Dynamic optimal matching method for online car-hailing aggregation platform

Abstract

The invention discloses a dynamic optimal matching method for an online car-hailing aggregation platform. The dynamic optimal matching method comprises the following steps: S1, defining variables required by approximate dynamic planning; s2, constructing an approximate dynamic programming matching model for maximizing the total revenue of the aggregation platform; and S3, based on the real-time analyzed variables required by the approximate dynamic programming, value approximation is performed on the expected profit in the approximate dynamic programming model by adopting an approximate dynamic programming algorithm, so that a platform vehicle driver and passenger matching strategy can be determined, and dynamic optimal matching can be realized. The dynamic approximate programming algorithm is introduced into the online car-hailing driver and passenger pricing matching strategy, more accurate pricing matching strategy adjustment is realized through real-time analysis of passenger demands, driver supply and market dynamics, the platform can continuously optimize the pricing matching strategy, the profit is maximized, the travel experience of the user is maintained, and the user experience is improved. The method not only improves the efficiency and stability of the whole system, but also can promote the reasonable utilization of resources, creates a win-win situation for drivers and passengers, and plays a positive role in the sustainable development of the platform.

Description

Technical Field

[0001] The present invention belongs to the technical field of transportation planning and management, and specifically relates to a dynamic optimal matching method for an online car-hailing aggregation platform. Background Art

[0002] Online taxi aggregation platforms refer to platforms that rely on Internet technology, cooperate with online taxi platform companies, match supply and demand information for passengers, and jointly provide online taxi booking services. This platform can optimize resource allocation through technical means, improve vehicle utilization efficiency, and provide users with more diverse and convenient travel options. With the popularization of mobile Internet and the popularization of the concept of sharing economy, online taxi aggregation platforms have developed rapidly, but they still face some challenges and problems, which are closely related to the platform's operating model and service characteristics.

[0003] Unlike traditional single online ride-hailing services, online ride-hailing aggregation platforms bring together services from multiple online ride-hailing companies, providing users with more choices. Users can compare prices, vehicle types, and estimated arrival times of different services in one application to make the most suitable choice for themselves. In addition, the drivers of online ride-hailing aggregation platforms are not exclusive drivers of a single service provider, but freelancers who can provide services to multiple service providers, which increases the flexibility and income source of drivers. However, due to the involvement of multiple service providers, the platform needs to deal with more complex scheduling and matching problems to ensure efficient docking between user needs and driver services. At the same time, the platform also needs to maintain a good user experience, handle user reviews and service quality monitoring to build and maintain user trust.

[0004] Existing customization methods for operating strategies of online ride-hailing aggregation platforms often only focus on immediate profits while ignoring expected future rewards. This approach may lead to suboptimal allocation of resources and thus fail to maximize platform revenue. Summary of the invention

[0005] In view of the above-mentioned deficiencies in the prior art, the dynamic optimal matching method for the online car-hailing aggregation platform provided by the present invention solves the problem that traditional matching methods are often unable to flexibly respond to real-time changes, resulting in uneven resource allocation and affecting user experience.

[0006] In order to achieve the above-mentioned invention object, the technical solution adopted by the present invention is: a dynamic optimal matching method for an online car-hailing aggregation platform, comprising the following steps:

[0007] S1. Define the variables required for approximate dynamic programming;

[0008] S2. Construct an approximate dynamic programming matching model to maximize the total revenue of the aggregation platform;

[0009] S3. Based on the variables required for approximate dynamic programming through real-time analysis, an approximate dynamic programming algorithm is used to approximate the value of the expected profit in the approximate dynamic programming model, and then the platform vehicle driver-passenger matching strategy is determined to achieve dynamic optimal matching.

[0010] Furthermore, in step S1, the required variables include state space, action space, immediate reward, exogenous information and state transfer function;

[0011] The state space includes resource information and demand information. The resource information indicates the number of idle vehicles on each vehicle platform and the bids for passenger demand before each decision. The demand information indicates the state of the passenger to be matched.

[0012] The action space is represented by the driver-passenger matching combinations that can be achieved in each iteration;

[0013] The instant reward refers to the reward obtained immediately after each decision;

[0014] The exogenous information represents new passenger demand information at the next iteration after the current decision is made and before the transition to a new state;

[0015] The independent variables of the state transfer function include the current state, the decision taken and the exogenous information, and the dependent variable is the next iteration state.

[0016] Furthermore, in step S2, the objective function of the constructed approximate dynamic programming matching model is expressed as:

[0017]

[0018] In the formula, π represents a selected strategy path, Π represents the set of all selectable strategy paths, k is the iteration number index, K represents the total number of iterations, and γ k represents the discount factor for the kth iteration, R k represents the passenger set at the kth iteration, S k represents the pre-decision state of the kth iteration, Represents the decision at the kth iteration under the selected strategy path.

[0019] Furthermore, the constraints of the approximate dynamic programming matching model include the number of idle vehicles on the platform, the passenger vehicle matching constraint, the passenger platform selection constraint, and the 0-1 constraint of the variable, which are respectively expressed as:

[0020]

[0021] In the formula, r represents the passenger, represents the passenger set at the kth iteration, x rprepresents the matching decision between passenger r and platform p, where p represents the vehicle platform, represents the vehicle platform set, v rp The decision variable representing whether passenger r chooses vehicle platform p.

[0022] Furthermore, in step S3, the method for determining the platform vehicle driver-passenger matching strategy is specifically as follows:

[0023] Use the value approximation in approximate dynamic programming to formulate the basis function vector And gradually update the coefficient vector θ by iterating the approximate dynamic programming model k , by the basis function vector The coefficient vector θ is updated with k The inner product of is used to approximate the expected profit of the aggregation platform, and then the immediate profit and expected profit are considered simultaneously before making a decision. On this basis, the platform vehicle driver-passenger matching strategy is determined to achieve dynamic optimal matching.

[0024] Furthermore, the iteration gradually updates the coefficient vector θ k The specific method is:

[0025] S31, initialize the decision state to The coefficient vector θ k Initialized to θ0;

[0026] S32. In each iteration of the approximate dynamic programming, take measures to satisfy the value Maximizing decision making Then, the state transfer function is used to transfer the pre-decision state to the new post-decision state;

[0027] The value Including decision making Immediate profit after And the basis function vector With the coefficient vector θ k-1 The future value of the state after the decision is approximated

[0028] S33, based on the generation of this iteration Based on the estimate of this value from the previous iteration Calculate the loss l between the two k ;

[0029] S34, according to the loss l k And regularization terms calculate the total gradient, and determine whether it exceeds the maximum gradient limit;

[0030] If yes, proceed to step S35;

[0031] If not, proceed to step S36;

[0032] S35, perform gradient clipping, and then proceed to step S36;

[0033] S36, the current total gradient is updated through Adam optimization momentum, and the updated coefficient vector θ is output under the current number of iterations k , and returns to step S32 to perform the next iteration.

[0034] Furthermore, in step S32, the decision taken satisfy:

[0035]

[0036] In the formula, γ represents the discount coefficient;

[0037] Instant Profit It is expressed as:

[0038]

[0039] In the formula, β p represents the profit ratio taken by the aggregation platform after the vehicle platform p accepts the order, q rp represents the price offered by vehicle platform p to passenger r, x rp represents the matching decision between passenger r and platform p;

[0040] Approximate future value It is expressed as:

[0041]

[0042] In step S33, the loss k It is expressed as:

[0043]

[0044] In step S34, the total gradient calculated It is expressed as:

[0045]

[0046] In the formula, represents the gradient calculated from the TD error, represents the regularized gradient;

[0047] In step S35, the clipped gradient It is expressed as:

[0048]

[0049] In the formula, represents the gradient norm limit;

[0050] In step S36, the updated coefficient vector is expressed as:

[0051]

[0052] In the formula, α k represents the learning rate of the kth iteration, represents the first-order momentum at the k-th iteration, represents the second-order momentum of the k-th iteration, and ε represents a positive number set to prevent the denominator from being 0.

[0053] Furthermore, the method also includes making action decisions for the approximate dynamic programming model through a greedy algorithm, and adjusting parameters of the approximate dynamic programming model by comparing decision profit parameters of the approximate dynamic programming algorithm and the greedy algorithm.

[0054] Furthermore, the decision profit parameters compared include the total platform profit and the average profit per platform driver;

[0055] Among them, the total profit of the platform s and the average profit per driver on the platform a The calculation formula is:

[0056]

[0057] In the formula, c represents the platform commission ratio, N represents the number of platform drivers, Indicates the total number of iterations.

[0058] The beneficial effects of the present invention are:

[0059] The dynamic optimal matching method for online car-hailing aggregation platforms provided by the present invention, as an intelligent optimization technology, provides a new idea for the dynamic matching of online car-hailing platforms, and can effectively cope with complex market demand and supply fluctuations. Its advantages are specifically reflected in:

[0060] (1) The present invention introduces the approximate dynamic programming algorithm (ADP) into the online car-hailing matching strategy. By analyzing passenger demand, driver supply and market dynamics in real time, an algorithm based on basis function value approximation is designed. By iteratively updating the coefficient vector, the optimal choice of the pre-decision state is achieved, aiming to maximize the comprehensive benefits of immediate profit and future value, and achieve more accurate matching adjustments.

[0061] (2) In the algorithm based on basis function value approximation provided by the present invention, a coefficient vector update mechanism using the Adam optimization algorithm is designed by combining the immediate reward and loss calculation to ensure the stability and convergence of the algorithm; at the same time, the state transfer function is used to realize the conversion from the pre-decision state to the post-decision state, forming an effective feedback loop, thereby enhancing the algorithm's adaptability and real-time responsiveness.

[0062] In summary, the iterative learning ability of the dynamic matching method provided by the present invention enables the platform to continuously optimize the matching strategy, maximize profits while maintaining the user's travel experience. Such a dynamic matching mechanism not only improves the efficiency and stability of the overall system, but also promotes the rational use of resources, creates a win-win situation for drivers and passengers, and plays a positive role in the sustainable development of the platform. BRIEF DESCRIPTION OF THE DRAWINGS

[0063] Figure 1 A flow chart of the dynamic optimal matching method for the online car-hailing aggregation platform provided by the present invention. DETAILED DESCRIPTION

[0064] The specific implementation modes of the present invention are described below so that those skilled in the art can understand the present invention. However, it should be clear that the present invention is not limited to the scope of the specific implementation modes. For those of ordinary skill in the art, as long as various changes are within the spirit and scope of the present invention as defined and determined by the attached claims, these changes are obvious, and all inventions and creations utilizing the concept of the present invention are protected.

[0065] The embodiment of the present invention provides a dynamic optimal matching method for an online car-hailing aggregation platform, such as Figure 1 As shown, the following steps are included:

[0066] S1. Define the variables required for approximate dynamic programming and collect relevant data;

[0067] S2. Construct an approximate dynamic programming matching model to maximize the total revenue of the aggregation platform;

[0068] S3. Based on the variables required for approximate dynamic programming through real-time analysis, an approximate dynamic programming algorithm is used to approximate the value of the expected profit in the approximate dynamic programming model, and then the platform vehicle driver-passenger matching strategy is determined to achieve dynamic optimal matching.

[0069] In the step S1 of the embodiment of the present invention, the required variables defined include state space, action space, immediate reward, exogenous information and state transfer function;

[0070] The state space includes resource information and demand information. The resource information indicates the number of idle vehicles on each vehicle platform and the quotation for passenger demand before each decision. The demand information indicates the state of the passenger to be matched. For example, the demand information specifically indicates the passenger's preference for each platform, that is, whether the platform is selected as an alternative.

[0071] The action space is represented by the driver-passenger matching combinations that can be achieved in each iteration; specifically, the platform will not refuse to take a passenger, and whether the privacy condition is partially matched depends on the passenger's signal;

[0072] Instant rewards refer to the rewards obtained immediately after each decision, which are determined by the platform's quotation for passenger demand and the specific decision of this iteration;

[0073] The exogenous information represents the new passenger demand information that arrives at the next iteration after the current decision is made and before the transition to a new state;

[0074] The independent variables of the state transfer function include the current state, the decision taken, and exogenous information, and its dependent variable is the next iteration state.

[0075] In step S2 of the embodiment of the present invention, the objective function of the constructed approximate dynamic programming matching model is expressed as:

[0076]

[0077] In the formula, π represents a selected strategy path, Π represents the set of all selectable strategy paths, k is the iteration number index, K represents the total number of iterations, and γ k represents the discount factor for the kth iteration, R k represents the passenger set at the kth iteration, S k represents the pre-decision state of the kth iteration, Represents the decision at the kth iteration under the selected strategy path.

[0078] In this embodiment, the constraints of the approximate dynamic programming matching model include the number of idle vehicles on the platform, the passenger vehicle matching constraint, the passenger platform selection constraint, and the 0-1 constraint of the variable, which are respectively expressed as:

[0079]

[0080] In the formula, r represents the passenger, represents the passenger set at the kth iteration, x rp represents the matching decision between passenger r and platform p, where p represents the vehicle platform, represents the vehicle platform set, v rp The decision variable representing whether passenger r chooses vehicle platform p.

[0081] Among the above constraints, the platform idle vehicle number constraint means that the number of passengers assigned to platform p does not exceed the number of available idle vehicles, the passenger-vehicle matching constraint means that for any passenger, he can be matched by a vehicle from at most one platform, and the passenger platform selection constraint means that passengers can only be assigned to the platforms they choose.

[0082] In step S3 of the embodiment of the present invention, the method for determining the platform vehicle driver-passenger matching strategy is specifically as follows:

[0083] Use the value approximation in approximate dynamic programming to formulate the basis function vector And gradually update the coefficient vector θ by iterating the approximate dynamic programming model k , by the basis function vector The coefficient vector θ is updated with k The inner product of is used to approximate the expected profit of the aggregation platform, and then the immediate profit and expected profit are considered simultaneously before making a decision. On this basis, the platform vehicle driver-passenger matching strategy is determined to achieve dynamic optimal matching.

[0084] Specifically, based on the above method, the coefficient vector θ is iteratively updated step by step in this embodiment. k The specific method is:

[0085] S31, initialize the decision state to The coefficient vector θ k Initialized to θ0;

[0086] S32. In each iteration of the approximate dynamic programming, take measures to satisfy the value Maximizing decision making Then, the state transfer function is used to transfer the pre-decision state to the new post-decision state;

[0087] The value Including decision making Immediate profit after And the basis function vector With the coefficient vector θ k-1 The future value of the state after the decision is approximated

[0088] Among them, the decision satisfy:

[0089]

[0090] In the formula, γ represents the discount coefficient;

[0091] Instant Profit It is expressed as:

[0092]

[0093] In the formula, β p represents the profit ratio taken by the aggregation platform after the vehicle platform p accepts the order, q rp represents the price offered by vehicle platform p to passenger r, x rp represents the matching decision between passenger r and platform p;

[0094] Approximate future value It is expressed as:

[0095]

[0096] S33, based on the generation of this iteration Based on the estimate of this value from the previous iteration Calculate the loss l between the two k ;

[0097] Among them, the loss l k It is expressed as:

[0098]

[0099] S34, according to the loss l k And regularization terms calculate the total gradient, and determine whether it exceeds the maximum gradient limit;

[0100] If yes, proceed to step S35;

[0101] If not, proceed to step S36;

[0102] Among them, the total gradient calculated It is expressed as:

[0103]

[0104] In the formula, represents the gradient calculated from the TD error, represents the regularized gradient;

[0105] S35, perform gradient clipping, and then proceed to step S36;

[0106] Among them, the clipped gradient It is expressed as:

[0107]

[0108] In the formula, represents the gradient norm limit;

[0109] S36, the current total gradient is updated through Adam optimization momentum, and the updated coefficient vector θ is output under the current number of iterations k, and return to step S32 to perform the next iteration;

[0110] Among them, the updated coefficient vector is expressed as:

[0111]

[0112] In the formula, α k represents the learning rate of the kth iteration, represents the first-order momentum at the k-th iteration, represents the second-order momentum of the k-th iteration, and ε represents a positive number set to prevent the denominator from being 0.

[0113] In an embodiment of the present invention, the dynamic optimal matching method also includes making action decisions on the approximate dynamic programming model through a greedy algorithm, and performing sensitivity analysis and parameter adjustment on the approximate dynamic programming model by comparing the decision profit parameters of the approximate dynamic programming algorithm and the greedy algorithm.

[0114] Specifically, we solve the above problem by Python and get the approximated θ k A greedy algorithm is also set up, in which the action with the greatest immediate profit is taken for each decision. The decision profit parameters compared include the total profit of the platform and the average profit per driver on the platform;

[0115] Among them, the total profit of the platform s and the average profit per driver on the platform a The calculation formula is:

[0116]

[0117] In the formula, c represents the platform commission ratio, N represents the number of platform drivers, Indicates the total number of iterations.

[0118] In the above process, the parameters involved in the sensitivity analysis include the driver arrival rate and the passenger arrival rate. For the driver arrival rate, when the passenger arrival rate is constant, the changes in the experimental results are analyzed by changing the driver arrival rate; for the passenger arrival rate, when the driver arrival rate is constant, the changes in the experimental results are analyzed by changing the passenger arrival rate.

[0119] In this embodiment, sensitivity analysis can help evaluate the stability and robustness of the system under different conditions, ensuring that the model can still provide reliable results even in an environment with high uncertainty. At the same time, by observing the experimental results under different arrival rates, it can provide a basis for parameter selection and tuning, further improving the model's prediction ability and practical application effect.

[0120] The present invention uses specific embodiments to illustrate the principles and implementation methods of the present invention. The description of the above embodiments is only used to help understand the method of the present invention and its core idea. At the same time, for those skilled in the art, according to the idea of ​​the present invention, there will be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as a limitation on the present invention.

[0121] Those skilled in the art will appreciate that the embodiments described herein are intended to help readers understand the principles of the present invention, and should be understood that the protection scope of the present invention is not limited to such specific statements and embodiments. Those skilled in the art can make various other specific variations and combinations that do not deviate from the essence of the present invention based on the technical revelations disclosed by the present invention, and these variations and combinations are still within the protection scope of the present invention.

Claims

1. A dynamic optimal matching method for an online car-hailing aggregation platform, characterized in that: The following steps are involved: S1. Define the variables required for approximate dynamic programming; S2. Construct an approximate dynamic programming matching model to maximize the total revenue of the aggregation platform; S3. Based on the variables required for approximate dynamic programming through real-time analysis, an approximate dynamic programming algorithm is used to approximate the value of the expected profit in the approximate dynamic programming model, and then the platform vehicle driver-passenger matching strategy is determined to achieve dynamic optimal matching.

2. The dynamic optimal matching method for online car-hailing aggregation platform according to claim 1 is characterized in that: In step S1, the required variables include state space, action space, immediate reward, exogenous information and state transfer function; The state space includes resource information and demand information. The resource information indicates the number of idle vehicles on each vehicle platform and the bids for passenger demand before each decision. The demand information indicates the state of the passenger to be matched. The action space is represented by the driver-passenger matching combinations that can be achieved in each iteration; The instant reward refers to the reward obtained immediately after each decision; The exogenous information represents new passenger demand information at the next iteration after the current decision is made and before the transition to a new state; The independent variables of the state transfer function include the current state, the decision taken and the exogenous information, and the dependent variable is the next iteration state.

3. The dynamic optimal matching method for online car-hailing aggregation platform according to claim 1 is characterized in that: In step S2, the objective function of the constructed approximate dynamic programming matching model is expressed as: In the formula, π represents a selected strategy path, Π represents the set of all selectable strategy paths, k is the iteration number index, K represents the total number of iterations, and γ k represents the discount factor for the kth iteration, R k represents the passenger set at the kth iteration, S k represents the pre-decision state of the kth iteration, Represents the decision at the kth iteration under the selected strategy path.

4. The dynamic optimal matching method for online car-hailing aggregation platform according to claim 3 is characterized in that: The constraints of the approximate dynamic programming matching model include the number of idle vehicles on the platform, the passenger vehicle matching constraint, the passenger platform selection constraint, and the 0-1 constraint of the variable, which are expressed as follows: In the formula, r represents the passenger, represents the passenger set at the kth iteration, x rp represents the matching decision between passenger r and platform p, where p represents the vehicle platform, represents the vehicle platform set, v rp The decision variable representing whether passenger r chooses vehicle platform p.

5. The dynamic optimal matching method for online car-hailing aggregation platform according to claim 4 is characterized in that: In step S3, the method for determining the platform vehicle driver-passenger matching strategy is specifically as follows: Use the value approximation in approximate dynamic programming to formulate the basis function vector And gradually update the coefficient vector θ by iterating the approximate dynamic programming model k , by the basis function vector The coefficient vector θ is updated with k The inner product of is used to approximate the expected profit of the aggregation platform, and then the immediate profit and expected profit are considered simultaneously before making a decision. On this basis, the platform vehicle driver-passenger matching strategy is determined to achieve dynamic optimal matching.

6. The dynamic optimal matching method for online car-hailing aggregation platform according to claim 5 is characterized in that: The iteration gradually updates the coefficient vector θ k The specific method is: S31, initialize the decision state to The coefficient vector θ k Initialized to θ0; S32. In each iteration of the approximate dynamic programming, take measures to satisfy the value Maximizing decision making Then, the state transfer function is used to transfer the pre-decision state to the new post-decision state; The value Including decision making Immediate profit after And the basis function vector With the coefficient vector θ k-1 The future value of the state after the decision is approximated S33, based on the generation of this iteration Based on the estimate of this value from the previous iteration Calculate the loss l between the two k ; S34, according to the loss l k And regularization terms calculate the total gradient, and determine whether it exceeds the maximum gradient limit; If yes, proceed to step S35; If not, proceed to step S36; S35, perform gradient clipping, and then proceed to step S36; S36, the current total gradient is updated through Adam optimization momentum, and the updated coefficient vector θ is output under the current number of iterations k , and returns to step S32 to perform the next iteration.

7. The dynamic optimal matching method for online car-hailing aggregation platform according to claim 6 is characterized in that: In step S32, the decision taken satisfy: In the formula, γ represents the discount coefficient; Instant Profit It is expressed as: In the formula, β p represents the profit ratio taken by the aggregation platform after the vehicle platform p accepts the order, q rp represents the price offered by vehicle platform p to passenger r, x rp represents the matching decision between passenger r and platform p; Approximate future value It is expressed as: In step S33, the loss k It is expressed as: In step S34, the total gradient calculated It is expressed as: In the formula, represents the gradient calculated from the TD error, represents the regularized gradient; In step S35, the clipped gradient It is expressed as: In the formula, represents the gradient norm limit; In step S36, the updated coefficient vector is expressed as: In the formula, α k represents the learning rate of the kth iteration, represents the first-order momentum at the k-th iteration, represents the second-order momentum of the k-th iteration, and ε represents a positive number set to prevent the denominator from being 0.

8. The dynamic optimal matching method for online car-hailing aggregation platform according to claim 6 is characterized in that: The method also includes making action decisions on the approximate dynamic programming model through a greedy algorithm, and adjusting parameters of the approximate dynamic programming model by comparing decision profit parameters of the approximate dynamic programming algorithm and the greedy algorithm.

9. The dynamic optimal matching method for online car-hailing aggregation platform according to claim 8 is characterized in that: The decision profit parameters compared include the total platform profit and the average profit per platform driver; Among them, the total profit of the platform s and the average profit per driver on the platform a The calculation formula is: In the formula, c represents the platform commission ratio, N represents the number of platform drivers, Indicates the total number of iterations.

Images