, and a table of the relationship between the order allocation effect and the number of drivers can be obtained. As can be seen from Table 10, the total revenue of the local search algorithm performs the best among all algorithms. At the same time, the order allocation revenue is positively correlated with the number of drivers. This is because the more drivers there are, the more orders can be processed at the same time. The local search algorithm, because it can calculate a more accurate waiting time, its order allocation effect is better than the greedy algorithm for the idle rate. It can be seen from the table that a larger number of served orders does not necessarily mean a larger final total revenue. This is because the larger the number of served orders, the more times the driver waits idle for the next order, resulting in an increase in the driver's idle time. Eventually, the revenue will be less than that of an algorithm with fewer served orders.

[0124] Table 10 Order Allocation Revenue

[0125]

[0126] The existing greedy algorithm for the idle rate is the same as that in Embodiment 4.1, and it can only consider the original situation of the number of drivers and the number of orders. In practice, for example, on rainy days, the number of orders is much larger than the number of drivers, and it does not have strong feasibility; while the local search algorithm proposed by the present invention analyzes all possible situations and analyzes the prediction results of the waiting time, and takes into account the impact of future order information on the current order allocation when allocating orders to drivers; combined with the supply-demand relationship of the order destination, preferentially allocate orders with a smaller idle rate to drivers, reduce the waiting time of drivers after completing orders, and at the same time the results show that analyzing the waiting time through queuing theory has high accuracy; propose to minimize the driver's idle time as the optimization goal to maximize the total revenue of all drivers. The results in Table 10 show that the local search algorithm proposed by the present invention has a better effect than the existing algorithms including the greedy algorithm for the idle rate and the random algorithm.

[0127] The protection scope of the present invention is not limited to the above embodiments. Without departing from the spirit and scope of the inventive concept, all changes and advantages that those skilled in the art can think of are included in the present invention, and the scope of protection is defined by the appended claims.

Claims

1. A vehicle dispatching method based on the supply-demand relationship of orders within a region, characterized in that It includes the following steps: Step (1): Divide the spatial area and predict the number of orders that will appear in the area in the next period of time; Step (2): Combine the predicted number of orders with the number of drivers, analyze the order supply-demand relationship in a specific area, and use queuing theory to analyze the empty-load time of the driver after completing an order. The empty-load time refers to the time required for the driver from completing an order to receiving the next order; Step (3): Combine the empty-load time of the driver after completing an order with the order mileage to analyze the order idle rate. The idle rate refers to the ratio of the empty-load time to the total time during which the driver picks up and drops off passengers; Dispatch orders for the driver according to the idle rate to ensure the maximization of the total income of all drivers; In step (3), the maximization of the total driver income means maximizing the total income of the driver when the income per unit time is constant during the driver's pick-up and drop-off of passengers, which is equivalent to minimizing the idle rate; The order dispatching algorithm for minimizing the idle rate is a local search algorithm; The local search algorithm first calculates a candidate order dispatching plan through a greedy algorithm oriented to the idle rate; Since dispatching the current order will affect the number of drivers in the area when the driver arrives at the order destination, ultimately resulting in a deviation in the calculation of the idle rate of subsequent orders; Therefore, the local search algorithm will traverse each driver in the candidate order dispatching plan and select an order with a smaller idle rate for it; The greedy algorithm oriented to the idle rate means calculating the idle rate for each order, and then sorting the orders in ascending order of the idle rate, and preferentially dispatching available drivers to handle orders with a smaller idle rate; During the process of dispatching a driver for an order, an order as close as possible to it needs to be dispatched for the order; 2. The vehicle order dispatching method based on the supply and demand relationship of orders within a region according to claim 1, wherein The vehicle order dispatching method takes one day as a complete order dispatching cycle, divides the 24 hours in a day into multiple time periods, and uniformly dispatches idle drivers at the end of each time period to handle the orders generated during the current time period; 3. The vehicle order dispatch method based on the supply-demand relationship of orders within a region according to claim 1, wherein In step (1), the division of the spatial area means dividing the space into n×n rectangular grids or areas of the same size according to the pre-set grid parameter n, so that the side length of each grid or area is 2 kilometers; Calculate the length and width of the space through the difference in the longitude and latitude of the upper left vertex and the lower right vertex of the space, and then divide the length and width by the grid parameter n to calculate the length and width of each sub-region rectangle; 4. The vehicle order dispatching method based on the supply-demand relationship of orders within a region according to claim 1, wherein In step (1), the prediction of the number of orders means combining the historical order data of the area and using a deep learning model to predict the number of orders in the area within a given time period; Obtain the number of drivers in the area during this period by adding the number of drivers in the current area to the number of drivers whose order destinations are in the current area during the time period to be predicted; 5. The vehicle order dispatching method based on the supply-demand relationship of orders within a region according to claim 1, wherein In step (2), the analysis of the supply-demand relationship and the driver waiting time means assuming that the generation rate of drivers in the order destination area in the next period of time is μ and the order generation rate is λ; If μ is greater than λ, it means that the supply of orders in this area is less than the demand, and the expected waiting time of the driver is longer; Otherwise, it means that the supply of orders is greater than the demand, and the expected waiting time of the driver is shorter; 6. The vehicle order dispatch method based on the supply and demand relationship of orders within a region according to claim 4, wherein The deep learning model is DeepST, which is trained with the order distribution data of the city; the order distribution data includes the following information: order ID, order appearance time, order completion time, longitude, latitude, order revenue, and maximum order waiting time.

7. The vehicle dispatching method based on the supply-demand relationship of orders within a region according to claim 1, wherein The available drivers represent the drivers who are not serving any orders at the current moment and are within a given distance from the target order; the given distance is set to be within two kilometers.

8. A vehicle dispatch system for implementing the vehicle dispatch method according to any one of claims 1-7, characterized in that, The system includes: A region division module, which is used to divide the total prediction region; A prediction module, which is used to predict the number of orders and the number of drivers in a certain future time period in a region; A demand analysis module, which is used to analyze the supply and demand relationship of orders in the region and predict the waiting time of drivers; An order dispatching module, which uses an order dispatching algorithm to dispatch idle drivers to handle the orders generated in the current time period at the end of each time period.

Images