Nonlinear surge pricing

Abstract

A system for determining a non-linear surge pricing is described. Typically, the system obtains historical data, as well as original non-linear surge pricing models and service parameters, related to service usage by a user and service provision by a provider, respectively. From the data, the system determines a user pricing elastic model and a provider pricing elastic model, and optimize the objective function based on a weighted combination of the predicted number of service uses and the predicted total amount of merchandise transactions (predicted GMV), an original non-linear surge pricing model, service parameters, and one or both of a user pricing elasticity model and a provider pricing elasticity model. The optimized objective function is an updated model for non-linear surge for generating an enhanced surge multiple for a new request for service use based on the optimized objective function and the original surge multiple.

Description

Technical Field

[0001] This invention generally relates to methods and systems for dynamically pricing service usage based on supply and demand interactions. More specifically, this invention relates to, but is not limited to, pricing the use of services (such as ride-hailing services) based on surges in demand. Background Technology

[0002] Nonlinear surge (NLS) is a feature typically designed to raise prices when demand increases, or when demand exceeds supply. The intention is to maintain the utilization rate of services while increasing the profits derived from their use. In the ride-hailing context, NLS is used to maintain the booked rate (BTR) for long-distance journeys.

[0003] NLS tends to remain static after startup, or is updated only very infrequently.

[0004] A surge pricing model updated by an agent can be generated from historical data to replace the coefficients and functions used in existing service levels. However, agent models are difficult to scale to new markets (such as new cities) because there is no historical data with optimized NLS functions that can be utilized in the agent, and such historical data cannot capture service parameters that are important now—for example, in one city, only the provision of services is captured, but not the service levels provided, such as the service levels of standard vehicles, large vehicles, and luxury vehicles.

[0005] The expectation is to overcome or partially solve at least one of the above problems, or at least provide a useful alternative. Summary of the Invention

[0006] This paper discloses a constrained nonlinear optimization framework that enables nonlinear surge (NLS) to shape demand more effectively compared to previous models. Demand can be shaped based on the optimized function to meet various business objectives with well-defined constraints—for example, not only maintaining booking rates (BTR) for long-distance journeys but also optimizing profitability. The proposed method is an automated, intelligent approach that adjusts service pricing based on service parameters. The scenario used in this paper is a ride-hailing scenario where the service is a journey or ride, and surge pricing involves journey costs based on journey parameters such as distance.

[0007] This paper discloses a system for determining nonlinear surge pricing, the system comprising: Memory; and A processing system, comprising one or more processors, with memory storing instructions, which, when executed by the processing system, cause the system to: To obtain historical data related to service usage by one or more users and service provision by one or more providers; Determine user pricing elasticity models and provider pricing elasticity models from data related to service use and service delivery; Obtain the original nonlinear surge pricing model and service parameters; The objective function is optimized based on a weighted combination of predicted service usage and predicted total merchandise volume (predicted GMV), one or both of the original nonlinear surge pricing model, user pricing elasticity model, and provider pricing elasticity model, as well as service parameters. Based on the optimized objective function and the surge multiple determined from the original surge pricing, the output is an enhanced surge multiple for new requests for service usage.

[0008] Data related to service usage and service provision may include service parameters for multiple requested service uses. The system can then determine the user pricing elasticity model and the provider pricing elasticity model through the following steps: Apply coefficients to service parameters to generate modified service parameters; and Linear regression was used on the modified service parameters.

[0009] In this regression analysis, the sign of the coefficients can be constrained to maintain the causal relationship between the service parameters and service usage for each request.

[0010] Requested service usage may include cancelled usage.

[0011] The system can identify the service level (e.g., standard, advanced, etc.) in new requests for service usage and output the boost multiplier based on the service level.

[0012] The system can receive one or more limit values, each defining a minimum or maximum surge multiple, and the objective function is optimized within these one or more limit values. These one or more limit values ​​can include the minimum and maximum ratio between the surge price determined from the original nonlinear surge pricing model and the surge price determined from the enhanced surge multiple.

[0013] The system can be configured to determine a weighted combination by applying a number between "0" and "1" to each of the predicted usage count and the predicted GMV. This might involve applying "0" to one of the predicted usage count and the predicted GMV, and "1" to the other.

[0014] Service parameters can include service levels, and the system optimizes the objective function separately for each service level. When historical data involves service usage across multiple requests, with each request including a service level, the system can determine separate user pricing elasticity models and provider pricing elasticity models for each service level. Therefore, the output augmentation multiplier can include a separate augmentation multiplier for each service level.

[0015] A method for determining nonlinear surge pricing is also disclosed, the method comprising: To obtain historical data related to service usage by one or more users and service provision by one or more providers; Determine user pricing elasticity models and provider pricing elasticity models from data related to service use and service delivery; Obtain the original nonlinear surge pricing model and service parameters; The objective function is optimized based on a weighted combination of predicted service usage and predicted total merchandise volume (predicted GMV), one or both of the original nonlinear surge pricing model, user pricing elasticity model, and provider pricing elasticity model, as well as service parameters. Based on the optimized objective function and the surge multiple determined from the original surge pricing, the output is an enhanced surge multiple for new requests for service usage.

[0016] Data related to service usage and service provision may include service parameters for multiple requested service uses. The method includes determining the user pricing elasticity model and the provider pricing elasticity model through the following steps: Apply coefficients to service parameters to generate modified service parameters; and Linear regression was applied to the modified service parameters. The signs of the coefficients are constrained to maintain the causal relationship between the service parameters and the service usage for each request.

[0017] In summary, this tutorial aims to upgrade NLS to enable its scalability to new cities and its optimization based on the latest market conditions and other objectives.

[0018] Advantageously, the method described in this paper uses a nonlinear model based on user and provider behavior to optimize surge pricing, which can be enhanced with current surge pricing models to remain relevant to new geographic regions or markets. Attached Figure Description

[0019] Embodiments of the invention will now be described by way of non-limiting examples with reference to the accompanying drawings, in which: Figure 1It is a method used to determine nonlinear surge pricing; Figure 2 The illustration shows the surge in ride-hailing scenarios; Figure 3 It demonstrates distance-based passenger surge behavior; and Figure 4 It is a system used to determine nonlinear surge pricing. Detailed Implementation

[0020] The proposed nonlinear surge pricing method extends the original surge pricing model based on current surge pricing schemes and recently captured data. In its implementation, the method for determining nonlinear surge pricing can utilize recent historical data (e.g., service usage data from the past month) and the current surge pricing model to provide an enhanced, target-based surge pricing multiple, thereby enabling accurate pricing of services based on demand.

[0021] The following discussion will be conducted with reference to a journey or ride using a ride-hailing service. However, the same teachings apply to other services with peak demand, such as hotel bookings during peak season, document delivery services during peak hours (mid-afternoon) compared to other times of day, and food delivery services around mealtimes compared to other times of day. Those skilled in the art will understand that the methods described herein are intended to be extended to other industries and services without departing from the teachings presented herein.

[0022] Figure 1 The broad features of this method are illustrated in the form of Method 100 for determining NLS pricing. Surge pricing uses prices for specific services based on surges in demand. The NLS pricing generated by Method 100 is a dynamic pricing tool used to determine surge pricing for transportation costs (or other services) based on various factors such as distance, passenger response to price changes, and driver response to price changes. User-configured parameters also exist, such as the user's primary objective (e.g., ride or GMV conversion) and different weights and coefficients. Specifically: The objective function used is a multi-objective function, with coefficients in both the simulated ride and the simulated GMV, resulting in a convex combination: a simulated_rides+b simulated_GMV, where a and b are user-configured parameters and a+b=1.

[0023] An example of the constraint used is the upper bound of NLS: NLS <c The original surge, and here, c is a user-configured parameter.

[0024] NLS uses these factors and user-configured parameters (such as service level) as input and outputs a surge multiple to modify transportation costs.

[0025] Figure 2 The illustration depicts surges in ride-hailing scenarios. Surges are clustered into buckets based on distance. Various distance metrics exist (such as physical distance and effective distance), where physical distance is the distance a passenger will travel between their origin and destination, and effective distance is defined as the NSNS cost per kilometer / minimum cost, where the NSNS cost refers to the cost without surcharges and surge fees. Figure 2 It is clearly shown that, without surges, the booking rate (BTR), defined as the number of bookings divided by the number of unique price-checking sessions (where a session refers to a user checking fares), decreases with increasing distance. This is expected because prices increase with distance. When surges are moderate (1.1x to 1.6x the NSNS fee), the BTR flattens across the distance bucket. When surges are high (>1.6), the BTR remains flat for shorter distances (5km to 10km) and then increases further with increasing distance. One possible reason for this curve shape is that the price elasticity of long-distance journeys is lower when surges are high, or during peak hours. This observation can be explained by the higher correlation of surges during periods of high demand, such as peak hours, when passengers are more willing to pay when they are in a hurry. If the objective function described in step 108 is designed to optimize revenue, addressing surge pricing during high surge periods could help capture... Figure 2 The additional benefit from the loss caused by the data-related NLS model.

[0026] One contributing factor to the surge is that drivers are often less willing to accept rides when fares are very low. This increases demand during periods when there wouldn't normally be a surge. In fact, many drivers turn their apps off entirely to avoid automatically accepting rides and then having to cancel them (which negatively impacts their satisfaction ratings).

[0027] For these reasons, it is difficult to estimate or understand the resulting surge response rate (BTR) if an alternative surge pricing model is developed from scratch. Deriving data solely from observed BTRs for our actual surge distribution is insufficient, as BTR responses for different surge distributions cannot be accurately predicted. To aid in predicting BTR, the underlying BTR can be inferred by filtering data based on a surge of 1.0 (i.e., no surge in demand). New NLS pricing models can also leverage BTR trends across time, such as peak vs. off-peak, weekdays vs. weekends, origin vs. destination combinations, and geographic regions.

[0028] To address the agnostic response of a new surge pricing model when developing such a model, method 100 models the behavior of both users and providers and uses the current surge pricing model when optimizing the objective function to generate a new enhanced surge pricing model from which an enhanced surge pricing can be computed. For this purpose, method 100 includes the following steps: 102: Obtain data; 104: Determine the user (passenger) pricing elasticity model and the provider (driver) pricing elasticity model from the data obtained in step 102; 106: Obtain the original NLS model and service parameters; 108: Optimize the objective function using the resilience model, the original NLS model, and service parameters to meet a predefined objective; and 110: Output an enhanced scalar factor based on the objective function and the scalar factor from the original NLS model.

[0029] Step 102 will typically involve retrieving historical data from a database. Data can be obtained through various other means, but many service providers are expected to include their own historical data in their databases. Historical data pertains to service usage by one or more passengers in question, as well as service provision by one or more providers. Typically, data will be derived from a large number of passengers and drivers. For this purpose, historical data can be used to model passenger and driver behavior.

[0030] Historical data is taken to include any necessary information that enables method 100 to be performed in its various embodiments. For example, service parameters may include the distance of the ride, the selling price or attempted selling price of the ride (where historical data includes rejected rides where passengers chose not to book), and the level of service (e.g., providing standard service, premium service, large vehicle, and other service types—the terms "level" and "type" of service will be used interchangeably without losing the generality of the term "type").

[0031] Step 104 involves determining the user pricing elasticity model (PAX model) and the provider pricing elasticity model (DAX model for driver behavior) from the data obtained in step 102. The PAX and DAX models model the responses of passengers to price changes and the responses of drivers to price changes.

[0032] The model is non-linear to ensure that surges are appropriate so that the objective function can be optimized across distance and surge levels. This avoids the situation where a single surge value is applied to all requests issued within a geohash-minute bucket (i.e., a distance bucket for ride requests issued within a specific geographic area). Otherwise, drivers might refuse shorter journeys (rides) in favor of longer, more profitable ones, and passengers might avoid longer journeys due to cost.

[0033] Therefore, the PAX and DAX models are based on machine learning techniques used to model the price elasticity (how demand responds to changes in service prices) behavior of both passengers and drivers. This approach infuses the enhanced NLS model with a more accurate understanding of the real market, enabling it to determine the optimal surge value.

[0034] Regarding the PAX model, the passenger's BTR (Breakthrough Rate of Transaction) is a proxy for the price elasticity of demand. Data shows that the PAX model trend differs for different distances, but generally exhibits a 'U'-shaped behavior with a valley that narrows as distance increases, such as... Figure 3 As shown. A possible reason for the narrower valley is that long-distance journeys within a certain geographic area have greater price inelasticity during peak hours (when surges are high), thus allowing for the application of higher peak pricing multiples without significant service utilization loss.

[0035] Accordingly, the data obtained in step 102 will typically include service parameters used by the service requested multiple times. Service parameters may include various parameters such as distance, price, service level (type), driver rating (assigning ratings to drivers based on passenger experience), time of day, weather, origin, destination, and other parameters. Step 104 will therefore involve applying coefficients to the service parameters to generate modified service parameters, and using linear regression on the modified service parameters to develop a PAX model (e.g., PAX and DAX models can be functions determined by regressing the service parameters and other terms mentioned herein). The regressed terms include the service parameters mentioned above, as well as interaction terms and nonlinear combinations of the service parameters (hence the term "modified" service parameters). These interactions and nonlinear combinations can be determined from contextual knowledge and causal reasoning.

[0036] The coefficients are designed to weight service parameters. These coefficients can be learned by mapping service parameters to machine learning models of ride usage or rejection. The signs of the coefficients are typically constrained to maintain a causal relationship between the ride parameters and ride usage for each ride request. In particular, not all requests will end with the passenger accepting the ride (i.e., using the service). Therefore, the data includes canceled or rejected uses, i.e., requests that the passenger rejected once the price for fulfilling the request was displayed.

[0037] The PAX model also adapts to different markets within the same geographic area. For example, some people will book only standard services, others only premium services, and some will book a mix of services. Therefore, a separate PAX model can be developed for each ride type, as price elasticity behavior can differ significantly across market segments. This similarly enables the generation of augmentation multipliers that depend on the type of service requested by the passenger.

[0038] The DAX model (Provider or Driver Resilience Model) similarly models the driver response to surges. It has been observed that fewer drivers are typically available when costs are low because they are unwilling to accept lower fares (prices). This explains why BTR increases (but only to a certain extent) as costs increase.

[0039] PAX and DAX models can be explicitly formulated as any appropriate type of model.

[0040] In some implementations, the PAX and DAX models are determined based on linear regression or curve fitting of features such as the original surge, distances and their transformations, and interaction terms. For example, these terms include distances. The surge and e^(-distance), where the determination of these interactions and nonlinear combinations can come from contextual knowledge and causal reasoning. Regression models can be generated separately for each of the various service levels and ride types (such as standard, premium, extra-large, pet-addictive, and others).

[0041] When studying the DAX model, the same service parameters and considerations can be taken into account.

[0042] The enhanced NLS model learned using method 100 utilizes the original NLS pricing model from the enhanced DAX and PAX models. For this purpose, step 106 involves obtaining the original NLS pricing model and service parameters. Some of the service parameters (such as service level) are user-configured, while others (such as distance from origin to destination, maximum waiting time, and others) are automatically determined.

[0043] The original NLS model will be developed using specific objectives. These objectives may differ from one or more objectives of the optimized model learned in step 108. For example, the original NLS model may involve maintaining the cross-distance BTR at a given surge level by reducing surges in longer-distance journeys. In contrast, the enhanced NLS multiplier can be generated through an objective function developed to maximize GMV, ride network efficiency, or throughput (e.g., driver allocation to some geographic locations based on demonstrated demand or predictions of these locations), to reduce passenger wait times.

[0044] Step 108 involves optimizing the objective function: a convex combination of simulated rides and GMV, a simulated_rides + b `simulated_GMV`, where `a` and `b` are user-configured parameters and `a+b=1`. For example, if the user's goal is simply to maximize GMV due to overall business objectives, then `a=0` and `b=1`. Optimization aims to achieve desired goals, such as maximizing the predicted total merchandise volume (GMV).

[0045] To maximize GMV, a safeguard is implemented for the number of rides for each predetermined region and time (e.g., geohash minutes) bucket. This safeguard is incorporated into the constraints of the optimization problem, where: the number of rides after the new NLS can deviate from the number of rides with the current final surge by a maximum predetermined tolerance. bookings_tolerance}

[0046] If the objective is to maximize the number of rides, then GMV protection measures are implemented again, where the total GMV after enhanced NLS can deviate from the GMV with the current final surge by a maximum of { gmv_tolerance}

[0047] Therefore, optimization may not be based on a single objective, but rather on multiple objectives—for example, a weighted combination of predicted service usage and predicted total merchandise volume (predicted GMV). Furthermore, other objectives can be added to the optimization model, or used to replace predicted rides and predicted GMV. For example, the objective function could be set to maximize the driver's effective acceptance rate (EAR), defined as the ratio of accepted bookings to the number of bookings broadcast to drivers. Other options can be set to allow for more flexible multi-objective approaches, such as maximizing average EAR and average GMV using functions like: max ( c 1 AverageEAR + c 2 average GMV ) When optimizing the objective function, the original nonlinear surge pricing model, service parameters, and one or both of the user pricing elasticity model and the provider pricing elasticity model are used. To ensure there is no mismatch between the enhanced NLS multiples for passengers and drivers, step 108 will typically use both the DAX model and the PAX model. This will avoid generating NLS multiples that are acceptable to passengers but unacceptable to drivers (e.g., too low) or acceptable to drivers but unacceptable to passengers (e.g., too high).

[0048] An objective function can apply a weighted combination of its inputs and / or efforts to achieve a weighted combination of its objectives. For example, if the objective function has multiple objectives (such as maximizing predicted GMV and maximizing predicted usage), the weighted combination of these objectives would involve applying a number between "0" and "1" to each of the predicted usage and predicted GMV. To restrict the objective function to achieving a single objective using the same framework, the weighted combination could include "0" for one of the predicted usage and predicted GMV, and "1" for the other. When weighting the inputs, ride type or distance can be given a higher weight compared to waiting time or other parameters.

[0049] When the objective function aims to maximize revenue or GMV, the formula for GMV can be given as follows:

[0050] In this context, assuming no booking cancellations, the bucket can be at the distance-hour level. In this case, the architecture used to learn or optimize the objective function might seek to manage a trade-off between increasing the BTR of longer journeys without excessively reducing the EAR of those longer journeys.

[0051] After generating the optimization model and completing the optimization process, the optimization coefficients of the NLS function are obtained. The NLS function takes into account the ride class distance and the original surge pricing included in new requests for the service (ride-hailing service) and determines the boost surge multiple. The boost surge multiple is output at step 110.

[0052] The optimized model can be further constrained to ensure it produces reasonable results—for example, without excessively depriving drivers of appropriate fees and without shocking the market compared to pricing determined by the original NLS model. For this purpose, method 100 may include receiving one or more limit values ​​at step 114. In practice, limit values ​​can be received at any time during or before optimizing the objective function. Each limit value defines a minimum or maximum scalation factor, and the objective function is optimized within these one or more limit values.

[0053] The limits can include low surge ratios (e.g., 0.5 for Singapore vehicles) and high surge ratios (e.g., 1.02 for Singapore vehicles), where the low and high surge ratios are the minimum and maximum ratios of the proposed NLS to the original surge, respectively. Therefore, the price of the enhanced NLS can be within a range […]. {lower_surge_ratio}, {upper_surge_ ratio} ] Within `original_surge`. Limits can also include low surge NLS ratios (e.g., 0.5 for Singapore vehicles) and high surge NLS ratios (e.g., 1.1 for Singapore vehicles), where the low and high surge NLS ratios are the minimum and maximum ratios of the proposed NLS to the current NLS, respectively. Therefore, the price of enhanced NLS can be within a range of […]. {lower_surge_nls_ratio}, {upper_surge_nls_ratio} ] Within `current_nls`. Other limits can be the NLS gap cap (NA for TT 199), i.e., the maximum difference between the proposed NLS and the current NLS—this is to avoid impacting the service market. Other limits are booking tolerance or GMV tolerance (e.g., 20% for TT199), i.e., the maximum allowable variation in bookings or GMV. Using these limits, the objective function can maximize a weighted combination of rides and GMV.

[0054] The optimization model and process output the coefficients of the NLS function. Therefore, step 108 outputs the coefficients of the NLS function or an NLS function optimized based on one or more objectives. For any subsequent ride request, at step 110, the NLS function uses the distance between the ride's origin and destination, along with the current surge, and outputs an enhanced surge multiplier. Where multiple vehicle types (e.g., standard, motorcycle, premium) are available, the NLS function can also use the vehicle type selected by the user for the ride and output an enhanced surge multiplier based on the vehicle type. In other implementations, multiple NLS functions may exist, each generating an enhanced surge multiplier for a different subset of rides; for example, one NLS function for standard vehicle rides and another for premium vehicle rides. This allows for optimization of the coefficients based on vehicle type or other desired parameters.

[0055] Step 110 involves outputting the enhanced surge multiple. The enhanced surge multiple is an automatic and intelligent method used to adjust the cost of each journey within different distance buckets, leveraging expert knowledge of the latest market behavior using ML techniques to optimize specific business objectives, including GMV and ride conversion. Ultimately, the cost is automatically adjusted via an NLS function that takes the original surge and distance as input and outputs the enhanced surge multiple. When the input includes ride type or other service levels, the enhanced NLS multiple can be based on the service level.

[0056] Figure 4 A block diagram of a system 100 for determining nonlinear surge pricing is illustrated. System 100 includes at least one processor 102, processor-accessible memory 104, and a network interface 108 to facilitate communication with computing devices 150 of multiple drivers and computing devices 160 of users, delivering enhanced surge pricing data to them. Program code 106, located in memory 104, includes processor-executable instructions for performing at least a portion of the methods described herein. Program code 106 implements: a resilience modeler 107 for generating passenger and driver behavior models (PAX and DAX); an optimizer 109 for optimizing an objective function; and an enhanced NLS calculator 111 for calculating enhanced NLS upon receiving a ride request, distance, and (where appropriate) ride type.

[0057] In particular, although Figure 4 While a standalone computer system is described, without departing from the intent of this disclosure, any such computer system may be distributed across multiple servers or devices, or some functions may be integrated into a single server or device. For illustration, although database 120 and memory 104 are separate, and database 120 is located at a remote location within system 100, database 120 and memory 104 may be in a single unit, one within the other, both may be distributed, or any other combination thereof.

[0058] Network 130 facilitates communication between various devices and may include one or more communication networks, including the Internet, cellular telephone networks, etc.

[0059] One or more databases 120 are also accessible to system 100. Database 120 includes records of passengers and drivers. Records may include historical data and related information related to rides taken by passengers or provided by drivers. Historical data related to rides may include the time, date, origin, destination, distance, vehicle type, weather, and other information.

[0060] System 100 optimizes ride pricing based on demand surges to achieve specific objectives, whether they are based on revenue, aim to clear the market, aim to redistribute supply across geographic regions, or other objectives.

[0061] Current approaches extend or enhance existing dynamic pricing components (enhanced NLS surge) to achieve new objectives. Specifically, NLS can enhance the initial surge pricing component, which can be designed for a specific objective (e.g., market clearing) or use a different optimization framework to achieve another objective (e.g., maximizing GMV). The dynamic pricing component can also achieve multiple objectives and is easily tunable for other objectives in the future roadmap. Currently, NLS can be used for maximizing GMV, maximizing ride conversion, or for surges during periods of supply shortage. Enhanced NLS surge models both passenger and driver behavior in a tractable manner, allowing for a good trade-off between accuracy, performance, and optimality across the entire product. The enhanced NLS model is customizable across time and location (or TT) and can be tuned for more aggressive or conservative settings to achieve single or multiple objectives.

[0062] It should be understood that numerous further modifications and arrangements of the various aspects of the described embodiments are possible. Therefore, the described aspects are intended to cover all such changes, modifications, and variations that fall within the spirit and scope of the appended claims.

[0063] Throughout this specification and the following claims, unless the context otherwise requires, the word “comprise” and its variations (such as “comprises” and “comprising”) shall be understood to implicitly include the specified integer or step or group of integers or steps, but not exclude any other integer or step or group of integers or steps.

[0064] References to any prior publications (or information derived therefrom) or any known matter in this specification are not and should not be construed as an acknowledgment or admission, or any form of implication, that such prior publications (or information derived therefrom) or known matter constitute part of the common general knowledge in the field covered by this specification.

Claims

1. A system for determining nonlinear surge pricing, comprising: Memory; as well as A processing system, including one or more processors, wherein the memory stores instructions that, when executed by the processing system, cause the system to: To obtain historical data related to service usage by one or more users and service provision by one or more providers; Determine the user pricing elasticity model and the provider pricing elasticity model from the data related to the use and provision of the service; Obtain the original nonlinear surge pricing model and service parameters; The objective function is optimized based on a weighted combination of the predicted number of times the service is used and the predicted total transaction value (GMV), one or both of the original nonlinear surge pricing model, the user pricing elasticity model and the provider pricing elasticity model, and the service parameters. Based on the optimized objective function and the surge multiple determined from the original surge pricing, the output is an enhanced surge multiple for new requests used by the service.

2. The system according to claim 1, wherein, The data related to the use and provision of the service includes service parameters used in multiple service requests. The system determines the user pricing elasticity model and the provider pricing elasticity model through the following steps: The coefficients are applied to the service parameters to generate the modified service parameters; as well as Linear regression is applied to the modified service parameters, wherein the signs of the coefficients are constrained to maintain a causal relationship between the service parameters and the service usage for each request.

3. The system according to claim 2, wherein, The requested service usage includes usage that has been cancelled.

4. The system according to claim 3, wherein, The system identifies the service level in new requests for the service and outputs the boost multiplier based on the service level.

5. The system according to claim 1, wherein, The system receives one or more limit values, each limit value defining a minimum or maximum scalar factor, and the objective function is optimized within the one or more limit values.

6. The system according to claim 5, wherein, The one or more limit values ​​include the minimum and maximum ratio between the surge price determined from the original nonlinear surge pricing model and the surge price determined from the enhanced surge multiple.

7. The system according to claim 1, wherein, The system is configured to determine a weighted combination by applying a number between "0" and "1" to each of the predicted usage count and the predicted GMV.

8. The system according to claim 7, wherein, The system is configured to determine the weighted combination by applying "0" to one of the predicted usage count and the predicted GMV, and "1" to the other of the predicted usage count and the predicted GMV.

9. The system according to claim 1, wherein, The service parameters include service levels, and the system optimizes the objective function for each service level.

10. The system according to claim 9, wherein, The historical data involves multiple service usage requests, each request including a service level, and wherein the system determines a separate user pricing elasticity model and a provider pricing elasticity model for each service level.

11. The system according to claim 9, wherein, The output of the enhancement surge multiplier includes outputting a separate enhancement surge multiplier for each service level.

12. A method for determining nonlinear surge pricing, comprising: To obtain historical data related to service usage by one or more users and service provision by one or more providers; Determine the user pricing elasticity model and the provider pricing elasticity model from the data related to the use and provision of the service; Obtain the original nonlinear surge pricing model and service parameters; The objective function is optimized based on a weighted combination of predicted service usage and predicted total merchandise volume (predicted GMV), one or both of the original nonlinear surge pricing model, the user pricing elasticity model, and the provider pricing elasticity model, as well as the service parameters. Based on the optimized objective function and the surge multiple determined from the original surge pricing, the output is an enhanced surge multiple for new requests used by the service.

13. The method according to claim 12, wherein, The data related to the use and provision of the service includes service parameters for multiple requests for the service, and the method includes determining the user pricing elasticity model and the provider pricing elasticity model through the following steps: The coefficients are applied to the service parameters to generate the modified service parameters; as well as Linear regression was applied to the modified service parameters. The signs of the coefficients are constrained to maintain a causal relationship between the service parameters and the service usage for each request.

14. The method according to claim 13, wherein, The requested service usage includes usage that has been cancelled.

15. The method of claim 14, further comprising identifying a service level in a new request for use of the service, wherein, Outputting the boost multiplier includes outputting the boost multiplier based on the service level.

16. The method of claim 12, further comprising receiving one or more limit values, each limit value defining a minimum or maximum surge multiple, and wherein, The objective function is optimized within one or more limit values.

17. The method according to claim 16, wherein, The one or more limit values ​​include the minimum and maximum ratio between the surge price determined from the original nonlinear surge pricing model and the surge price determined from the enhanced surge multiple.

18. The method of claim 12, further comprising determining the weighted combination by applying a number between "0" and "1" to each of the predicted number of uses and the predicted GMV.

19. The method of claim 18, further comprising determining the weighted combination by applying "0" to one of the predicted usage count and the predicted GMV, and applying "1" to the other of the predicted usage count and the predicted GMV.

20. The method according to claim 12, wherein, The service parameters include service levels, and the method optimizes the objective function for each service level.

21. The method according to claim 20, wherein, The historical data involves multiple requests for service usage, each request including a service level, and the method includes determining a separate user pricing elasticity model and a provider pricing elasticity model for each service level.

22. The method according to claim 20, wherein, The method outputs the boost multiplier by outputting a separate boost multiplier for each service level.

Images