Online car-hailing discount decision-making method and device and computer equipment

Abstract

The invention relates to an online car-hailing discount decision-making method and device and computer equipment, and the method comprises the steps: determining target input data based on order supply information obtained in advance; inputting the target input data into a pre-constructed decision support model for decision processing, and obtaining a target decision result including the number and proportion of preferential listening orders opened corresponding to platform revenue maximization; and sending recommendation information for recommending a user of the target terminal to start a preferential service to at least one target terminal according to the target decision result. According to the method, the complex relationship in the data can be deeply analyzed, and the preferential listening order starting number and proportion corresponding to the platform revenue maximization can be accurately predicted as the target decision result. The method not only helps the server to maximize economic benefits, but also gives consideration to the benefits of driver groups while guaranteeing the platform benefits.

Description

[0001] Related Applications

[0002] This application claims the priority of a Chinese patent application with the application number 2024117919375 and the title "A Method, Device and Computer Equipment for Determining Fixed-Price Decisions in Online Ride-Hailing Services" filed on December 06, 2024, the entire content of which is incorporated herein by reference. Technical Field

[0003] This application relates to the technical field of data processing and analysis, and particularly to a method, device and computer equipment for making preferential decisions in online ride-hailing services. Background Art

[0004] With the rapid development of Internet technology, the online ride-hailing industry has emerged and grown rapidly, becoming one of the important transportation modes for people's daily travel. Among them, preferential services are an emerging business model in the online ride-hailing industry.

[0005] However, currently, the setting of preferential order receiving is manually turned on and off by the driver. It is difficult for the driver to adjust the relevant settings in a timely manner during driving. In most cases, the driver chooses to keep it on or off all the time after going online. This makes the setting of preferential order receiving unable to be adjusted in real time according to the actual supply and demand situation. It is difficult to meet the needs of the server and the driver for efficient and intelligent decision-making and achieving the maximization of overall revenue. Summary of the Invention

[0006] Based on this, it is necessary to provide an online ride-hailing preferential decision-making method, device and computer equipment that can solve the above problems for the above technical problems.

[0007] In a first aspect, this application provides an online ride-hailing preferential decision-making method, which includes:

[0008] Determine target input data based on pre-acquired order supply information;

[0009] Input the target input data into a pre-constructed decision support model for decision-making processing to obtain a target decision result; the target decision result includes the number and proportion of preferential order receiving openings corresponding to the maximization of platform revenue;

[0010] Send recommendation information to at least one target terminal according to the target decision result, and the recommendation information is used to recommend that the users of the target terminal turn on the preferential service.

[0011] In one of the embodiments, the above target input data includes global spatio-temporal information, environmental variables and the number of drivers before decision-making; the decision support model includes a one-to-one connection layer, an encoder layer, a decision layer and a fully connected layer. Inputting the target input data into the pre-constructed decision support model for decision-making processing to obtain a target decision result includes:

[0012] Input the environmental variables into a one-to-one connection layer for feature extraction processing to obtain feature data; the feature extraction includes at least one of geographical location feature extraction, time feature extraction, user feature extraction, and order-related feature extraction;

[0013] Input the number of drivers before decision-making into the decision layer for binary classification decision-making processing to obtain decision information;

[0014] Input the global spatio-temporal information into the encoder layer for global spatio-temporal supply and demand information extraction processing to obtain a global spatio-temporal supply and demand vector;

[0015] Input the feature data, the global spatio-temporal supply and demand vector, and the decision information into the fully connected layer for feature integration and feature extraction processing to obtain the target decision result.

[0016] In one embodiment, the above decision information includes the proportions of various types of drivers to turn on and turn off the preferential order listening; input the number of drivers before decision-making into the decision layer for binary classification decision-making processing to obtain decision information, including:

[0017] Perform a weighted summation operation on the pre-acquired weight parameters and the number of drivers before decision-making to obtain an intermediate result;

[0018] Input the intermediate result into a pre-constructed activation function for binary classification decision-making processing to obtain the proportions of various types of drivers to turn on and turn off the preferential order listening.

[0019] In one embodiment, the above method further includes:

[0020] Perform data preprocessing on the pre-acquired historical operation data to obtain training data; the data preprocessing includes format processing, hexagonal grid coding conversion processing, and data partitioning processing;

[0021] Input the training data into the initial prediction model for prediction processing to obtain an initial prediction result;

[0022] Adjust the parameters in the initial prediction model according to the initial prediction result and the mean absolute value function to determine the decision support model.

[0023] In one embodiment, the above method further includes:

[0024] Calculate the maximum platform revenue value according to the target decision result and the pre-established objective function; the objective function includes the corresponding relationship between the total platform revenue and at least one of the order price, the platform commission rate, and the operating cost.

[0025] In one embodiment, the training process of the above objective function includes:

[0026] Obtain multiple samples, where the samples include the number of drivers of various types before decision-making, environmental information, and the actual income data of the drivers;

[0027] For each iteration process, determine the initial function value according to the number of drivers of various types before decision-making, environmental information, and the actual income data of the drivers;

[0028] Update the parameters of the initial function according to the initial function value and the optimization algorithm;

[0029] When the current number of iterations meets the maximum number of iterations, determine the initial function at the current number of iterations as the objective function.

[0030] In one embodiment, the above-mentioned determining the initial function value according to the number of drivers of various types before decision-making, environmental information, and the actual income data of the drivers includes:

[0031] Obtain the driver type weight coefficient, the environmental type weight coefficient, and the penalty term coefficient;

[0032] Multiply the number of drivers of various types before decision-making by the driver type weight coefficient to obtain the first function value;

[0033] Multiply the environmental information by the environmental type weight coefficient to obtain the second function value;

[0034] Conduct income analysis processing on the actual income data of the drivers to obtain the income value, and multiply the income value by the penalty term coefficient to obtain the penalty value;

[0035] Determine the initial function value according to the first function value, the second function value, and the penalty value.

[0036] In one embodiment, the above-mentioned determining the target input data based on the pre-obtained order supply information includes:

[0037] Perform data preprocessing on the order supply information to obtain the target input data; the data preprocessing includes at least one of space-time partitioning, driver classification, driver status judgment, and driver order-taking statistics.

[0038] In a second aspect, the present application also provides a preferential decision-making device for online car-hailing. The device includes:

[0039] A data determination module, configured to determine target input data based on the pre-obtained order supply information;

[0040] A decision-making module, configured to input the target input data into a pre-constructed decision support model for decision-making processing to obtain a target decision result; the target decision result includes the number and proportion of preferential order-taking openings corresponding to the maximization of platform income;

[0041] An information sending module, configured to send recommendation information to at least one target terminal according to a target decision result, where the recommendation information is used to recommend that a user of the target terminal enable a preferential service.

[0042] Thirdly, the present application further provides a computer device. The computer device includes a memory and a processor. The memory stores a computer program, and when the processor executes the computer program, the following steps are implemented:

[0043] Determine target input data based on pre-acquired order supply information;

[0044] Input the target input data into a pre-constructed decision support model for decision processing to obtain a target decision result; the target decision result includes the opening quantity and ratio of preferential order receiving corresponding to the maximum platform revenue;

[0045] Send recommendation information to at least one target terminal according to the target decision result, where the recommendation information is used to recommend that a user of the target terminal enable a preferential service.

[0046] Fourthly, the present application further provides a computer-readable storage medium. On the computer-readable storage medium, a computer program is stored, and when the computer program is executed by a processor, the following steps are implemented:

[0047] Determine target input data based on pre-acquired order supply information;

[0048] Input the target input data into a pre-constructed decision support model for decision processing to obtain a target decision result; the target decision result includes the opening quantity and ratio of preferential order receiving corresponding to the maximum platform revenue;

[0049] Send recommendation information to at least one target terminal according to the target decision result, where the recommendation information is used to recommend that a user of the target terminal enable a preferential service.

[0050] Fifthly, the present application further provides a computer program product. The computer program product includes a computer program, and when the computer program is executed by a processor, the following steps are implemented:

[0051] Determine target input data based on pre-acquired order supply information;

[0052] Input the target input data into a pre-constructed decision support model for decision processing to obtain a target decision result; the target decision result includes the opening quantity and ratio of preferential order receiving corresponding to the maximum platform revenue;

[0053] Send recommendation information to at least one target terminal according to the target decision result, where the recommendation information is used to recommend that a user of the target terminal enable a preferential service.

[0054] The above-mentioned online car-hailing discount decision-making method, device and computer equipment determine target input data based on the order supply information obtained in advance; input the target input data into a pre-constructed decision support model for decision-making processing to obtain a target decision result including the number and proportion of discount order acceptance openings corresponding to the maximization of platform revenue; and send recommendation information for recommending users of at least one target terminal to turn on the discount service according to the target decision result. This application can deeply analyze the complex relationships in the data and accurately predict the number and proportion of discount order acceptance openings corresponding to the maximization of platform revenue as the target decision result. This not only helps the server achieve the maximization of economic benefits, but also takes into account the interests of the driver group while ensuring the platform revenue. Brief Description of the Drawings

[0055] Figure 1 It is an application environment diagram of the online car-hailing discount decision-making method in an embodiment;

[0056] Figure 2 It is a schematic flowchart of the online car-hailing discount decision-making method in an embodiment;

[0057] Figure 3 It is a schematic flowchart of the process of obtaining the target decision result in an embodiment;

[0058] Figure 4 It is a schematic flowchart of the process of obtaining the decision information in an embodiment;

[0059] Figure 5 It is a schematic flowchart of the decision support model training process in an embodiment;

[0060] Figure 6 It is a schematic flowchart of the training process of the objective function in an embodiment;

[0061] Figure 7 It is a schematic flowchart of the process of determining the initial function value in an embodiment;

[0062] Figure 8 It is a three-dimensional coordinate encoding format diagram of a hexagonal grid in an embodiment;

[0063] Figure 9 It is a schematic diagram of a neural network prediction model in an embodiment;

[0064] Figure 10 It is a structural block diagram of the online car-hailing discount decision-making device in an embodiment;

[0065] Figure 11 It is an internal structure diagram of a computer device in an embodiment. Detailed Description of the Embodiments

[0066] In order to make the objectives, technical solutions and advantages of the present application more clear and understandable, the present application will be further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0067] First, before specifically introducing the technical solutions of the embodiments of the present application, the technical background on which the embodiments of the present application are based will be introduced.

[0068] With the increasingly fierce competition in the online car-hailing market, providing preferential services has become a way to enhance the competitiveness of the platform and attract passengers. The online car-hailing platform needs to reasonably set the number and proportion of different types of drivers who can receive orders with discounts in different regions and time slices to ensure that the supply matches the demand at different levels while maximizing the service revenue. The existing settings for online car-hailing orders with discounts only rely on the individual preferences of drivers, resulting in vicious competition, lack of a dynamic adjustment mechanism, mismatch between supply and demand stratification, and inability to achieve global optimality.

[0069] Preferential services are a new business model in the online car-hailing market. Compared with the traditional mileage-based billing model, preferential services charge a fixed fee based on the pick-up and drop-off points of passengers, and additional costs such as congestion and detours during the journey will be borne by the platform. Therefore, the profit of a single service is lower than that of the traditional billing model. Such services meet the needs of passengers for transparent pricing, improve the competitiveness and attractiveness of the platform, but at the cost of reducing the revenue of a single service.

[0070] Current preferential order receiving settings for online car-hailing services often rely on drivers' personal experience and habits, and cannot respond to the spatio-temporal changes in demand and supply from a global system perspective. There are the following four problems: 1. There is vicious competition. For individual drivers, turning on preferential order receiving can increase the probability of receiving orders, but this advantage will gradually disappear as the number of drivers choosing preferential order receiving within the region increases. In extreme cases, all drivers in the region choose preferential order receiving. At this time, choosing preferential order receiving does not bring advantages to drivers, but instead reduces the service profit of drivers due to changes in the billing rules. 2. There is a lack of a dynamic adjustment mechanism. Currently, the preferential order receiving settings are manually turned on and off by drivers. It is difficult for drivers to adjust the relevant settings in a timely manner during driving. In most cases, drivers choose to keep them on or off after logging in. 3. There is a mismatch between supply and demand stratification. Preferential orders and traditional mileage-based billing orders correspond to different demands and supplies. When passengers call for online car-hailing services, they can select different service types, such as only choosing preferential services, only choosing traditional mode services, and choosing both services at the same time. Under the multi-level passenger demand types, it is easy to cause a mismatch in service supply between different levels. For example, no driver in the region chooses preferential order receiving, but there is a large number of passenger demands for only preferential services in the region. The stratification mismatch between supply and demand will reduce service efficiency and passenger satisfaction. 4. It is impossible to achieve global optimality. The individual behavior of drivers cannot guarantee the maximization of overall benefits.

[0071] Based on this, the present application provides an online car-hailing preferential decision-making method, device, and computer device, aiming to solve the above technical problems.

[0072] The online car-hailing preferential decision-making method provided by the embodiments of the present application can be applied to an application environment as Figure 1 shown. Among them, the terminal 102 communicates with the server 104 through the network. The data storage system can store the data that the server 104 needs to process. The data storage system can be integrated on the server 104, or placed in the cloud or other network servers. The server determines the target input data based on the pre-obtained order supply information; inputs the target input data into a pre-constructed decision support model for decision-making processing, and obtains a target decision result including the number and proportion of preferential order receiving openings corresponding to the maximization of platform revenue; and sends recommendation information for recommending the users of the target terminal to turn on preferential services to at least one target terminal according to the target decision result. Among them, the terminal 102 can be, but is not limited to, a smart phone, a tablet computer, a smart in-vehicle device, etc. The server 104 can be implemented by an independent server or a server cluster composed of multiple servers.

[0073] In one embodiment, as Figure 2 shown, an online car-hailing preferential decision-making method is provided. This method is applied to Figure 1Taking the server in [it] as an example for illustration, it includes the following steps:

[0074] S201, determine the target input data based on the pre-obtained order supply information.

[0075] Among them, the order supply information refers to a series of data information related to online car-hailing orders, covering the starting location, destination location, order placement time, order type (such as preferential orders, ordinary orders, etc.), order price, the order demand quantity and driver supply quantity in different time periods within each region, etc. These information comprehensively reflect the order supply and demand situation in the online car-hailing operation scenario and are an important basis for subsequent decision-making processes.

[0076] The target input data is a data set obtained after a series of preprocessing operations on the pre-obtained order supply information. It has undergone data cleaning, format standardization, feature extraction, etc., to make it more suitable for input into the decision support model for analysis and decision-making, and contains content such as encoded geographical location information, standardized time information, driver- and order-related features, etc.

[0077] Data collection in the embodiments of this application: Comprehensively collect various order supply information from the database of the online car-hailing platform, including the detailed information of the order (starting location, destination location, order placement time, order type, order price, etc.), the order demand quantity and driver supply quantity in different time periods within each region, etc. These data come from the actual records during the online car-hailing operation process, ensuring the authenticity and comprehensiveness of the data.

[0078] Data preprocessing can include data cleaning operations on the collected order supply information, removing outliers (such as unreasonable order prices, obviously incorrect order placement times, etc.) and missing values (processing the missing parts through appropriate filling methods, such as mean filling, median filling, etc.).

[0079] Unify different formats of time data into the standard timestamp format, perform standardized encoding on categorical data such as order type and driver identity (for example, encode preferential orders as 1, ordinary orders as 2; encode full-time drivers as 1, part-time drivers as 2, etc.), and process geographical location information according to a specific encoding method (such as hexagonal grid encoding based on the UberH3 standard), so that the data can meet the requirements of subsequent model input.

[0080] Extract relevant features from the cleaned and standardized order supply information, such as the spatio-temporal distribution features of orders (by analyzing the distribution rules of orders in different time intervals and geographical regions), the behavioral pattern features of drivers (analyzing their order acceptance preferences, the conversion rules between busy and idle in different time periods and regions based on the driver's historical order acceptance records), etc. These extracted features, together with other processed data, constitute the target input data.

[0081] Another implementation: In addition to collecting regular order supply information from the platform database, relevant supplementary information provided by third-party data sources can also be integrated, such as real-time traffic information provided by certain geographical information platforms, population flow data in specific regions, etc. These additional data can further enrich the understanding of the order supply situation.

[0082] S202, input the target input data into a pre-constructed decision support model for decision-making processing to obtain the target decision result.

[0083] Among them, the decision support model is a comprehensive model specifically constructed for the preferential decision-making of online car-hailing. It integrates a variety of advanced technologies and algorithm components, such as a deep learning prediction model including an input layer, an encoder layer, a decoder layer, an output layer, etc., and a neuron optimizer composed of a one-to-one connection layer, a decision construction layer, and an objective function construction layer. This model can deeply analyze and process the input data, comprehensively consider various factors, and finally output the optimal decision result regarding the opening of preferential order acceptance.

[0084] Among them, the target decision result includes the number and proportion of preferential order acceptance openings corresponding to the maximization of platform revenue, which is obtained by inputting the target input data into the decision support model for decision-making processing. Specifically, it clearly gives the number and proportion of preferential order acceptance openings that can achieve the maximization of platform revenue in the current operation scenario, that is, it determines how many drivers should open the preferential order acceptance function and the proportion of these drivers among all drivers, providing a scientific and reasonable decision-making basis for the online car-hailing platform and drivers.

[0085] In the embodiments of the present application, the determined target input data is accurately input into the model according to the input requirements of each layer of the decision support model. For example, for the deep learning prediction model part including an input layer, an encoder layer, a decoder layer, and an output layer, the target input data is input into the input layer in a suitable vector form, enabling it to be processed layer by layer inside the model. For instance, the encoder layer performs deep feature extraction, the decoder layer performs feature reconstruction and prediction preparation, and finally, the estimated driver order acceptance information is output at the output layer. For the neuron optimizer part, the estimated driver order acceptance information output from the deep learning prediction model and other decision-related parts of the target input data (such as environmental variables like geographical location information and time information) are input into the one-to-one connection layer, and then processed successively through the decision construction layer and the objective function construction layer.

[0086] In the deep learning prediction model, through the operations and processing of each layer, the potential relationships and feature information in the target input data are mined, and relevant information such as the number of orders accepted by each type of driver after deciding whether to turn on the preferential order listening in different spatio-temporal slices is predicted.

[0087] In the neuron optimizer, the one-to-one connection layer sorts out the input data and extracts key environmental variables. The decision construction layer performs binary classification decision processing through an improved Softmax activation function to determine the quantity tendency of a certain type of driver to turn on or not turn on the preferential order listening. The objective function construction layer, by means of a custom loss function, replaces the original neural network training objective of minimizing the prediction error with maximizing the objective function, and additionally adds a penalty term corresponding to the uneven driver earnings in the objective function in the form of Lagrangian relaxation. After comprehensively considering various factors, the opening quantity and proportion of the preferential order listening corresponding to the maximization of the platform revenue are obtained as the target decision result.

[0088] Another implementation method: A distributed computing architecture can be adopted to split the target input data into multiple sub-datasets and input them into the replicas of the decision support model on different computing nodes simultaneously. This can improve the data processing speed, especially suitable for the case of large-scale online car-hailing operation data. Each computing node processes according to the same model architecture and parameter settings, and finally, the processing results of each node are summarized and integrated.

[0089] S203, sending recommendation information to at least one target terminal according to the target decision result; the recommendation information is used to recommend that the user of the target terminal turn on the preferential service.

[0090] Among them, in the online car-hailing operation system, the target terminal refers to various terminal devices related to the online car-hailing business, including but not limited to the mobile client used by passengers (such as mobile phone APP), the mobile device where the driver-side APP used by the driver is located, etc. These terminal devices are the carriers for receiving and displaying recommended information, so that users and drivers can make corresponding decisions based on the recommendations.

[0091] In the embodiment of the present application, according to the target decision result, that is, the opening quantity and ratio of preferential order receiving corresponding to the maximum platform revenue, recommended information for different target terminals is generated. For the mobile client used by passengers, the generated recommended information may include the advantages of preferential services (such as fixed price, more secure journey, etc.), the quantity and ratio of drivers currently opening preferential order receiving, so as to attract passengers to choose preferential services. For the mobile device where the driver-side APP used by the driver is located, the generated recommended information may include the expected revenue increase of opening preferential order receiving, the ratio of other drivers currently opening preferential order receiving, etc., to encourage drivers to turn on the preferential order receiving function.

[0092] The generated recommended information can be accurately sent to at least one target terminal through the communication module of the online car-hailing platform. The communication module can select an appropriate communication protocol (such as HTTP, WebSocket, etc.) for information transmission according to factors such as the device type and network connection status of the target terminal, to ensure that the recommended information can reach the target terminal users in a timely and accurate manner.

[0093] The above online car-hailing preferential decision-making method, device and computer device determine target input data based on the pre-acquired order supply information; input the target input data into a pre-constructed decision support model for decision-making processing to obtain a target decision result including the opening quantity and ratio of preferential order receiving corresponding to the maximum platform revenue; and send recommended information for recommending users of the target terminal to turn on preferential services to at least one target terminal according to the target decision result. The present application can deeply analyze the complex relationships in the data and accurately predict the opening quantity and ratio of preferential order receiving corresponding to the maximum platform revenue as the target decision result. This not only helps the server achieve the maximization of economic benefits, but also takes into account the interests of the driver group while ensuring the platform revenue.

[0094] In an exemplary embodiment, based on the above embodiment, please refer to Figure 3 , the target input data of the embodiment of the present application includes global spatio-temporal information, environmental variables and the number of drivers before decision-making; the decision support model includes a one-to-one connection layer, an encoder layer, a decision layer and a fully connected layer. The embodiment of the present application relates to the process of inputting the target input data into a pre-constructed decision support model for decision-making processing to obtain a target decision result, including the following steps:

[0095] S301. Input the environmental variables into a one-to-one connection layer for feature extraction processing to obtain feature data.

[0096] Among them, feature extraction includes at least one of geographical location feature extraction, time feature extraction, user feature extraction, and order-related feature extraction.

[0097] In the embodiments of the present application, first, ensure that the format of the environmental variable data meets the input requirements of the one-to-one connection layer. For geographical location information, perform encoding processing according to a preset encoding method (such as hexagonal grid encoding based on the UberH3 standard) so that it is input in a standard encoding format; for time information, uniformly convert it into a standard timestamp format; for user information and order-related information, perform normalization encoding (such as encoding different types of users into digital forms according to specific rules and performing similar encoding on order types, etc.).

[0098] The one-to-one connection layer uses built-in geographical information processing algorithms to analyze the input geographical location information (such as the starting location of an order, the destination location, the current location of the driver, etc.). By calculating the distance between geographical locations, the supply-demand relationship characteristics of the areas to which they belong (such as a certain area is a commercial center, and the order demand is usually high but the driver supply may also be large), etc., geographical location features are extracted. For example, extract the distance feature between the starting location and the destination location of an order, the order density feature in the area where it is located (that is, the number of orders per unit area), etc.

[0099] For time information (such as the current time point, the order placement time, whether it is a working day, etc.), by analyzing the rules of time series data, time features are extracted. For example, determine whether the current time is in the peak period, off-peak period, or night period, calculate the time interval between the order placement time and the current time, and analyze the change trend of order demand in different time periods, etc., so as to obtain time features such as strong order demand during the peak period and low order demand during the night period.

[0100] For user information (such as user type, user consumption habits, etc.), the one-to-one connection layer analyzes the user's consumption preferences (such as whether they prefer discounted orders, whether they often take taxis in a specific area, etc.), the user's activity level (such as the taxi-taking frequency within a certain time period), etc. based on data such as the user's historical order records, and extracts user features. For example, determining that a certain user is a price-sensitive user and often chooses a more preferential travel mode is a kind of user feature.

[0101] Analyze the price range distribution of orders (e.g., the prices of preferential orders are relatively fixed, while the prices of ordinary orders may fluctuate greatly) and the distribution rules of order types (e.g., the proportion of preferential orders is higher in certain regions) from order-related information (such as order type, order price, etc.), and extract order-related features. For example, obtain order-related features such as the average price of preferential orders in a certain region and the quantity ratio of ordinary orders to preferential orders.

[0102] In S302, the number of drivers before decision-making is input into the decision-making layer for binary classification decision-making processing to obtain decision-making information.

[0103] When the decision-making layer in the embodiment of the present application performs binary classification decision-making processing, first, the weight parameters of the improved Softmax activation function are initialized. These weight parameters can be randomly initialized by assigning random initial values within a certain range (for example, for a certain weight parameter, its initial value is randomly set within the range of [-0.5, 0.5]). Then, through the improved Softmax formula, it is further processed to convert it into two output values, corresponding to the "probability" (not strictly a probability here, but a relative decision-making tendency degree) of a certain type of driver to turn on the preferential order receiving and the "probability" of not turning on the preferential order receiving, respectively.

[0104] Based on the operation results of the above-mentioned improved Softmax activation function, preliminary decision-making results on whether various types of drivers turn on or do not turn on the preferential order receiving are obtained. That is, the classification situation of whether each type of driver is more inclined to turn on or not turn on the preferential order receiving is clarified, which is an important part of the decision-making information.

[0105] Another implementation method: In addition to using the improved Softmax activation function for binary classification decision-making processing, other advanced binary classification algorithms can also be used, such as the binary classification version of the support vector machine (SVM), logistic regression, etc. Taking the support vector machine as an example, first, feature selection and preprocessing need to be performed on the number of drivers before decision-making data to make it meet the input requirements of the SVM. Then, through the trained SVM model, according to the input data features, the drivers are divided into two categories: those who turn on the preferential order receiving and those who do not turn on the preferential order receiving, and at the same time, the decision scores corresponding to each category can be obtained (similar to the relative decision-making tendency degree output by Softmax).

[0106] In S303, the global spatio-temporal information is input into the encoder layer for global spatio-temporal supply and demand information extraction processing to obtain the global spatio-temporal supply and demand vector.

[0107] In the embodiments of the present application, the global spatio-temporal information is prepared in the input format required by the encoder layer. For the time information part, ensure that it has been uniformly converted into the standard timestamp format; for the geographical area information part, encode it according to the preset encoding method (such as the hexagonal grid encoding based on the Uber H3 standard) so that it can be input into the encoder layer in a standardized encoding format.

[0108] The encoder layer first divides the input global spatio-temporal information and subdivides it according to the two dimensions of time and space. For example, in the time dimension, a day is divided into several time periods (such as peak hours, off-peak hours, night hours, etc.); in the space dimension, the entire operation area is divided into multiple small grid areas according to the geographical area encoding. Then, the supply and demand information such as the number of order demands and the number of driver supplies within each spatio-temporal slice (i.e., the combination of each specific time period and specific small grid area) is statistically analyzed.

[0109] By analyzing the supply and demand information within each spatio-temporal slice, relevant features are extracted, such as the supply-demand imbalance feature that the order demand is strong but the driver supply is insufficient within a certain spatio-temporal slice, and the change trend feature of the supply-demand relationship between different spatio-temporal slices. Then, these extracted features are encoded according to a certain order and rules to form a vector form, that is, the global spatio-temporal supply-demand vector. This vector can comprehensively reflect the dynamic supply-demand relationship in the entire operation area in the spatio-temporal dimension and provide important input data for the subsequent processing of the fully connected layer.

[0110] S304, input the feature data, the global spatio-temporal supply-demand vector, and the decision-making information into the fully connected layer for feature integration and feature extraction processing to obtain the target decision result.

[0111] In the embodiments of the present application, each neuron in the fully connected layer is connected to all neurons in the previous layer, forming a dense connection network. When the input data enters the fully connected layer, each neuron will perform a weighted sum processing on the input information from the feature data, the global spatio-temporal supply-demand vector, and the decision-making information.

[0112] After completing the weighted sum, the neurons in the fully connected layer will also perform a non-linear transformation on the integrated input information to increase the expression ability of the model. Usually, an activation function is used to achieve this purpose.

[0113] In the embodiments of the present application, the fully connected layer performs comprehensive feature integration and feature extraction processing on the feature data, the global spatio-temporal supply-demand vector, and the decision-making information, which can fully explore the potential relationships between these data and the deep feature information related to the target decision result. Whether through conventional weighted summation, non-linear transformation, and deep feature mining methods, or more advanced processing methods based on attention mechanisms, ensemble learning, etc., it helps to more accurately determine target decision results such as the opening quantity and proportion of preferential order listening, so as to provide a more accurate decision-making basis for the online car-hailing platform to maximize the platform revenue.

[0114] In an exemplary embodiment, based on the above embodiment, please refer to Figure 4 , the decision-making information in the embodiments of the present application includes the proportions of different types of drivers who turn on and turn off the preferential order listening; the process of inputting the number of drivers before decision-making into the decision-making layer for binary classification decision-making processing to obtain the decision-making information includes the following steps:

[0115] S401, perform a weighted summation operation on the pre-acquired weight parameters and the number of drivers before decision-making to obtain an intermediate result.

[0116] In the embodiments of the present application, after obtaining the number of drivers before decision-making data, it is first grouped. The drivers are divided into different groups according to different attributes of the drivers (such as service duration, credit rating, driving vehicle type, etc.), and the drivers within each group have similar characteristics. Then, corresponding weight parameters are determined for each group respectively. These weight parameters are also determined comprehensively based on factors such as historical operation data, platform policies, and relevant experience, but will more carefully consider the specific attributes of each group of drivers.

[0117] For example, for the group of full-time drivers with a long service duration and a high credit rating, its weight parameter may be relatively high, reflecting the importance of such drivers in decision-making; while for the group of part-time drivers with a small vehicle type and a short service duration, its weight parameter may be relatively low.

[0118] S402, input the intermediate result into a pre-constructed activation function for binary classification decision-making processing to obtain the proportions of different types of drivers who turn on and turn off the preferential order listening.

[0119] In the embodiments of the present application, in addition to the improved Softmax activation function, the decision function of the support vector machine (SVM) can also be considered as an alternative activation function for binary classification decision-making. Before using the SVM decision function, it is necessary to train the SVM, collect the actual order-receiving situations of different types of drivers under different settings in the historical operation data (including the situations of turning on and not turning on the preferential order-receiving), and use them as training samples. By adjusting the kernel function of the SVM (such as linear kernel, polynomial kernel, radial basis kernel, etc.) and related parameters (such as penalty coefficient, etc.), the SVM can accurately distinguish the order-receiving tendencies of different types of drivers under different settings.

[0120] In the embodiments of the present application, a weighted sum operation is performed on the number of drivers before decision-making, and binary classification decision-making is performed in combination with the activation function, so as to accurately obtain the proportions of different types of drivers turning on and turning off the preferential order-receiving. These probability information provides a precise decision-making basis for the online car-hailing platform to formulate the preferential order-receiving opening strategy, enabling the platform to scientifically and reasonably determine the number of different types of drivers turning on or turning off the preferential order-receiving in different spatio-temporal slices according to the situations of different types of drivers, so as to maximize the platform's revenue while taking into account the interests of drivers and passengers.

[0121] In an exemplary embodiment, based on the above embodiment, please refer to Figure 5 , the method of the embodiments of the present application further includes the following steps:

[0122] S501, perform data preprocessing on the pre-acquired historical operation data to obtain training data.

[0123] Among them, the data preprocessing includes format processing, hexagonal grid coding conversion processing, and data partitioning processing.

[0124] After the embodiments of the present application obtain the historical operation data from the database of the online car-hailing platform, they first process various inconsistent formats in the data. For example, for time data, different format time records (such as date-time format, timestamp format, etc.) are uniformly converted into the standard timestamp format to facilitate accurate analysis and processing in the time dimension later.

[0125] For categorical data, such as driver identity (full-time, part-time, etc.), order type (preferential order, ordinary order, etc.), etc., perform standardized coding processing. Different categories are respectively assigned unique integer codes. For example, full-time drivers are coded as 1, part-time drivers are coded as 2; preferential orders are coded as 1, ordinary orders are coded as 2, etc., so that the categorical data can participate in subsequent calculations and analyses in digital form.

[0126] For the geographical location information in historical operation data, such as the starting location, destination location of orders, and the location of drivers, etc., a hexagonal grid coding method is adopted for processing. Based on specific geographical area division criteria (such as the hexagonal grid division of the UberH3 standard), each geographical location is mapped to the corresponding hexagonal grid cell, and a unique code is assigned to each grid cell.

[0127] Through this coding conversion, the geographical location information is transformed into a form that is more convenient for processing and analysis, which can better reflect the relationships between geographical locations and the supply and demand situations in different regions. For example, by counting information such as the number of orders and the number of drivers in each grid cell, the operation popularity and supply-demand balance in different regions can be analyzed.

[0128] The historical operation data after format processing and hexagonal grid coding conversion is divided into different subsets according to a certain ratio, usually divided into a training set, a test set, and a validation set. Common division ratios are 7:2:1 (training set:test set:validation set), etc.

[0129] The training set will be used to be input into the initial prediction model for training later, enabling the model to learn the rules and features in the data; the test set is used to preliminarily evaluate the performance of the model during the training process to assist in judging whether problems such as overfitting occur in the model; the validation set is used to verify the effect of the finally generated model after the model training is completed to ensure that the model can also show good performance on the data that has not participated in the training.

[0130] S502, Input the training data into the initial prediction model for prediction processing to obtain the initial prediction result.

[0131] In the embodiments of this application, the training data is input into the initial prediction model in the form of batches (batch) in sequence. When inputting a batch of training data each time, the model calculates according to the current parameter settings and the input data, and through the interaction of neurons in each layer, operations such as feature extraction and transformation are performed on the input data, and finally the initial prediction result regarding the driver's decision-making situation is obtained at the output layer.

[0132] S503, Adjust the parameters in the initial prediction model according to the initial prediction result and the mean absolute value function to determine the decision support model.

[0133] In the embodiments of this application, first, obtain the actual driver decision-making situation data, which can be extracted from the historical operation data of the online car-hailing platform and corresponds to the training data used for training, that is, the actual driver decision-making situation under the same spatio-temporal slice.

[0134] Then, compare the initial prediction result with the actual driver's decision situation, and use the mean absolute value function to calculate the difference between the two, that is, the loss value. According to the calculated loss value, use an optimization algorithm to adjust the parameters in the initial prediction model. By continuously repeating this update process, as the training data is continuously input and the loss value is continuously optimized, the parameters of the initial prediction model will gradually approach the optimal value, thereby continuously improving the prediction performance of the model. When the prediction performance of the model reaches a certain standard (such as the performance on the test set and the validation set meets the requirements), the model at this time is the decision support model.

[0135] In the embodiments of the present application, through data preprocessing of historical operation data, including steps such as format processing, hexagonal grid coding conversion processing, and data partitioning processing, the original historical operation data that is not standardized in format and difficult to effectively utilize can be transformed into training data that is more suitable for model training and analysis. This greatly improves the usability of the data and lays a good foundation for building an accurate decision support model.

[0136] In an exemplary embodiment, based on the above embodiment, the method of the embodiments of the present application further includes calculating the maximum platform revenue value according to the target decision result and a pre-established objective function; the objective function includes the corresponding relationship between the total platform revenue and at least one of the order price, the platform commission rate, and the operation cost.

[0137] In the embodiments of the present application, in addition to obtaining the number and proportion of preferential order acceptance openings in the target decision result, the driver group is further segmented, and the drivers are divided into different groups according to different attributes of the drivers (such as service duration, credit rating, driving vehicle type, etc.). The drivers within each group have similar characteristics. Then, for each driver group, the estimated number of successful order pickups under different order types is determined respectively (obtained through comprehensive analysis based on historical data, current operation conditions, etc.).

[0138] For the objective function, on the basis of considering factors such as the total platform revenue, the order price, the platform commission rate, and the operation cost, some additional influencing factors are introduced, such as the degree of driver revenue imbalance, the order cancellation rate, and the passenger satisfaction. By quantifying these factors (such as determining the quantification index of the driver revenue imbalance degree according to historical data, calculating the order cancellation rate according to the order cancellation records, and calculating the passenger satisfaction according to the passenger evaluations), and incorporating them into the construction of the objective function.

[0139] In the embodiments of the present application, by calculating according to the target decision result and the pre-established objective function, the maximum platform revenue value can be accurately obtained. This calculation method comprehensively considers factors closely related to platform revenue such as order price, platform commission rate, and operating cost, as well as the relationships between them, making the evaluation of platform revenue more accurate and providing a reliable basis for the online car-hailing platform to understand its profit situation under different decision-making scenarios.

[0140] In an exemplary embodiment, based on the above embodiment, please refer to Figure 6 , the training process of the objective function in the embodiments of the present application includes the following steps:

[0141] S601, Obtain a plurality of samples, where the samples include the number of drivers of various types before decision-making, environmental information, and actual driver revenue data.

[0142] In the embodiments of the present application, format-uniform processing is performed on the collected samples of the number of drivers of various types before decision-making. For example, quantity records in different formats (such as some recorded as integers and some may have decimal parts) are uniformly converted into integer form for subsequent calculation and processing.

[0143] Perform standardization processing on the environmental information. Re-encode the geographical location information according to a specific coding method (such as hexagonal grid coding based on the UberH3 standard), convert the time point information into the standard timestamp format, and perform normalization processing on numerical data such as real-time demand information and supply-demand situations in each region, so that its value range is within the interval [0,1], improving the consistency and comparability of the data.

[0144] Perform cleaning processing on the actual driver revenue data. Check whether there are outliers (such as too high or too low revenue values may be caused by data entry errors or special situations), and use appropriate processing methods (such as mean filling, median filling, etc.) to process the outliers to ensure the rationality and reliability of the data.

[0145] S602, For each iteration process, determine the initial function value according to the number of drivers of various types before decision-making, environmental information, and actual driver revenue data.

[0146] In the embodiments of the present application, in each iteration process, the currently obtained number of drivers of various types before decision-making, environmental information, and actual driver revenue data are substituted into the expression of the initial function for calculation to determine the initial function value.

[0147] S603, Update the parameters of the initial function according to the initial function value and the optimization algorithm.

[0148] Embodiments of the present application select appropriate optimization algorithms to update the parameters of the initial function. Common optimization algorithms include Stochastic Gradient Descent (SGD), Adagrad algorithm, Adadelta algorithm, etc. Taking the Stochastic Gradient Descent algorithm as an example, before use, its relevant parameters need to be determined, such as the learning rate. The learning rate is used to control the speed of parameter update, and usually, appropriate values are determined based on experience or through some preliminary experiments. Then, according to the update formula of the Stochastic Gradient Descent algorithm, each parameter is updated.

[0149] S604. When the current iteration count meets the maximum iteration count, determine the initial function at the current iteration count as the target function.

[0150] In the training process of the target function in embodiments of the present application, a counter is set to record the current iteration count. Starting from the beginning of training, each time a complete round of parameter update process (as described in step S603) is completed, the iteration count counter is incremented by 1.

[0151] Meanwhile, the value of the maximum iteration count is preset in advance. This maximum iteration count is determined comprehensively based on factors such as experience, data scale, and expectations for training effects. For example, for large-scale online car-hailing operation data and a relatively high expected accuracy of the target function, the maximum iteration count may be set to 1000 times; while for relatively small-scale data or preliminary exploratory training, the maximum iteration count may be set to 500 times, etc.

[0152] After each iteration ends, compare the current iteration count with the maximum iteration count. When the current iteration count reaches or exceeds the maximum iteration count, it is considered that the condition for determining the target function is met.

[0153] At this time, directly determine the initial function at the current iteration count as the target function. That is to say, after multiple rounds of iterative training before, continuously updating the parameters of the initial function according to the sample data (step S603), when reaching the maximum iteration count, the current initial function that has been optimized and adjusted multiple times is regarded as the target function that has completed training. It will be used in subsequent relevant application scenarios such as online car-hailing discount decisions and can relatively accurately reflect the impacts of various factors on aspects such as platform revenue and driver revenue.

[0154] In an exemplary embodiment, based on the above embodiment, please refer to Figure 7 , the process of determining the initial function value in embodiments of the present application involves the number of various types of drivers, environmental information, and actual driver revenue data before the decision, including the following steps:

[0155] S701. Obtain the driver type weight coefficient, environmental type weight coefficient, and penalty term coefficient.

[0156] In the embodiments of the present application, the above coefficients are determined in advance before the training of the objective function and can be stored in the database of the online car-hailing platform in a specific data format (such as in tabular form, where each row corresponds to a driver type or an environment type, and the columns are relevant information such as coefficient values). When determining the initial function value for each iterative calculation, these coefficients are directly read from the database to ensure their consistency and accuracy.

[0157] S702. Multiply the number of drivers of each type before decision-making by the weight coefficient of the driver type to obtain a first function value.

[0158] In the embodiments of the present application, according to the rules of vector multiplication, multiply the number of drivers of each type before decision-making by the weight coefficient of the driver type to calculate the first function value.

[0159] S703. Multiply the environmental information by the weight coefficient of the environment type to obtain a second function value.

[0160] In the embodiments of the present application, the environmental information is combined with the corresponding weight coefficient to obtain a second function value that reflects the other part of the contribution of environmental factors to the objective function value after considering its weight.

[0161] S704. Perform revenue analysis processing on the actual driver revenue data to obtain a revenue value, and multiply the revenue value by the penalty term coefficient to obtain a penalty value.

[0162] In the embodiments of the present application, the actual driver revenue data is obtained from the database of the online car-hailing platform or other relevant data sources. These data are statistically obtained based on information such as the driver's identity identifier, order type, and time of completing the order, and represent the actual revenues of different types of drivers in different time periods and different order types (such as discounted orders, ordinary orders, etc.). For example, for a discounted order completed by a full-time driver, calculate his actual revenue on this order (considering factors such as platform commission); for an ordinary order completed by a part-time driver, calculate his actual revenue in the same way. Analyze and process the obtained actual driver revenue data to obtain a revenue value that can comprehensively reflect the revenue status of drivers in actual operations. Various analysis methods can be used, such as calculating the average revenue of different types of drivers in different time periods, calculating the standard deviation of the revenues of each type of driver to measure the revenue fluctuation, etc., and then according to these analysis results, combine them through a certain formula or algorithm to obtain the revenue value.

[0163] S705. Determine the initial function value according to the first function value, the second function value, and the penalty value.

[0164] In the embodiments of the present application, within the overall framework of objective function training, a calculation method for the initial function value is preset to be associated with the general structure of the objective function. Generally speaking, the determination of the initial function value is for subsequent iterative training to gradually optimize the parameters of the objective function so that it can accurately reflect the impacts of various factors in the online car-hailing operation on aspects such as platform revenue and driver revenue. According to the expression of the preset initial function, substituting the already calculated first function value, second function value, and penalty value into the calculation can obtain the initial function value.

[0165] In the embodiments of the present application, by comprehensively considering the first function value (related to the number of drivers), the second function value (related to environmental information), and the penalty value (related to the actual income situation of drivers) to determine the initial function value, it can comprehensively and accurately reflect the impacts of various factors in the online car-hailing operation on the overall operation effect. Whether in the conventional static weight setting method or the dynamic weight adjustment method, factors in different aspects can be reasonably incorporated into the calculation of the initial function value, making the initial function value an effective indicator to reflect the comprehensive effect of various factors in the current operation scenario and providing a reliable basis for the subsequent training and optimization of the objective function.

[0166] In an exemplary embodiment, based on the above embodiments, the embodiments of the present application relate to the process of determining target input data based on pre-acquired order supply information, including: performing data preprocessing on the order supply information to obtain target input data.

[0167] Among them, the data preprocessing includes at least one of time-space division, driver classification, driver status judgment, and driver order-taking statistics.

[0168] In the embodiments of the present application, the release time information of the order is extracted from the order supply information, and the time dimension is divided according to a certain time interval standard. For example, a day can be divided into several time periods, such as peak hours (7:00 - 9:00 in the morning, 5:00 - 7:00 in the afternoon), off-peak hours (9:00 in the morning - 5:00 in the afternoon, 7:00 - 10:00 in the evening), night hours (10:00 in the evening - 7:00 in the next morning), etc. Through this time division, it is possible to better analyze the laws and characteristics of order supply in different time periods. For example, the order demand is strong during peak hours and relatively stable during off-peak hours.

[0169] For the starting position and destination position information in the order supply information, a suitable geographical area division method is used for space division. Such as Figure 8As shown, a common method is to divide the entire operation area into multiple hexagonal grid units based on the UberH3 standard. Each unit corresponds to a specific geographical area. In this way, each order can be mapped to the corresponding grid unit, facilitating the analysis of spatial distribution characteristics such as the density of order supply, the supply-demand balance, etc. in different regions.

[0170] Classify drivers according to different attributes of the drivers. These attributes can include but are not limited to the driver's identity (full-time driver, part-time driver), service duration (e.g., classified as new drivers (service duration less than a certain duration, such as 3 months), experienced drivers (service duration greater than or equal to 3 months)), vehicle type driven (e.g., small car, medium car, large car), etc. By classifying drivers, it is possible to more meticulously analyze differences in the order-taking situations, preferences, etc. of different types of drivers when facing order supply. For example, full-time drivers may be more inclined to take high-value orders during peak hours, while part-time drivers may pay more attention to the flexibility of order-taking.

[0171] Obtain the real-time status information of drivers from the order supply information and relevant platform data, such as whether the driver is online, whether they are in a busy state (picking up or dropping off passengers), whether they are resting, etc. By accurately judging the driver's status, it is possible to understand the actual situation of available drivers for order-taking. For example, during a certain period, if a large number of drivers are in a busy state, then the order supply may be relatively excessive, affecting the order matching efficiency. At the same time, combined with the spatio-temporal division and driver classification, it is also possible to analyze the response of different types of drivers to order supply at different times and in different states.

[0172] Based on the order supply information and the order matching records of the platform, count indicators such as the number of orders taken and the order-taking success rate of different types of drivers in different time periods and different regions. For example, count the number of orders taken and the order-taking success rate of full-time drivers during peak hours in a specific region. Through these statistical data, it is possible to intuitively understand the order-taking ability and efficiency of drivers in different situations, providing an important basis for subsequent analysis of the relationship between order supply and driver order-taking.

[0173] After the above data preprocessing operations such as spatio-temporal division, driver classification, driver status judgment, and driver order-taking statistics, integrate the processed various data in a certain format to form the target input data. For example, the relevant data of different time periods, different regions, and different types of drivers can be organized into a multi-dimensional array or a data table form, where each row represents a specific spatio-temporal - driver type combination, and each column corresponds to different attribute indicators (such as the number of orders taken, the order-taking success rate, the order supply volume, etc.). Such target input data can comprehensively and accurately reflect various key factors in the operation of online car-hailing services and can be used as an effective input for subsequent analysis.

[0174] In an exemplary embodiment, based on the above embodiments, based on Figure 9 the neural network prediction model shown, the method of the embodiments of the present application further includes the following steps:

[0175] Step 1: Perform data preprocessing on the pre-acquired historical operation data to obtain training data; input the training data into the initial prediction model for prediction processing to obtain an initial prediction result; adjust the parameters in the initial prediction model according to the initial prediction result and the mean absolute value function to determine the decision support model; the decision support model includes a one-to-one connection layer, an encoder layer, a decision layer, and a fully connected layer;

[0176] Step 2: Obtain samples of the number of drivers of various types before decision-making, environmental information samples, and samples of the actual income data of drivers; for each iteration process, obtain the driver type weight coefficient, the environmental type weight coefficient, and the penalty term coefficient; perform a product operation on the number of drivers of various types before decision-making and the driver type weight coefficient to obtain a first function value; perform a product operation on the environmental information and the environmental type weight coefficient to obtain a second function value; perform income analysis processing on the actual income data of drivers to obtain an income value, and perform a product operation on the income value and the penalty term coefficient to obtain a penalty value; determine the initial function value according to the first function value, the second function value, and the penalty value;

[0177] Step 3: Update the parameters of the initial function according to the initial function value and the optimization algorithm; in the case where the current iteration number meets the maximum iteration number, determine the initial function at the current iteration number as the target function;

[0178] Step 4: Perform data preprocessing on the order supply information to obtain target input data; input the environmental variables in the target input data into the one-to-one connection layer for feature extraction processing to obtain feature data;

[0179] Step 5: Input the number of drivers before decision-making into the decision layer, and the decision layer performs a weighted summation operation on the pre-acquired weight parameters and the number of drivers before decision-making in the target input data to obtain an intermediate result; input the intermediate result into the pre-constructed activation function for binary classification decision processing to obtain the proportions of various types of drivers to open and close the preferential order listening;

[0180] Step 6: Input the global spatio-temporal information in the target input data into the encoder layer for global spatio-temporal supply and demand information extraction processing to obtain a global spatio-temporal supply and demand vector;

[0181] Step 7: Input the feature data, the global spatio-temporal supply and demand vector, and the decision information into the fully connected layer for feature integration and feature extraction processing to obtain a target decision result; calculate the maximum platform income value according to the target decision result and the pre-established target function.

[0182] It should be understood that although the steps in the flowcharts involved in the above-described embodiments are sequentially shown according to the indications of the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless there is a clear indication in this article, the execution of these steps has no strict order limit, and these steps can be executed in other orders. Moreover, at least a part of the steps in the flowcharts involved in the above-described embodiments may include multiple steps or multiple stages. These steps or stages are not necessarily executed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be executed alternately or in turn with at least a part of other steps or steps or stages in other steps.

[0183] Based on the same inventive concept, an embodiment of the present application also provides a ride-hailing discount decision-making device for implementing the ride-hailing discount decision-making method described above. The solution provided by this device for solving problems is similar to the solution described in the above method. Therefore, the specific limitations in one or more embodiments of the ride-hailing discount decision-making device provided below can refer to the limitations on the ride-hailing discount decision-making method in the above text, and will not be repeated here.

[0184] In one embodiment, as Figure 10 shown, a ride-hailing discount decision-making device is provided, including:

[0185] A data determination module 801, configured to determine target input data based on pre-acquired order supply information;

[0186] A decision-making module 802, configured to input the target input data into a pre-constructed decision support model for decision-making processing to obtain a target decision result; the target decision result includes the number and proportion of discounted order acceptance openings corresponding to maximizing platform revenue;

[0187] An information sending module 803, configured to send recommendation information to at least one target terminal according to the target decision result, and the recommendation information is used to recommend that the users of the target terminal turn on the discount service.

[0188] In one of the embodiments, the above target input data includes global spatio-temporal information, environmental variables, and the number of drivers before decision-making; the decision support model includes a one-to-one connection layer, an encoder layer, a decision layer, and a fully connected layer. The above decision-making module includes:

[0189] A feature extraction unit, configured to input the environmental variables into the one-to-one connection layer for feature extraction processing to obtain feature data; the feature extraction includes at least one of geographical location feature extraction, time feature extraction, user feature extraction, and order-related feature extraction;

[0190] A binary classification decision unit for inputting the number of drivers before decision into the decision layer for binary classification decision processing to obtain decision information;

[0191] An information extraction unit for inputting global spatio-temporal information into the encoder layer for global spatio-temporal supply and demand information extraction processing to obtain a global spatio-temporal supply and demand vector;

[0192] An integration and extraction unit for inputting feature data, the global spatio-temporal supply and demand vector, and decision information into a fully connected layer for feature integration and feature extraction processing to obtain a target decision result.

[0193] In one embodiment, the above decision information includes the proportions of various types of drivers opening and closing preferential order listening; the binary classification decision unit includes:

[0194] A summation operation sub-unit for performing weighted summation operation processing on pre-acquired weight parameters and the number of drivers before decision to obtain an intermediate result;

[0195] A binary classification decision sub-unit for inputting the intermediate result into a pre-constructed activation function for binary classification decision processing to obtain the proportions of various types of drivers opening and closing preferential order listening.

[0196] In one embodiment, the above device further includes:

[0197] A data preprocessing module for preprocessing pre-acquired historical operation data to obtain training data; the data preprocessing includes format processing, hexagonal grid coding conversion processing, and data partitioning processing;

[0198] A prediction module for inputting the training data into an initial prediction model for prediction processing to obtain an initial prediction result;

[0199] A parameter adjustment module for adjusting the parameters in the initial prediction model according to the initial prediction result and the mean absolute value function to determine a decision support model.

[0200] In one embodiment, the above device further includes:

[0201] A revenue determination module for calculating the maximum platform revenue value according to the target decision result and a pre-established target function; the target function includes the corresponding relationship between the total platform revenue and at least one of the order price, the platform commission rate, and the operating cost.

[0202] In one embodiment, the above device includes:

[0203] A sample acquisition module for acquiring a plurality of samples, where the samples include the number of various types of drivers before decision, environmental information, and driver actual revenue data;

[0204] An initial value determination module, configured to determine an initial function value for each iteration process according to the number of drivers of various types before decision-making, environmental information, and actual driver revenue data;

[0205] An update module, configured to update the parameters of the initial function according to the initial function value and an optimization algorithm;

[0206] A function determination module, configured to determine the initial function at the current iteration as the target function when the current iteration count meets the maximum iteration count.

[0207] In one embodiment, the above initial value determination module includes:

[0208] A coefficient acquisition unit, configured to acquire a driver type weight coefficient, an environmental type weight coefficient, and a penalty term coefficient;

[0209] A first function value acquisition unit, configured to perform a multiplication process on the number of drivers of various types before decision-making and the driver type weight coefficient to obtain a first function value;

[0210] A second function value acquisition unit, configured to perform a multiplication process on the environmental information and the environmental type weight coefficient to obtain a second function value;

[0211] A penalty value determination unit, configured to perform revenue analysis processing on the actual driver revenue data to obtain a revenue value, and perform a multiplication process on the revenue value and the penalty term coefficient to obtain a penalty value;

[0212] An initial value determination unit, configured to determine the initial function value according to the first function value, the second function value, and the penalty value.

[0213] In one embodiment, the above data determination module includes:

[0214] A data determination unit, configured to perform data preprocessing on the order supply information to obtain target input data; the data preprocessing includes at least one of space-time partitioning, driver classification, driver status judgment, and driver order receiving statistics.

[0215] Each module in the above online car-hailing discount decision device can be implemented in whole or in part by software, hardware, and their combination. The above modules can be embedded in or independent of a processor in a computer device in the form of hardware, or stored in a memory in the computer device in the form of software, so as to facilitate the processor to call and execute the operations corresponding to the above respective modules.

[0216] In one embodiment, a computer device is provided. The computer device can be a server, and its internal structure diagram can be as Figure 11As shown in the figure. The computer device includes a processor, a memory, an input / output interface (Input / Output, abbreviated as I / O), and a communication interface. Among them, the processor, the memory, and the input / output interface are connected through a system bus, and the communication interface is connected to the system bus through the input / output interface. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program, and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The input / output interface of the computer device is used to exchange information between the processor and external devices. The communication interface of the computer device is used to communicate with external terminals through a network connection. When the computer program is executed by the processor, it implements a method for making preferential decisions for online car-hailing.

[0217] Those skilled in the art can understand that Figure 11 the structure shown in the figure is only a block diagram of some structures related to the solution of this application, and does not constitute a limitation on the computer device to which the solution of this application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine some components, or have different component arrangements.

[0218] In one embodiment, a computer device is provided, including a memory and a processor. A computer program is stored in the memory. When the processor executes the computer program, the following steps are implemented:

[0219] Based on the pre-acquired order supply information, determine the target input data;

[0220] Input the target input data into a pre-constructed decision support model for decision-making processing to obtain a target decision result; the target decision result includes the number and proportion of preferential order-taking openings corresponding to the maximization of platform revenue;

[0221] According to the target decision result, send recommendation information to at least one target terminal. The recommendation information is used to recommend that the users of the target terminal turn on the preferential service.

[0222] In one embodiment, the above-mentioned target input data includes global spatio-temporal information, environmental variables, and the number of drivers before decision-making. The decision support model includes a one-to-one connection layer, an encoder layer, a decision layer, and a fully connected layer. When the processor executes the computer program, the following steps are also implemented:

[0223] Input the environmental variables into the one-to-one connection layer for feature extraction processing to obtain feature data; the feature extraction includes at least one of geographical location feature extraction, time feature extraction, user feature extraction, and order-related feature extraction;

[0224] Input the number of drivers before decision-making into the decision-making layer for binary classification decision-making processing to obtain decision-making information;

[0225] Input the global spatio-temporal information into the encoder layer for global spatio-temporal supply and demand information extraction processing to obtain the global spatio-temporal supply and demand vector;

[0226] Input the feature data, the global spatio-temporal supply and demand vector, and the decision-making information into the fully connected layer for feature integration and feature extraction processing to obtain the target decision result.

[0227] In one embodiment, when the processor executes the computer program, the following steps are further implemented:

[0228] Perform a weighted summation operation on the pre-acquired weight parameters and the number of drivers before decision-making to obtain an intermediate result;

[0229] Input the intermediate result into a pre-constructed activation function for binary classification decision-making processing to obtain the proportions of various types of drivers to turn on and turn off the preferential order receiving.

[0230] In one embodiment, when the processor executes the computer program, the following steps are further implemented:

[0231] Perform data preprocessing on the pre-acquired historical operation data to obtain training data; the data preprocessing includes format processing, hexagonal grid coding conversion processing, and data partitioning processing;

[0232] Input the training data into the initial prediction model for prediction processing to obtain the initial prediction result;

[0233] Adjust the parameters in the initial prediction model according to the initial prediction result and the mean absolute value function to determine the decision support model.

[0234] In one embodiment, when the processor executes the computer program, the following steps are further implemented:

[0235] Calculate the maximum platform revenue value according to the target decision result and the pre-established objective function; the objective function includes the corresponding relationship between the total platform revenue and at least one of the order price, the platform commission rate, and the operating cost.

[0236] In one embodiment, when the processor executes the computer program, the following steps are further implemented:

[0237] Obtain multiple samples, where the samples include the number of various types of drivers before decision-making, environmental information, and the actual revenue data of the drivers;

[0238] For each iteration process, determine the initial function value according to the number of various types of drivers before decision-making, environmental information, and the actual revenue data of the drivers;

[0239] Update the parameters of the initial function according to the initial function value and the optimization algorithm;

[0240] When the current iteration count meets the maximum iteration count, determine the initial function at the current iteration count as the objective function.

[0241] In one embodiment, when the processor executes the computer program, the following steps are also implemented:

[0242] Obtain the driver type weight coefficient, the environmental type weight coefficient, and the penalty term coefficient;

[0243] Multiply the number of drivers of each type before decision-making by the driver type weight coefficient to obtain the first function value;

[0244] Multiply the environmental information by the environmental type weight coefficient to obtain the second function value;

[0245] Conduct income analysis processing on the actual income data of the driver to obtain the income value, and multiply the income value by the penalty term coefficient to obtain the penalty value;

[0246] Determine the initial function value according to the first function value, the second function value, and the penalty value.

[0247] In one embodiment, when the processor executes the computer program, the following steps are also implemented:

[0248] Perform data preprocessing on the order supply information to obtain the target input data; the data preprocessing includes at least one of spatio-temporal partitioning, driver classification, driver status judgment, and driver order receiving statistics.

[0249] According to some embodiments of the present application, a computer program product is also provided. When the computer program is executed by a processor, the above method can be implemented. The computer program product includes one or more computer instructions. When these computer instructions are loaded and executed on a computer, part or all of the above method can be implemented in accordance with the process or function described in the embodiments of the present application.

[0250] According to some embodiments of the present application, a non-transitory computer-readable storage medium including instructions is also provided, such as a memory including instructions. The above instructions can be executed by a processor of an electronic device to complete the above method. For example, the non-transitory computer-readable storage medium can be a ROM, a random access memory (RAM), a CD-ROM, a magnetic tape, a floppy disk, and an optical data storage device, etc.

[0251] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use, and processing of relevant data need to comply with the relevant laws, regulations, and standards of relevant countries and regions.

[0252] Those of ordinary skill in the art can understand that all or part of the processes in the methods of the above embodiments can be completed by instructing relevant hardware through a computer program. The computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above methods. Among them, any reference to a memory, database, or other medium used in the embodiments provided in this application can include at least one of non-volatile and volatile memories. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetoresistive random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM), etc. The databases involved in the embodiments provided in this application can include at least one of relational databases and non-relational databases. Non-relational databases can include distributed databases based on blockchain, etc., and are not limited thereto. The processors involved in the embodiments provided in this application can be general-purpose processors, central processors, graphics processors, digital signal processors, programmable logic devices, data processing logics based on quantum computing, etc., and are not limited thereto.

[0253] The technical features of the above embodiments can be combined arbitrarily. For the sake of brevity of description, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, it should be considered as the scope described in this specification.

[0254] The above-described embodiments merely represent several implementation manners of the present application. The description thereof is relatively specific and detailed, but it should not be construed as a limitation on the patent scope of the present application. It should be noted that for those of ordinary skill in the art, without departing from the concept of the present application, several modifications and improvements can still be made, and these all fall within the protection scope of the present application. Therefore, the protection scope of the present application shall be subject to the appended claims.

Claims

1. A method for making preferential decisions for online car-hailing services, characterized in that, The method includes: Determining target input data based on pre-acquired order supply information; Inputting the target input data into a pre-constructed decision support model for decision-making processing to obtain a target decision result; the target decision result includes the opening quantity and ratio of preferential order listening corresponding to the maximization of platform revenue; Sending recommendation information to at least one target terminal according to the target decision result, where the recommendation information is used to recommend that the users of the target terminal turn on the preferential service.

2. The method according to claim 1, characterized in that, The target input data includes global spatio-temporal information, environmental variables, and the number of drivers before decision-making; the decision support model includes a one-to-one connection layer, an encoder layer, a decision layer, and a fully connected layer. Inputting the target input data into the pre-constructed decision support model for decision-making processing to obtain a target decision result includes: Inputting the environmental variables into the one-to-one connection layer for feature extraction processing to obtain feature data; the feature extraction includes at least one of geographical location feature extraction, time feature extraction, user feature extraction, and order-related feature extraction; Inputting the number of drivers before decision-making into the decision layer for binary classification decision-making processing to obtain decision information; Inputting the global spatio-temporal information into the encoder layer for global spatio-temporal supply and demand information extraction processing to obtain a global spatio-temporal supply and demand vector; Inputting the feature data, the global spatio-temporal supply and demand vector, and the decision information into the fully connected layer for feature integration and feature extraction processing to obtain the target decision result.

3. The method according to claim 2, wherein The decision information includes the ratios of various types of drivers to turn on and turn off preferential order listening. Inputting the number of drivers before decision-making into the decision layer for binary classification decision-making processing to obtain decision information includes: Performing a weighted summation operation on pre-acquired weight parameters and the number of drivers before decision-making to obtain an intermediate result; Inputting the intermediate result into a pre-constructed activation function for binary classification decision-making processing to obtain the ratios of various types of drivers to turn on and turn off preferential order listening.

4. The method according to claim 2, wherein The method further includes: Performing data preprocessing on pre-acquired historical operation data to obtain training data; the data preprocessing includes format processing, hexagonal grid coding conversion processing, and data partitioning processing; Inputting the training data into an initial prediction model for prediction processing to obtain an initial prediction result; Adjusting the parameters in the initial prediction model according to the initial prediction result and the mean absolute value function to determine the decision support model.

5. The method according to any one of claims 1-4, characterized in that, The method further includes: Calculating the maximum platform revenue value according to the target decision result and a pre-established target function; the target function includes the corresponding relationship between the total platform revenue and at least one of the order price, the platform commission ratio, and the operation cost.

6. The method according to claim 5, wherein The training process of the target function includes: Obtaining a plurality of samples, where the samples include the number of various types of drivers before decision-making, environmental information, and driver actual revenue data; For each iteration process, determining an initial function value according to the number of various types of drivers before decision-making, the environmental information, and the driver actual revenue data; Updating the parameters of the initial function according to the initial function value and an optimization algorithm; When the current iteration count meets the maximum iteration count, determine the initial function at the current iteration count as the target function.

7. The method according to claim 6, wherein The determining the initial function value according to the number of drivers of various types before decision-making, the environmental information, and the actual driver revenue data includes: Obtain the driver type weight coefficient, the environmental type weight coefficient, and the penalty term coefficient; Perform a product operation on the number of drivers of various types before decision-making and the driver type weight coefficient to obtain a first function value; Perform a product operation on the environmental information and the environmental type weight coefficient to obtain a second function value; Perform revenue analysis processing on the actual driver revenue data to obtain a revenue value, and perform a product operation on the revenue value and the penalty term coefficient to obtain a penalty value; Determine the initial function value according to the first function value, the second function value, and the penalty value.

8. The method according to claim 1, wherein The determining the target input data based on the pre-obtained order supply information includes: Perform data preprocessing on the order supply information to obtain the target input data; the data preprocessing includes at least one of space-time partitioning, driver classification, driver status judgment, and driver order acceptance statistics.

9. A preferential decision-making device for online car-hailing, characterized in that, The apparatus includes: A data determination module, configured to determine target input data based on pre-obtained order supply information; A decision module, configured to input the target input data into a pre-constructed decision support model for decision processing to obtain a target decision result; the target decision result includes the number and proportion of preferential order acceptance openings corresponding to maximizing platform revenue; An information sending module, configured to send recommendation information to at least one target terminal according to the target decision result, where the recommendation information is used to recommend that the user of the target terminal enable preferential services.

10. A computer device, comprising a memory and a processor, the memory storing a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 8.

Images