Online car-hailing order dispatching method considering driver characteristics based on artificial intelligence

Abstract

The invention discloses an online car-hailing order dispatching method considering driver characteristics based on artificial intelligence, and relates to the technical field of intelligent scheduling of online car-hailing, and the method comprises the steps: associating vehicle track data preprocessed based on a large language model with order data, and obtaining driver order receiving data; in consideration of driver characteristics, a driver order receiving preference prediction model is constructed by utilizing machine learning, parameter calibration is completed, a fuzzy control model among driver age, working duration and order receiving reaction time is established in combination with a fuzzy reasoning technology, and an influence factor of the driver order receiving reaction time is calculated; building a supply-demand relationship evaluation index, judging a peak period and a flat period according to a supply-demand relationship, flexibly converting an order sending mode, and calculating order receiving reaction time of a driver under different order sending strategies; a utility function is constructed by taking minimization of a vehicle scheduling distance and order transaction time as targets, and an order dispatching scheme is optimized by using a hill-climbing algorithm HC. Based on the artificial intelligence technology, the platform order sending efficiency and the user satisfaction degree are remarkably improved.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent dispatching of online car-hailing vehicles, and in particular to an online car-hailing vehicle dispatching method based on artificial intelligence and taking into account driver characteristics. Background Art

[0002] With the rapid development of the online car-hailing industry, order allocation efficiency has become a core issue affecting user experience and platform operations. Improving the quality of order allocation requires optimization from both the order allocation target and the order allocation strategy.

[0003] In terms of dispatching objectives, traditional dispatching technologies are mainly based on simple rules or single optimization objectives. For example, by focusing on the needs of a single group, dispatching orders in a way that minimizes the driver's dispatch distance, minimizes the passenger's waiting time, or maximizes the platform's order completion rate. However, such methods are difficult to achieve an effective balance between drivers, passengers, and platforms, resulting in the overall efficiency and adaptability of the dispatching system being limited. Although the rise of artificial intelligence technology has promoted the development of online car-hailing dispatching research in the direction of multi-objective optimization, existing multi-objective optimization technologies are mostly limited to the deployment and algorithm improvement of the platform dispatching system, and have not yet considered the impact of the supply-side, i.e., driver characteristics, on dispatching efficiency.

[0004] In terms of dispatching strategies, the existing technical strategies for promotion and implementation are mainly divided into two types: assignment mode and grabbing mode. Among them, the assignment mode means that the driver can only accept a single order assigned by the system at the same time, while the grabbing mode allows the driver to freely grab orders through the multi-order interface at the same time. However, a single mode is often difficult to achieve multiple dispatching goals at the same time. In addition, the existing methods ignore the differences in market conditions and driver behavior in different time periods, which may lead to order backlogs and reduced driver order acceptance efficiency during peak periods, while resource waste may occur during off-peak periods. Therefore, there is an urgent need for a hybrid dispatching strategy that can adapt to the needs of multiple groups during specific time periods. Summary of the invention

[0005] The technical problem to be solved by the present invention is to overcome the shortcomings of the prior art and provide an online car-hailing dispatching method based on artificial intelligence that takes into account driver characteristics. By considering the order acceptance behavior characteristics and reaction time characteristics in the driver's characteristics, machine learning is used to predict the driver's order acceptance probability, and fuzzy control technology is used to calculate the reaction time influencing factors, and the impact of different supply and demand relationships, dispatching strategies and individual physiological factors of drivers on reaction time are comprehensively considered; based on the hill climbing algorithm (HC) under artificial intelligence technology, the optimal dispatching plan is searched under the multi-objective constraints of reducing vehicle dispatching distance, increasing the platform order transaction rate, that is, shortening the passenger waiting time; by dynamically adjusting the target weights and parameter settings, it adapts to the needs of different scenarios, achieves an effective balance of the interests of the driver, passengers and the platform, and improves dispatching efficiency and user experience.

[0006] The present invention adopts the following technical solutions to solve the above technical problems:

[0007] According to the present invention, an online car-hailing dispatching method based on artificial intelligence and considering driver characteristics comprises:

[0008] Step A: pre-processing vehicle trajectory data and order data based on the large language model;

[0009] Associating the preprocessed vehicle trajectory data with the order data to obtain the driver's order acceptance data;

[0010] Step B: Considering the characteristics of the driver's order-taking behavior and reaction time, based on the driver's order-taking data, machine learning is used to build a driver's order-taking preference prediction model and complete parameter calibration. At the same time, fuzzy reasoning technology is combined to establish a fuzzy control model between the driver's age, working hours and order-taking reaction time by setting fuzzy rules;

[0011] The real-time orders are processed through the driver's order acceptance preference prediction model, and the driver's acceptance probability value for the new order is output;

[0012] Based on the fuzzy control model, the factors affecting the response time of different drivers at different times are calculated;

[0013] Step C: according to the number D of idle drivers within the preset time interval Δt i 、Current order quantity O c , and predicted number of new orders O f , construct the supply and demand evaluation index E s Determine the market supply and demand status, divide the peak period and the off-peak period according to the market supply and demand relationship, and adjust the order dispatching strategy in real time;

[0014] The driver's reaction time for accepting orders is calculated by combining the supply and demand relationship in different markets, the psychological differences of drivers in accepting orders under dispatching strategies, the probability of drivers accepting new orders, and the factors affecting the reaction time of different drivers at different times.

[0015] Step D: Calculate the order completion time E[t o ], in order to minimize the vehicle dispatch distance E[d o ] and order execution time E[t o ] as the goal, construct a multi-objective utility function;

[0016] Based on the hill climbing algorithm HC in artificial intelligence technology, the dispatching plan is optimized to find the average multi-objective utility function value The smallest dispatch plan.

[0017] As a further optimization solution of the online car-hailing order dispatching method considering driver characteristics based on artificial intelligence according to the present invention, in step A, the preprocessing includes outlier deletion, missing value filling, and effective trajectory extraction; among them, outlier deletion refers to automatically detecting and deleting those with position drift and abnormal speed in the order data and vehicle trajectory data based on semantic analysis and spatio-temporal pattern recognition; missing value filling refers to using spatio-temporal reasoning and context data completion technology to complete the missing time or position points in the order data and vehicle trajectory data; effective trajectory extraction refers to extracting the effective pick-up and drop-off trajectories of the vehicle through the order time period.

[0018] As a further optimization solution of the online car-hailing order dispatching method considering driver characteristics based on artificial intelligence according to the present invention, in step A, the association between the preprocessed vehicle trajectory data and the order data includes: associating the personal information of the driver who receives the order with the detailed information of the order through the ID number to obtain the driver order-receiving data; the driver order-receiving data includes order feature data and driver attribute data, the order feature data includes the order start position, end position, price, travel distance, and the time period where the order is located; the driver attribute data includes the initial position of the vehicle, the driver's order-receiving frequency, and the dispatching distance.

[0019] As a further optimization solution of the online car-hailing order dispatching method considering driver characteristics based on artificial intelligence according to the present invention, the process of establishing a fuzzy control model includes:

[0020] Step B1, defining the input and output variables of the fuzzy control model:

[0021] The values of the control input and output variables are between [0, 1];

[0022] Taking the driver's age and working hours as input variables, which are used to characterize the influence of the driver's age and fatigue level on the order-taking response time. Among them, the working hours are used to represent the driver's fatigue level; the value range of the driver's age is [18, 60] years old, which is divided into three age groups: young [18, 35], prime [35, 45], and middle-aged [45, 60]. The corresponding fuzzy control input intervals for these three age groups are [0, 0.42], [0.42, 0.65], and [0.65, 1] respectively; the value range of the working hours is [0, 18] hours, and the fatigue level is divided into three groups according to the duration: mild fatigue [0, 1.8], moderate fatigue [1.8, 10.8], and severe fatigue [10.8, 18]. The corresponding fuzzy control input intervals for these three fatigue levels are [0, 0.1], [0.1, 0.6], and [0.6, 1] respectively; the response time impact factor is divided into seven intervals: very small interval [0, 0.15], small interval [0.15, 0.30], slightly small interval [0.30, 0.45], medium interval [0.45, 0.60], slightly large interval [0.6, 0.75], large interval [0.75, 0.9], and very large interval [0.9, 1].

[0023] Taking the order-taking response time impact factor as the output variable, which is used to represent the increment of the benchmark response time due to driver heterogeneity;

[0024] Step B2: Determine the input and output membership functions of the fuzzy control model according to the membership relationship among the driver's age, working hours, and response time impact factor;

[0025] Step B3: Design fuzzy rules, which include the first rule - the ninth rule. Among them, the first rule: when the driver's age belongs to the young group and the fatigue level is mild fatigue, the response time impact factor belongs to the very small interval; the second rule: when the driver's age belongs to the young group and the fatigue level is moderate fatigue, the response time impact factor belongs to the small interval; the third rule: when the driver's age belongs to the young group and the fatigue level is severe fatigue, the response time impact factor belongs to the slightly small interval; the fourth rule: when the driver's age belongs to the prime group and the fatigue level is mild fatigue, the response time impact factor belongs to the small interval; the fifth rule: when the driver's age belongs to the prime group and the fatigue level is moderate fatigue, the response time impact factor belongs to the medium interval; the sixth rule: when the driver's age belongs to the prime group and the fatigue level is severe fatigue, the response time impact factor belongs to the slightly large interval; the seventh rule: when the driver's age belongs to the middle-aged group and the fatigue level is mild fatigue, the response time impact factor belongs to the slightly small interval; the eighth rule: when the driver's age belongs to the middle-aged group and the fatigue level is moderate fatigue, the response time impact factor belongs to the large interval; the ninth rule: when the driver's age belongs to the middle-aged group and the fatigue level is severe fatigue, the response time impact factor belongs to the very large interval;

[0026] Step B4: Calculate through fuzzy inference and output the influencing factor of the driver's order acceptance reaction time.

[0027] As a further optimization scheme of the online car-hailing order assignment method considering driver characteristics based on artificial intelligence of the present invention, the order assignment strategies include an assignment mode, a grab-and-assign mode, and a mixed mode. Among them, the assignment mode means that the driver can only accept a single order assigned by the online car-hailing order assignment platform at the same moment. The grab-and-assign mode means that the driver can freely grab orders on the multi-order interface at the same time. The mixed mode means that the online car-hailing order assignment platform screens out multiple drivers for each order, and multiple drivers simultaneously perform the order grabbing operation on this order;

[0028] The driver's order acceptance reaction time includes the driver's order acceptance decision time and the driver's order acceptance action time.

[0029] As a further optimization scheme of the online car-hailing order assignment method considering driver characteristics based on artificial intelligence of the present invention, construct a supply-demand evaluation index E s Judge the market supply-demand state, divide the peak period and the off-peak period according to the supply-demand relationship, and adjust the order assignment strategy in real time; including:

[0030] Step C1: Calculate the supply-demand evaluation index E s ;

[0031]

[0032] Step C2: Convert the order assignment strategy in real time: initially adopt the mixed mode, and set the current time period to be the off-peak period; set a time interval Δt, and calculate E every Δt interval s ; Set the threshold μ of E s When the duration of the state where E s is greater than μ reaches the preset time ΔT, and ΔT is the longest time that E s can exceed the threshold μ in the off-peak period state, it is judged as the peak period, and automatically switch to the assignment mode, and stop calculating E in this assignment mode s ; Set the residence time C in the assignment mode, and return to the mixed mode to re-loop after the residence time C ends.

[0033] As a further optimization scheme of the online car-hailing order assignment method considering driver characteristics based on artificial intelligence of the present invention, in step C, two improved AFT models are used to calculate the driver's order acceptance reaction time respectively; specifically as follows:

[0034] Calculate the order acceptance reaction time t of the i-th driver accepting the o-th order in different periods o,i :

[0035] In the peak period, the platform adopts a one-to-one assignment form, to,i It follows a Gaussian distribution, and the specific formula is as follows:

[0036] t o,i =(1 + v o,i ) * [T s * exp(-α * (p o.i - 0.5)) + T r

[0037] During the flat peak period and the platform stage, a one-to-many mixed order dispatching form is adopted, and t o,i follows an exponential decay, and the specific formula is as follows:

[0038] t o,i =(1 + v o,i ) * [T s * exp(-α * p o.i ) + T r

[0039] Among them, v o,i is the response time impact factor for the i-th driver to accept the o-th order, and p o,i is the order acceptance probability of the i-th driver for the o-th order; T s is the maximum reference time for the driver to make an order acceptance decision; α represents the importance degree of p o,i ; T r is the time from when the driver makes an order acceptance decision until the brain sends an order acceptance action instruction to the fingertip to click the screen to accept the order;

[0040] A driver can only receive one order at the same time.

[0041] As a further optimization scheme of the online car-hailing order dispatching method considering driver characteristics based on artificial intelligence according to the present invention, in step D, a multi-objective utility function is constructed, including:

[0042] Step D1: For the o-th order at the current moment, select I drivers for order dispatching to determine the order dispatching driver set; where I >= 1;

[0043] Step D2: Considering the possibility of driver order rejection, calculate the vehicle dispatching distance E[d o corresponding to the o-th order;

[0044]

[0045] Among them, d o,j is the dispatching distance between the o-th order and the j-th driver, and p o,j is the order acceptance probability of the j-th driver for the o-th order;

[0046] Step D3: Considering the possibility of driver order rejection, calculate the transaction time E[t​​o ;

[0047]

[0048] where t o,j is the order acceptance response time for the j-th driver to accept the o-th order;

[0049] Step D4. Calculate the average multi-objective utility function value of all orders at the current moment

[0050] where N is the number of orders at the current moment; D max is the farthest distance for order assignment; T max is the maximum waiting time of the passenger.

[0051] As a further optimization scheme of the online car-hailing order assignment method considering driver characteristics based on artificial intelligence of the present invention, based on the hill climbing algorithm HC in artificial intelligence technology, optimize the order assignment scheme to find the order assignment scheme with the smallest average multi-objective utility function value, including: Step D51. For the o-th order at the current moment, calculate the order acceptance probability value of idle drivers for the new order, sort them from largest to smallest according to the order acceptance probability, and select the first I drivers as the initial order assignment objects to form an initial order assignment scheme;

[0052] Step D52. Generate a new order assignment scheme by adjusting the matching relationship between orders and drivers; the specific matching relationship is as follows:

[0053] Explore the neighborhood solution space of the current order assignment scheme by exchanging the matching combinations of orders and drivers to find the optimal matching method;

[0054] Step D53. Calculate the corresponding

[0055] of the new order assignment scheme and compare it with the current order assignment scheme; if the corresponding to the new order assignment scheme is less than the corresponding to the current order assignment scheme then replace the current order assignment scheme with the new order assignment scheme; otherwise, retain the current order assignment scheme and continue to search the neighborhood solution space;

[0056] Step D54. Repeat steps D52 to D53 until the of the order assignment scheme no longer improves. At this time, the obtained order assignment scheme is: the order assignment scheme with the smallest average multi-objective utility function value Step D54. Repeat steps D52 to D53 until the

[0057] Compared with the prior art, the present invention adopts the above technical solutions and has the following technical effects:

[0058] (1) The present invention is based on artificial intelligence technology, uses a large language model for data processing, predicts the probability of drivers accepting orders through machine learning, and uses the hill climbing algorithm HC to optimize the dispatching plan, thereby realizing intelligent dispatching under multi-objective constraints, significantly improving vehicle dispatch efficiency and order completion rate, that is, the waiting time for passengers to receive orders, while balancing the interests of the drivers, passengers and the platform.

[0059] (2) This invention introduces the concept of reaction time for accepting orders under the influence of driver characteristics into the field of online car-hailing dispatch optimization for the first time, and comprehensively considers the time consumed by the driver from the appearance of an order to clicking the screen to accept or reject the order. By analyzing the impact of different ages, working hours and periods on drivers, the accuracy and scientificity of time calculation in the dispatch process are further improved.

[0060] (3) The present invention proposes an innovative dispatch strategy conversion method that can better adapt to the differences in driver behavior caused by changes in supply and demand in different time periods. This method optimizes vehicle resource utilization by dynamically converting dispatch modes, significantly increasing order volume.

[0061] (4) The multi-objective balancing method proposed in the present invention has wide applicability and can be extended to multi-objective balancing problems in other scenarios, providing new technical ideas and methodological support for solving complex optimization problems. BRIEF DESCRIPTION OF THE DRAWINGS

[0062] Figure 1 It is a flow chart of a method for dispatching online car-hailing orders based on artificial intelligence and taking into account driver characteristics of the present invention;

[0063] Figure 2 The present invention relates to a principle diagram of a dispatching strategy; wherein (a) is a dispatching mode, and (b) is a mixed mode;

[0064] Figure 3 It is a flow chart of the hill climbing algorithm of the present invention. DETAILED DESCRIPTION

[0065] In order to further understand the present invention, preferred embodiments of the present invention are described below in conjunction with examples. However, it should be understood that these descriptions are only for further illustrating the features and advantages of the present invention, rather than limiting the claims of the present invention.

[0066] The following embodiments take Chengdu as an example, but the application of the present invention is not limited to the scope of the embodiments. The combination of different embodiments, the mutual replacement of some technical features in different embodiments, and the mutual replacement of the same or similar prior art means with some technical features in the embodiments are also within the scope of the description and protection of the present invention.

[0067] This embodiment provides a ride-hailing order dispatching method based on artificial intelligence that takes into account driver characteristics, with reference to Figure 1 as shown below, including the following steps:

[0068] A. Association of vehicle trajectory and order information: Based on the large language model, perform data preprocessing: clean missing values and outliers in the data, and extract the passenger-carrying trajectory segments in the vehicle data; match the processed vehicle and order data, and associate the personal information of the driver taking the order with the detailed information of the order;

[0069] Based on the large language model, perform data preprocessing; associate the preprocessed vehicle trajectory data with the order data to obtain driver order-taking data; specifically including the following steps:

[0070] Step A1) Data preprocessing: mainly includes outlier deletion, missing value filling, and effective trajectory extraction; based on semantic analysis and spatio-temporal pattern recognition, automatically detect and delete the position drifts and speed anomalies in the order data and vehicle trajectory data; use spatio-temporal reasoning and context data completion technology to complete the missing time or position points in the data; extract the effective pick-up and drop-off trajectories of the vehicle through the order time period.

[0071] Step A2) Association between data: Associate the personal information of the driver taking the order with the detailed information of the order through the ID number to obtain driver order-taking data.

[0072] Among them, the driver order-taking data includes order feature data and driver attribute data. The order feature data includes: order start position, end position, price, travel distance, and order time period; the driver attribute data includes: vehicle initial position, driver order-taking frequency, and dispatching distance;

[0073] In this embodiment, some of the driver order-taking data covered after association is shown in Table 1:

[0074] Table 1 Partial associated data

[0075] Order ID Start Billing Time End Billing Time Pick-up Location Latitude and Longitude Drop-off Location Latitude and Longitude Order Price Driver ID Order Receiving Time Track Point Latitude and Longitude Dispatch Distance

[0076] Step B. Driver characteristic research: Consider the driver's order-taking behavior characteristics and response time characteristics. Based on the driver order-taking data, use machine learning to construct a driver order-taking preference prediction model, and at the same time combine fuzzy inference technology to establish a relationship model between driver age, working hours, and order-taking response time through fuzzy rules to complete the calibration of model parameters; perform inference calculations on real-time orders through the order-taking preference prediction model, output the order-taking probability value of the driver for a new order, and calculate the order-taking response time influence factor of different drivers at each moment based on the fuzzy control model;

[0077] Considering the characteristics of drivers' order-taking behavior and response time, the specific process is as follows:

[0078] Step B1) Based on the drivers' order-taking behavior in the past 30 days, combined with order feature data and driver attribute data, use the Logistic Regression (LR) method in machine learning for training, model the drivers' behavior patterns and calibrate parameters; use the trained prediction model to perform inference calculations on real-time orders and output the order-taking probability of drivers for new orders.

[0079] Step B2) Establish a fuzzy control model and determine the influencing factors of drivers' order-taking response time. The specific process is as follows:

[0080] Step B21) Define the input and output variables of the model: 1) The values of the control input and output variables are between [0, 1]; 2) Use the driver's age and working hours as input variables to represent the influence of the driver's age and fatigue level on the order-taking response time, where the working hours are used to represent the driver's fatigue level: the longer the working hours, the higher the driver's fatigue level; the value range of the driver's age is [18, 60] years old, divided into three age groups: young [18, 35], prime [35, 45], and middle-aged [45, 60], and the corresponding fuzzy control input intervals are [0, 0.42], [0.42, 0.65], and [0.65, 1] respectively; the value range of the working hours is [0, 18] hours, and the fatigue level is divided into three groups according to the duration: mild fatigue [0, 1.8], moderate fatigue [1.8, 10.8], and severe fatigue [10.8, 18], and the corresponding fuzzy control input intervals are [0, 0.1], [0.1, 0.6], and [0.6, 1] respectively; 3) Use the influencing factor of the order-taking response time as the output variable to represent the increment of the benchmark response time due to driver heterogeneity.

[0081] The influencing factor of the response time is divided into seven intervals: very small interval [0, 0.15], small interval [0.15, 0.30], slightly small interval [0.30, 0.45], medium interval [0.45, 0.60], slightly large interval [0.6, 0.75], large interval [0.75, 0.9], and very large interval [0.9, 1].

[0082] Step B22) Determine the membership functions of the fuzzy controller input and output: Define the membership relationships between the driver's age, working hours, and the influencing factor of the response time and each fuzzy set respectively.

[0083] In this embodiment, the input membership function adopts a mixed form of triangle and trapezoid to more accurately describe the variable characteristics in the case of fuzzy classification boundaries of age and working hours; the output membership function adopts a triangle form.

[0084] Step B23) Design fuzzy rules, as shown in Table 2 below:

[0085] Table 2 Control Rules of Fuzzy Logic System

[0086] Rule Number Driver Age Range Fatigue Level Reaction Time Impact Factor Range Rule 1 Youth Mild Very Small Rule 2 Youth Moderate Small Rule 3 Youth Severe Slightly Small Rule 4 Prime Mild Small Rule 5 Prime Moderate Medium Rule 6 Prime Severe Slightly Large Rule 7 Middle-aged Mild Slightly Small Rule 8 Middle-aged Moderate Large Rule 9 Middle-aged Severe Very Large

[0087] Step B24) Output the reaction time impact factor: Through fuzzy inference calculation, output the impact factor of the driver's order acceptance reaction time.

[0088] C. Dispatch Strategy Selection and Reaction Time Calculation:

[0089] There are mainly three dispatch strategies: the assignment mode, the grab-and-assignment mode, and the mixed mode. The assignment mode means that the driver can only accept a single order assigned by the system at the same moment. The grab-and-assignment mode means that the driver can freely grab orders on the multi-order interface at the same time. The mixed mode means that the platform screens out multiple relatively matching drivers for each order, and the matching drivers perform the operation of grabbing the order. The driver's order acceptance reaction time is divided into two parts: the driver's order acceptance decision time and the driver's order acceptance action time.

[0090] According to the number of idle drivers D within the preset time interval Δt i 、the current number of orders O c 、the predicted number of new orders O f Judge the market status, and adopt two different dispatch strategies, the assignment mode and the mixed mode, at different times; Figure 2 Both of the two dispatch strategies presented in Figure 2 (a) in is the assignment mode, Figure 2 (b) in is the mixed mode;

[0091] Construct the supply-demand evaluation index E s Judge the market supply-demand status, divide the peak period and the off-peak period according to the supply-demand relationship, and adjust the dispatch strategy in real time; the specific process is as follows:

[0092] Step C1) According to the number of idle drivers D within the preset time interval Δt i 、the current number of orders O c 、the predicted number of new orders O f Construct the supply-demand evaluation index E s , judge whether the current period is in the peak period or the off-peak period, specifically including the following steps;

[0093] Step C11) Calculate the supply-demand evaluation index E s , the specific formula is as follows:

[0094]

[0095] Among them, Of Estimated from the order data and driver data at this moment in the past 30 days;

[0096] Step C12) Assignment strategy conversion setting: Initially adopt the hybrid mode, and the default period is in the off-peak period; Set the time interval Δt, and calculate E every Δt interval s ; Set E s Threshold μ of, when E s The duration of the state greater than μ reaches the preset time ΔT, and ΔT is the longest time that E s Can exceed the threshold μ in the off-peak period state, judge it as the peak period, and automatically switch to the assignment mode. In this assignment mode, stop calculating E s ; Set the residence time C in the assignment mode. After the residence time C ends, return to the hybrid mode and loop again.

[0097] Among them, the time interval Δt is set to 20s, the threshold μ is taken as 1.2, the duration ΔT is set to 5 minutes, and the residence time C is set to 30 minutes;

[0098] Use different assignment strategies under different supply-demand relationships, and use two improved AFT models to calculate the corresponding driver's order acceptance response time under different assignment strategies. The specific process is as follows:

[0099] Calculate the order acceptance response time t for the i-th driver to accept the o-th order in different periods o,i :

[0100] (1) Peak period: In the peak period, due to the small number of drivers, the platform adopts a one-to-one assignment form to increase the order response rate; t o,i Follows a Gaussian distribution, and the specific formula is as follows:

[0101] t o,i =(1 + v o,i ) * [T s * exp(-α * (p o.i - 0.5)) + T r

[0102] (2) Off-peak period: In the off-peak period, due to the small number of orders, the platform adopts a one-to-many hybrid assignment form to shorten the order transaction time; t o,i Follows an exponential decay, and the specific formula is as follows:

[0103] t o,i =(1 + v o,i ) * [T s * exp(-α * p o.i ) + T r

[0104] Among them, v o,i ​​is the reaction time factor of the i-th driver accepting the o-th order, p o,i is the probability of the i-th driver accepting the o-th order; T s The maximum benchmark time for the driver to make an order decision; α represents p o,i The importance of T r After the driver makes a decision to accept an order, the time from the brain sending the order action command to the fingertips clicking the screen to accept the order; in any case, the driver can only accept one order at a time; preferably, T s The value of is 20s, the value of α is 1.42, and T r The value is 1.21s.

[0105] D. Determine the dispatching plan: In this example, we select the vehicle trajectory data and order data of Chengdu for one day to simulate the dispatching platform to minimize the vehicle dispatching distance E[d o ] and order execution time E[t o ] Construct a multi-objective utility function for the target and calculate the average multi-objective utility function value of all orders placed at different times of the day Use the hill climbing algorithm HC to optimize the dispatching plan and select The smallest dispatching plan; by adjusting the parameter σ, dynamically adapt to the weight distribution of the two goals in different scenarios and platform development stages;

[0106] Constructing a multi-objective utility function for different dispatching plans at the same time in different time periods includes the following steps:

[0107] Step D1) Determine the set of drivers for dispatching orders: for the oth order placed at the current moment, select I drivers for dispatching orders (I>=1); preferably, the value of I during peak hours is 1, and the value of I during off-peak hours is 3;

[0108] Step D2) Calculate the vehicle dispatch distance E[d o ]: Considering the possibility of the driver rejecting the order, calculate the vehicle dispatch distance corresponding to the oth order. The specific formula is as follows:

[0109]

[0110] Among them, d o,j is the dispatch distance between the oth order and the jth driver, p o,j is the probability of the jth driver accepting the oth order;

[0111] Step D3) Calculate the order execution time E[t o ]: Considering the possibility of the driver rejecting the order, calculate the transaction time of the oth order. The specific formula is as follows:

[0112]

[0113] where t o,j is the order acceptance response time for the j-th driver to accept the o-th order;

[0114] Step D4) Calculate the average multi-objective utility function value of all orders at the current moment The specific formula is as follows:

[0115]

[0116] where N is the number of orders at the current moment; D max is the farthest distance for order assignment; T max is the longest waiting time for passengers; preferably, D max takes the value of 6000m, T max takes the value of 120s, and σ takes the value of 1;

[0117] Step D5) Use the hill climbing algorithm HC to find the order assignment plan with the smallest average multi-objective utility function value as shown in Figure 3 The specific process is as follows:

[0118] Step D51. For the o-th order at the current moment, calculate the order acceptance probability values of idle drivers for the new order, sort them from largest to smallest, and select the top I drivers as the initial order assignment objects to form an initial order assignment plan;

[0119] Step D52. Generate a new order assignment plan by adjusting the matching relationship between orders and drivers; the specific matching relationship is as follows:

[0120] Explore the neighborhood solution space of the current order assignment plan by exchanging the matching combinations of orders and drivers to find the optimal matching method;

[0121] Step D53. Calculate the corresponding to the new order assignment plan and compare it with the current order assignment plan; if the corresponding to the new order assignment plan is less than the corresponding to the current order assignment plan, then replace the current order assignment plan with the new order assignment plan; otherwise, retain the current order assignment plan and continue to search the neighborhood solution space;

[0122] Step D54. Repeat Step D52 to Step D53 until the of the order assignment plan no longer improves. At this time, the obtained order assignment plan is: the order assignment plan with the smallest average multi-objective utility function value

[0123] ​In this example, the time range of the one-day online car-hailing data orders selected in Chengdu is from 6 to 23 o'clock. The average multi-objective utility function values, average dispatching distances (in meters, m), and average order transaction times (in seconds, s) under the optimal order dispatching plan at each moment within each hour are shown in Table 3 as follows:

[0124] Table 3 Corresponding to each optimal order dispatching plan within each hour from 6 to 23 o'clock E[d o , E[t o

[0125]

[0126] The description and application of the present invention here are illustrative, and it is not intended to limit the scope of the present invention to the above embodiments. The related descriptions of effects or advantages in the specification may not be reflected in actual experimental examples due to uncertainties in specific condition parameters or other factors. The related descriptions of effects or advantages are not used to limit the scope of the invention. Modifications and changes to the disclosed embodiments here are possible, and various components of substitution and equivalence for those embodiments are well-known to those of ordinary skill in the art. Those skilled in the art should clearly understand that the present invention can be implemented in other forms, structures, arrangements, proportions, and with other components, materials, and parts without departing from the spirit or essential characteristics of the present invention. Other modifications and changes can be made to the disclosed embodiments here without departing from the scope and spirit of the present invention.

[0127] The above is only the specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any changes or substitutions that can be easily thought of by those skilled in the art within the technical scope disclosed by the present invention should be covered by the protection scope of the present invention.​

Claims

1. A ride-hailing order dispatching method based on artificial intelligence and considering driver characteristics, characterized in that Including: Step A: Based on the large language model, preprocess the vehicle trajectory data and order data; Associate the preprocessed vehicle trajectory data with the order data to obtain driver order-receiving data; Step B: Considering the characteristics of drivers' order-receiving behaviors and response time characteristics, based on the driver order-receiving data, use machine learning to construct a driver order-receiving preference prediction model and complete parameter calibration. At the same time, combined with fuzzy inference technology, establish a fuzzy control model between driver age, working hours, and order-receiving response time by setting fuzzy rules; Process real-time orders through the driver order-receiving preference prediction model and output the order-receiving probability value of the driver for new orders; Calculate the order-receiving response time influence factors of different drivers at each moment based on the fuzzy control model; Step C: Based on the number of idle drivers D within the preset time interval Δt i , the current number of orders O c , and the predicted number of new orders O f , construct the supply-demand evaluation index E s Judge the market supply-demand status, divide the peak period and the off-peak period according to the market supply-demand relationship, and adjust the order assignment strategy in real time; Combining different market supply and demand relationships, the psychological differences of drivers in the order assignment strategy, the order-receiving probability value of the driver for new orders, and the order-receiving response time influence factors of different drivers at each moment, calculate the driver order-receiving response time; Step D: Calculate the order transaction time E[t o using the driver's order acceptance response time, and minimize the vehicle scheduling distance E[d o and the order transaction time E[t o to construct a multi-objective utility function; Optimize the dispatching plan based on the hill climbing algorithm HC in artificial intelligence technology to find the average multi-objective utility function value The dispatching plan with the minimum value.

2. The method for dispatching orders for online car-hailing considering driver characteristics based on artificial intelligence according to claim 1, wherein In Step A, the preprocessing includes outlier deletion, missing value filling, and effective trajectory extraction; among them, outlier deletion refers to automatically detecting and deleting those with position drift and abnormal speed in the order data and vehicle trajectory data based on semantic analysis and spatio-temporal pattern recognition; missing value filling refers to using spatio-temporal reasoning and context data completion technology to complete the missing time or position points in the order data and vehicle trajectory data; effective trajectory extraction refers to extracting the effective pick-up and drop-off trajectories of the vehicle through the order time period.

3. The method for dispatching orders for online car-hailing considering driver characteristics based on artificial intelligence according to claim 1, wherein In Step A, the association between the preprocessed vehicle trajectory data and the order data includes: associating the personal information of the order-receiving driver with the detailed information of the order through the ID number to obtain driver order-receiving data; the driver order-receiving data includes order feature data and driver attribute data, the order feature data includes the order start position, end position, price, travel distance, and the time period where the order is located; the driver attribute data includes the vehicle initial position, driver order-receiving frequency, and dispatching distance.

4. The method for dispatching orders for online car-hailing considering driver characteristics based on artificial intelligence according to claim 1, wherein The process of establishing the fuzzy control model includes: Step B1: Define the input and output variables of the fuzzy control model: The values of the control input and output variables are between [0, 1]; Taking the driver's age and working hours as input variables to characterize the impact of driver age and fatigue level on the order-taking response time. Among them, the working hours are used to represent the driver's fatigue level; the value range of the driver's age is [18, 60] years old, divided into three age groups: young [18, 35], prime [35, 45], and middle-aged [45, 60]. The corresponding fuzzy control input intervals for these three age groups are [0, 0.42], [0.42, 0.65], and [0.65, 1] respectively; the value range of the working hours is [0, 18] hours. According to the duration, the fatigue level is divided into three groups: mild fatigue [0, 1.8], moderate fatigue [1.8, 10.8], and severe fatigue [10.8, 18]. The corresponding fuzzy control input intervals for these three fatigue levels are [0, 0.1], [0.1, 0.6], and [0.6, 1] respectively; the response time impact factor is divided into seven intervals: very small interval [0, 0.15], small interval [0.15, 0.30], slightly small interval [0.30, 0.45], medium interval [0.45, 0.60], slightly large interval [0.6, 0.75], large interval [0.75, 0.9], and very large interval [0.9, 1]; Taking the order-taking response time impact factor as the output variable to represent the increment of the benchmark response time due to driver heterogeneity; Step B2: Determine the input and output membership functions of the fuzzy control model according to the membership relationship among the driver's age, working hours, and response time impact factor; Step B3: Design fuzzy rules, including the first rule - the ninth rule. Among them, the first rule: when the driver's age belongs to the young group and the fatigue level is mild fatigue, the response time impact factor belongs to the very small interval; the second rule: when the driver's age belongs to the young group and the fatigue level is moderate fatigue, the response time impact factor belongs to the small interval; the third rule: when the driver's age belongs to the young group and the fatigue level is severe fatigue, the response time impact factor belongs to the slightly small interval; the fourth rule: when the driver's age belongs to the prime group and the fatigue level is mild fatigue, the response time impact factor belongs to the small interval; the fifth rule: when the driver's age belongs to the prime group and the fatigue level is moderate fatigue, the response time impact factor belongs to the medium interval; the sixth rule: when the driver's age belongs to the prime group and the fatigue level is severe fatigue, the response time impact factor belongs to the slightly large interval; the seventh rule: when the driver's age belongs to the middle-aged group and the fatigue level is mild fatigue, the response time impact factor belongs to the slightly small interval; the eighth rule: when the driver's age belongs to the middle-aged group and the fatigue level is moderate fatigue, the response time impact factor belongs to the large interval; the ninth rule: when the driver's age belongs to the middle-aged group and the fatigue level is severe fatigue, the response time impact factor belongs to the very large interval; Step B4: Calculate through fuzzy reasoning and output the order-taking response time impact factor of the driver.

5. The method for dispatching orders for online car-hailing considering driver characteristics based on artificial intelligence according to claim 1, characterized in that, The order assignment strategies include the assignment mode, the grab-and-assignment mode, and the hybrid mode. Among them, the assignment mode means that a driver can only accept a single order assigned by the online car-hailing order assignment platform at the same moment. The grab-and-assignment mode means that a driver can freely grab orders on the multi-order interface at the same moment. The hybrid mode means that the online car-hailing order assignment platform screens out multiple drivers for each order, and multiple drivers simultaneously perform the order-grabbing operation on this order; The driver's order reception response time includes the driver's order reception decision time and the driver's order reception action time.

6. The method for dispatching online car-hailing orders considering driver characteristics based on artificial intelligence according to claim 1, wherein Construct the supply and demand evaluation index E s Judge the market supply and demand status, divide the peak period and the flat peak period according to the supply and demand relationship, and adjust the order assignment strategy in real time; It includes: Step C1: Calculate the supply and demand evaluation index E s ; Step C2, real-time conversion of dispatch strategy: Initially adopt a hybrid mode, set the current period as the off-peak period; set a time interval Δt, and calculate E every Δt interval s ; Set E s threshold μ of, when E s is greater than μ and the duration of this state reaches the preset time ΔT, where ΔT is the longest time that E s can exceed the threshold μ in the off-peak period state, it is judged as the peak period, and automatically switch to the assignment mode, and stop calculating E in this assignment mode s ; Set the residence time C in the assignment mode, and return to the hybrid mode to loop again after the residence time C ends.

7. The method for dispatching online car-hailing orders considering driver characteristics based on artificial intelligence according to claim 1, wherein, In step C, two improved AFT models are used to calculate the driver's order reception response time respectively; specifically as follows: Calculate the pick-up response time t when the i-th driver accepts the o-th order by period o,i : During peak hours, the platform adopts a one-to-one assignment format. o,i It is Gaussian distributed, and the specific formula is as follows: t o,i = (1 + v o,i ) * [T s * exp(-α * (p o.i - 0.5)) + T r ​ During the off-peak period, the platform adopts a one-to-many hybrid order dispatching form, t o,i shows exponential decay, and the specific formula is as follows: t o,i = (1 + v o,i ) * [T s * exp(-α * p o.i ) + T r ​ Among them, v o,i is the response time impact factor for the i-th driver to accept the o-th order, p o,i is the probability for the i-th driver to accept the o-th order; T s is the maximum reference time for the driver to make an order acceptance decision; α represents the importance of p o,i ; T r is the time from when the driver makes an order acceptance decision until the brain sends an order acceptance action instruction to the fingertip to click the screen to accept the order; A driver can only receive one order at the same moment.

8. The method for dispatching orders for online car-hailing considering driver characteristics based on artificial intelligence according to claim 7, wherein In step D, a multi-objective utility function is constructed, including: Step D1: For the o-th order at the current moment, select I drivers for order assignment to determine the order assignment driver set; where I >= 1; Step D2: Calculate the vehicle dispatching distance E[d o corresponding to the o-th order while considering the possibility of the driver rejecting the order; where d o,j is the scheduling distance between the o-th order and the j-th driver, and p o,j is the acceptance probability of the j-th driver for the o-th order; Step D3. Calculate the transaction time E[t o of the o-th order considering the possibility of the driver rejecting the order; where t o,j is the order receiving response time for the j-th driver to receive the o-th order; Step D4: Calculate the average multi-objective utility function value of all orders at the current moment Among them, N is the number of orders at the current moment; D max is the farthest distance for dispatching orders; T max is the longest waiting time of passengers.

9. The method for dispatching orders for online car-hailing considering driver characteristics based on artificial intelligence according to claim 1, characterized in that, Optimizing the order assignment plan based on the hill climbing algorithm HC in artificial intelligence technology to find the average multi-objective utility function value The order assignment plan with the minimum value, including: Step D51: For the o-th order at the current moment, calculate the order reception probability values of idle drivers for the new order, and sort them from largest to smallest in terms of the order reception probability, and select the top I drivers as the initial order assignment objects to form an initial order assignment plan; Step D52: Generate a new order assignment plan by adjusting the matching relationship between orders and drivers; the matching relationship is specifically as follows: Explore the neighborhood solution space of the current order assignment plan by exchanging the matching combinations of orders and drivers to find the optimal matching method; Step D53, calculate the corresponding to the new dispatch plan and compare it with the current dispatch plan; if the corresponding to the new dispatch plan is less than the corresponding to the current dispatch plan, then replace the current dispatch plan with the new dispatch plan; otherwise, retain the current dispatch plan and continue to search the neighborhood solution space; Step D54. Repeat Step D52 to Step D53 until the dispatch plan no longer improves. At this time, the obtained dispatch plan is the one with the smallest average multi-objective utility function value.

Images