Optimization method and system for establishing marketing activity strategy based on vehicle dispatching data

Abstract

The invention discloses an optimization method and system for establishing a marketing activity strategy based on vehicle dispatching data. The invention relates to the technical field of online car hailing. The method comprises the following steps: collecting order completion data a, driving data b and duration data c of a driver from a database of an online car-hailing platform to form a data set D; and generating feature data D'according to the data set D, wherein the feature data D 'comprises statistical information of the historical order completion amount a'and the driving duration c' of the driver. Configuring basic information of the main activity, wherein the basic information comprises requirements of a time period d, a city e and a region f; and after all the tenants finish auditing and the activity starting time is up, starting a timed task to monitor the activity state. And generating a daily sub-activity strategy according to the rule of the main activity and the feature data generated upstream. According to the invention, through real-time data analysis and intelligent strategy generation, dynamic management and optimization of online car-hailing driver activities can be realized, so that the overall operation efficiency of an online car-hailing platform is improved. And the platform can better deal with the market change and adjust the operation strategy in time.

Description

Technical Field

[0001] The present invention relates to the technical field of online car-hailing, and specifically to strategy optimization of online car-hailing marketing activities, and in particular to an optimization method and system for building a marketing activity strategy based on vehicle dispatch data. Background Art

[0002] In the current online car-hailing industry, core information such as the driver's activity time, driving time, and order-running mode show certain limitations to the existing activity scheduling mechanism, which is mainly reflected in the lack of activity flexibility. Most activity arrangements need to be issued in advance, resulting in a lag in strategy adjustment. This situation highlights the current system's lack of intelligent activity issuance based on the driver's actual activity time, which in turn affects the timeliness of strategy adjustment and the ability to fully utilize operational data.

[0003] In order to meet the above challenges, the current technical solutions urgently need to be innovated and optimized. The core idea is to deeply optimize the strategy based on the existing activity delivery link and introduce a new strategy for dynamically delivering reward activities based on activity periods. The goal is to dynamically identify the most suitable time window for delivering reward activities by analyzing the driver's activity period data in real time, thereby achieving instant incentives and guidance for driver behavior.

[0004] To this end, the present invention proposes an optimization method and system for building a marketing activity strategy based on vehicle dispatch data. Summary of the invention

[0005] In view of this, the present invention hopes to provide an optimization method and system for building marketing activity strategies based on vehicle dispatch data, so as to solve or alleviate the technical problems existing in the prior art, namely, how to respond to market changes in a timely manner based on real-time data analysis capabilities and realize dynamic adjustment of marketing strategies (for example, during the evening peak of online car-hailing, activities are dynamically issued according to the strategy to stimulate drivers to dispatch vehicles and complete orders during the peak period, improve the response rate during the peak period, and achieve the effect of increasing the frequency during the period). The technical solution of the present invention is implemented as follows:

[0006] First, the optimization method of building marketing activity strategy based on vehicle data:

[0007] 1. Overview:

[0008] The present invention aims to achieve dynamic management and optimization of online car-hailing driver activities through real-time data analysis and intelligent strategy generation. It first collects and pre-processes the core data of drivers such as order completion, vehicle dispatch and duration, and then generates feature data and stores it in an efficient database. Based on this data, the solution can configure and review the main activities, and use machine learning models to predict driver behavior to formulate more accurate activity strategies. Subsequently, according to the main activity rules and feature data, daily sub-activity strategies are automatically generated, and budget allocation and leveling are performed to ensure the economy and feasibility of the activities. The strategies are issued to the corresponding drivers so that they can dispatch vehicles and complete orders according to the strategies, thereby improving the timeliness of the activities and the participation of drivers.

[0009] (II) Technical solution:

[0010] 2.1 Step S1, data collection and preprocessing:

[0011] The driver’s order completion data a, vehicle dispatch data b and duration data c are collected from the database of the online car-hailing platform to form a data set D.

[0012] 2.2 Step S2, feature generation and storage:

[0013] Generate feature data D' based on data set D, including the driver's historical completed orders a' and driving time c' statistics. Including:

[0014] Given a data set D = (a i ,b i ,c i )|i=1,...,n}, where a i represents the order completion data of the i-th driver, b i represents the dispatch data of the i-th driver, c i represents the duration data of the i-th driver. The goal of step S2 is to generate a feature data set D'={(a' i ,c' i )|i=1,...,n}, where a' i and c' i Respectively represent the statistics of the historical number of completed orders and driving time of the i-th driver:

[0015] (1) Historical order volumea i Calculation of ′: Use the sliding window average method to calculate the average number of completed orders of the i-th driver in the past k time units (such as days, hours, etc.):

[0016]

[0017] Among them, a j It represents the number of orders completed by the driver in the jth time unit, and k is the size of the sliding window.

[0018] (2) Distance travel time c i Calculation of ′: The sliding window average method is also used to calculate the average driving time of the i-th driver in the past l time units:

[0019]

[0020] c j It represents the driver’s driving time in the jth time unit, and l is the size of the sliding window.

[0021] Then write it to the ClickHouse database and use the MergeTree engine to aggregate and query the data.

[0022] 2.3 Step S3, main activity configuration and review:

[0023] Configure the basic information of the main activity, including time period d, city e and region f requirements; based on the feature data D', use the treelite model to predict the driver's next-day order volume A and driving time C, then review the main activity based on the platform and tenants to determine whether the activity is approved, and generate additional reward information according to the main activity rules;

[0024] 2.3.1 Step S300, perform prediction task:

[0025] The treelite model is an XGBoost model:

[0026]

[0027] in is the predicted number of orders completed the next day. is the predicted trip time. θ represents the parameters of the model, including the structure of all trees and the weights of leaf nodes. K is the number of decision trees, q k is the structure data, x i is the i-th sample (a' i ,c' i ); XGBoostModel is the XGBoost model, which has the following form:

[0028] 1) The XGBoost model consists of K decision trees, each of which is represented by f k (x); for each decision tree f k (x), which all include structural data q k and leaf node weight f(w) k .

[0029] 2) First, the structure data q k Map the input space to leaf nodes:

[0030] q k =R m →{1,2,...,T k};

[0031] Among them, T k is the number of leaf nodes of the kth decision tree. → represents a mapping operation. R represents a set of real numbers, and m is a positive integer representing the dimension of the vector. Therefore, R m The space is an m-dimensional real vector space, in which each element is an m-dimensional real vector.

[0032] 3) Specify the leaf node weight w k ∈R Tk is the score associated with each leaf node of the kth decision tree, so let the prediction output of the XGBoost model be the sum of the predictions of all trees:

[0033]

[0034] in, is a two-dimensional vector,

[0035] 2.3.2 Step S301, review and generate additional reward information:

[0036] The first review standard: predicted next day's order volume A i Must be greater than or equal to N orders;

[0037] Second review standard: predicted dispatch time C i Must be greater than or equal to M hours;

[0038] If both audit criteria 1 and audit criteria 2 are met, the activity passes; otherwise, the activity fails.

[0039] The first additional reward condition: Drivers who pass the activity, if the predicted number of orders completed the next day is A i If the number of orders is greater than or equal to N, an additional reward of n yuan will be given.

[0040] Second additional reward condition 2: Drivers who pass the activity, if the predicted driving time is C i If it is greater than or equal to M hours, an additional reward of m yuan will be given.

[0041] For drivers who meet the additional reward conditions, reward information including the reward amount is generated and notified to the driver.

[0042] 2.4 Step S4, strategy generation and budget allocation:

[0043] When all tenants have completed the review and the activity starts at the time, start the scheduled task to monitor the activity status. Generate daily sub-activity strategies based on the rules of the main activity and the feature data generated upstream.

[0044] Based on the leveling strategy algorithm, the generated sub-activity strategy s is budgeted and leveled.

[0045] 2.4.1 Step S400, generating daily sub-activity strategies:

[0046] S4000, initialize the empty set S t Used to store sub-activity strategies;

[0047] S4001, for each rule r in the main active rule set R, execute the following in a loop:

[0048] According to the rule r and the characteristic data F of the day t , calculate whether the conditions for satisfying the rule are established;

[0049] If the condition is met, generate the corresponding sub-activity strategy s according to rule r and add it to the set S t middle;

[0050] S4002, return set S t As a collection of sub-activity strategies for the day.

[0051] 2.4.2 Step S401, execute the leveling strategy algorithm:

[0052] The input of this step is: sub-activity strategy set S t (where t represents the current date), and the available budget B for the day t ;

[0053] The output of this step is: the set of sub-activity strategies S after budget allocation t ′;

[0054] The execution process of the leveling strategy algorithm in step S401 is:

[0055] S4010, Initialize budget balance B remaining =B t .

[0056] S4011, initialize the empty set S t ′ is used to store the sub-activity strategy after budget allocation;

[0057] S4012, for the set S t For each sub-activity strategy s in , execute it in a loop:

[0058] Calculate the budget requirement b for strategy s s ;

[0059] If b s ≤B remaining , then allocate the budget to strategy s and add it to the set S t '; Update budget balance B remaining =B remaining -b s ; if B remaining >0, then for the set S t ′ to further level the budget to ensure full utilization of the budget:

[0060] a. Calculate the budget adjustment coefficient k for each strategy s s , which is determined based on the strategy’s priorities and effect predictors;

[0061] b. For set S t For each strategy s in ″, execute it cyclically:

[0062] I. Calculate the budget increase Δb for strategy s s :

[0063]

[0064] Where K is the sum of all strategy budget adjustment coefficients.

[0065] II. Update the budget of strategy s to b s +Δb s ;

[0066] III. Calculate whether the updated budget exceeds the maximum budget limit of the strategy. If not, proceed to the next step; if yes, reduce the adjustment coefficient k in a stepwise or iterative manner. s ;

[0067] c. Return the sub-activity strategy set S after budget adjustment t ″.

[0068] S4013, return set S t ′ is the set of sub-activity strategies after budget allocation.

[0069] 2.5 Step S5, policy storage and distribution:

[0070] When a sub-activity is started, the sub-activity strategy s is sent to the corresponding driver. The driver can view and participate in the activity on the client, and then the driver can dispatch the vehicle and complete the order according to the sent strategy.

[0071] (III) Mechanism of solving technical problems:

[0072] By collecting and preprocessing the core data of online ride-hailing drivers' orders, trips and duration in real time, feature data is generated and stored in an efficient database. This data can be used to configure and review main activities. At the same time, a machine learning model (treelite) is used to predict the driver's next-day order volume and trip duration, so as to more accurately grasp market changes and driver behavior trends.

[0073] According to the rules of the main activity and the feature data generated upstream, it can automatically generate daily sub-activity strategies, allocate and balance budgets, and ensure the economic and feasibility of the activities. In this process, the adjustment and optimization of strategies can respond to market changes in real time. For example, if the response rate of online car-hailing services decreases during the evening peak hours, the plan can quickly generate and issue corresponding activity strategies to stimulate drivers to complete orders during peak hours.

[0074] The strategies are sent to the corresponding drivers, who can receive and display them in real time, so that they can dispatch vehicles and complete orders according to the latest strategies. This not only improves the timeliness of the activities, but also greatly increases the participation of drivers, thus achieving dynamic adjustment and optimization of marketing strategies.

[0075] Secondly, we will build an optimization system for marketing activity strategies based on vehicle dispatch data:

[0076] like Figure 4 As shown, the system includes a processor and a memory connected to the processor, wherein the memory stores program instructions, and when the program instructions are executed by the processor, the processor executes the optimization method as described above. The processor is connected to,

[0077] (1) A data collection module responsible for collecting core operational information from the online ride-hailing platform’s database in real time: including drivers’ order completion data, vehicle dispatch data, and duration data. The key elements are:

[0078] (1.1) Data source: online car-hailing platform database;

[0079] (1.2) Data types: order completion data, vehicle dispatch data and duration data;

[0080] (1.3) Preprocessing operations: data cleaning, format unification and missing value processing;

[0081] (2) Feature generation and storage module for generating feature data: including the driver’s historical order completion volume and driving time statistics. The key elements are:

[0082] (2.1) Feature type: historical statistics of completed orders and vehicle dispatch time;

[0083] (2.2) Storage method: ClickHouse database, MergeTree engine;

[0084] (3) The main activity configuration and review module is responsible for configuring the basic information of the main activity: including time period, city, region and other requirements. At the same time, the machine learning model (treelite) is used to predict the number of orders completed by the driver the next day and the length of time the driver will be on duty, providing data support for strategy generation. The configured main activity needs to be reviewed by the platform and tenants to determine whether the activity is approved, and to add reward information according to the main activity rules. The key elements are:

[0085] (3.1) Activity configuration: time period, city and region;

[0086] (3.2) Prediction model: treelite machine learning model;

[0087] (3.3) Audit process: platform audit and tenant audit;

[0088] (4) Start the strategy generation and budget allocation module for scheduled tasks to monitor the activity status: When all tenants have completed the review and the activity start time has arrived, this module will start the scheduled task to monitor the activity status. According to the rules of the main activity and the feature data generated upstream, the daily sub-activity strategy is automatically generated. At the same time, the budget is allocated and leveled for the generated strategy based on the leveling strategy to ensure the economy and feasibility of the activity. The key elements are:

[0089] (4.1) Strategy generation: based on main activity rules and feature data;

[0090] (4.2) Budget allocation: leveling strategy;

[0091] (4.3) Scheduled tasks: monitor activity status;

[0092] (5) The strategy storage and distribution module stores the generated sub-activity strategy in the strategy result table: When the sub-activity is started, this module is responsible for distributing the strategy to the corresponding driver, who can view and participate in the activity on the client. The driver dispatches the vehicle and completes the order according to the distributed strategy, thereby realizing the dynamic adjustment of the marketing strategy. The key elements are:

[0093] (5.1) Policy storage: policy result table;

[0094] (5.2) Policy issuance: Displayed on the driver side;

[0095] (5.3) Driver operation: dispatching vehicles and / or completing orders.

[0096] Compared with the prior art, the present invention has the following beneficial effects:

[0097] 1. Improve operational efficiency: Through real-time data analysis and intelligent strategy generation, the present invention can achieve dynamic management and optimization of online ride-hailing driver activities, thereby improving the overall operational efficiency of the online ride-hailing platform. It helps the platform to better respond to market changes and adjust operational strategies in a timely manner to meet the travel needs of passengers.

[0098] 2. Improve service quality: The present invention can predict driver behavior based on real-time data and formulate more accurate activity strategies accordingly. It helps to stimulate drivers to dispatch vehicles to complete orders during peak hours, improve response rates, and thus improve passengers' travel experience and service quality. At the same time, by dynamically adjusting marketing strategies, the platform can better meet drivers' needs and improve driver satisfaction and loyalty.

[0099] 3. Enhance market competitiveness: The solution of the present invention uses data-driven and intelligent means to enable online car-hailing platforms to respond to market changes more flexibly and adjust marketing strategies in a timely manner. It helps the platform stand out in the fiercely competitive market environment, attract more passengers and drivers, and thus increase market share and brand influence.

[0100] 4. Optimize resource allocation: Through real-time data analysis and budget allocation and leveling mechanism, the present invention can ensure the economy and feasibility of activities. This helps the platform to allocate resources more reasonably, avoid resource waste and over-investment, and thus achieve sustainable development. BRIEF DESCRIPTION OF THE DRAWINGS

[0101] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments or technical descriptions will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0102] Figure 1 It is a schematic diagram of the method flow of the present invention;

[0103] Figure 2 It is a schematic diagram of the service execution process of the present invention;

[0104] Figure 3 It is a schematic diagram of the execution flow of the leveling strategy algorithm of S401 of the present invention;

[0105] Figure 4 It is a system schematic diagram of the present invention. DETAILED DESCRIPTION

[0106] In order to make the above-mentioned purposes, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are described in detail below in conjunction with the accompanying drawings. In the following description, many specific details are set forth to facilitate a full understanding of the present invention. However, the present invention can be implemented in many other ways different from those described herein, and those skilled in the art can make similar improvements without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below;

[0107] It should be noted that the various embodiments in this specification are described in a progressive manner, and each embodiment focuses on the differences from other embodiments, and the same or similar parts between the various embodiments can be referred to each other. For the device disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and the relevant parts can be referred to the method part description.

[0108] Explanation of relevant terms:

[0109] (1) Completed order data: refers to the number of orders completed by online ride-hailing drivers or related information. It is an important indicator for evaluating driver workload and platform operating efficiency.

[0110] (2) Dispatch data: This records the time, location, status, and other information of online ride-hailing drivers, which is used to analyze the drivers’ dispatch habits and the platform’s capacity distribution.

[0111] (3) Duration data: Data reflecting the online time or driving time of online ride-hailing drivers, which helps to understand the drivers’ workload and the stability of the platform’s transportation capacity.

[0112] (4) ClickHouse database: a high-performance column-based database suitable for real-time data analysis and large-scale data query, providing fast data reading and writing speeds.

[0113] (5) MergeTree engine realizes data aggregation and query: a storage engine in ClickHouse database that can efficiently aggregate and query data, support real-time data analysis and rapid response.

[0114] (6) Main activities and their basic information (time period, cities and regions): the main marketing activities launched by the online ride-hailing platform, including basic information such as the time period of the activity, the cities covered and the specific regions.

[0115] (7) (Main Event) Rules: Specific terms and conditions that govern how the Main Event is conducted, how drivers participate, and how rewards are distributed.

[0116] (8) Reward information: Details of rewards provided by the online ride-hailing platform to encourage drivers to participate in the activity, including reward type, amount, and issuance conditions.

[0117] (9) Upstream: In the process of data processing and strategy generation, it refers to the link that provides input data or information, such as data collection module or feature generation module.

[0118] (10) Sub-activity strategy: A detailed and preset marketing strategy for a specific time period, city or region, generated based on the main activity rules and feature data.

[0119] (11) Leveling strategy: A strategy used to balance the order volume and rewards between different regions, time periods, or drivers to ensure the fairness and economy of the activity.

[0120] Embodiment 1: Figure 1 As shown, this embodiment will provide an optimization method for building a marketing activity strategy based on vehicle dispatch data, including the following execution steps.

[0121] In this embodiment, regarding step S1, data collection and preprocessing: collect the driver's order completion data (number of orders completed per day), vehicle dispatch data (number of vehicle dispatches per day) and duration data (daily online duration) from the database of the online car-hailing platform. Form data set D:

[0122] D={(a i ,b i ,c i )|i=1,...,n},a i represents the order completion data of the i-th driver, b i represents the dispatch data of the i-th driver, c i Represents the duration data of the i-th driver.

[0123] In this embodiment, regarding step S2, feature generation and storage: feature data D' is generated according to the data set D, including statistical information of the driver's historical order completion volume a' and vehicle dispatch time c'.

[0124] Specifically, the sliding window average method is used to calculate the average number of orders completed by each driver in the past 7 days. For example, for driver i, his historical number of completed orders a i ' is the average number of completed orders in the past 7 days. At the same time, the sliding window average method is also used to calculate the average driving time of each driver in the past 7 days. For example, for driver i, his driving time ci' is the average driving time in the past 7 days. Including:

[0125] Given a data set D = (a i ,b i ,c i )|i=1,...,n}, where a i represents the order completion data of the i-th driver, b i represents the dispatch data of the i-th driver, c irepresents the duration data of the i-th driver. The goal of step S2 is to generate a feature data set D'={(a' i ,c' i )|i=1,...,n}, where a' i and c' i Respectively represent the statistics of the historical number of completed orders and driving time of the i-th driver:

[0126] (1) Historical order volumea i Calculation of ′: Use the sliding window average method to calculate the average number of completed orders of the i-th driver in the past k time units (such as days, hours, etc.):

[0127]

[0128] Among them, a j It represents the number of orders completed by the driver in the jth time unit, and k is the size of the sliding window.

[0129] (2) Distance travel time c i Calculation of ′: The sliding window average method is also used to calculate the average driving time of the i-th driver in the past l time units:

[0130]

[0131] c j It represents the driver’s driving time in the jth time unit, and l is the size of the sliding window.

[0132] Then write it to the ClickHouse database and use the MergeTree engine to aggregate and query the data.

[0133] In this embodiment, regarding step S3, main activity configuration and review: configure basic information of the main activity, including time period d, city e and region f requirements;

[0134] Exemplarily, the basic information of the main activity is set, such as the activity period is 10:00-17:00 on weekdays, the city is XX city, and the area is XX district.

[0135] Then, based on the feature data D', the treelite model is used to predict the number of orders A and the driving time C of the driver on the next day. Then, the main activity is reviewed based on the platform and tenants to determine whether the activity is passed, and additional reward information is generated according to the main activity rules;

[0136] Specifically, in step S300, the prediction task is performed:

[0137] 1) The XGBoost model consists of K decision trees, each of which is represented by f k (x); for each decision tree fk (x), which all include structural data q k and leaf node weight f(w) k .

[0138] 2) First, the structure data q k Map the input space to leaf nodes:

[0139] q k =R m →{1,2,...,T k};

[0140] Among them, T k is the number of leaf nodes of the kth decision tree. → represents a mapping operation. R represents a set of real numbers, and m is a positive integer representing the dimension of the vector. Therefore, R m The space is an m-dimensional real vector space, in which each element is an m-dimensional real vector.

[0141] 3) Specify leaf node weights is the score associated with each leaf node of the kth decision tree, so let the prediction output of the XGBoost model be the sum of the predictions of all trees:

[0142]

[0143] in, is a two-dimensional vector,

[0144] Further, step S301, review and generate additional reward information: set the review criteria as the predicted number of orders completed the next day is greater than or equal to 10 orders, and the predicted driving time is greater than or equal to 8 hours. If the driver meets the review criteria, the activity is passed and additional reward information is generated.

[0145] For example, drivers who meet the requirements can receive an additional 50 yuan reward.

[0146] In this embodiment, regarding step S4, strategy generation and budget allocation: when all tenants have completed the review and the activity start time has arrived, start the scheduled task to monitor the activity status. Generate a daily sub-activity strategy s based on the rules of the main activity and the feature data generated upstream.

[0147] For example, for a driver in area xx, if his historical order completion volume is high and his driving time is stable, an exclusive sub-activity strategy can be generated for him.

[0148] Then, based on the leveling strategy algorithm, the generated sub-activity strategy s is budget allocated and leveled.

[0149] For example, if the budget for the day is limited, priority will be given to drivers with better historical performance to ensure full utilization of the budget.

[0150] Specifically, step S400 generates daily sub-activity strategies:

[0151] S4000, initialize the empty set S t Used to store sub-activity strategies;

[0152] S4001, for each rule r in the main active rule set R, execute the following in a loop:

[0153] According to the rule r and the characteristic data F of the day t , calculate whether the conditions for satisfying the rule are established;

[0154] If the condition is met, generate the corresponding sub-activity strategy s according to rule r and add it to the set S t middle;

[0155] S4002, return set S t As a collection of sub-activity strategies for the day.

[0156] Specifically, in step S401, the leveling strategy algorithm is executed:

[0157] The input of this step is: sub-activity strategy set S t (where t represents the current date), and the available budget B for the day t ;

[0158] The output of this step is: the set of sub-activity strategies S after budget allocation t ′;

[0159] The execution process of the leveling strategy algorithm in step S401 is:

[0160] S4010, Initialize budget balance B remaining =B t .

[0161] S4011, initialize the empty set S t ′ is used to store the sub-activity strategy after budget allocation;

[0162] S4012, for the set S t For each sub-activity strategy s in , execute it in a loop:

[0163] Calculate the budget requirement b for strategy s s ;

[0164] If b s ≤B remaining , then allocate the budget to strategy s and add it to the set S t'; Update budget balance B remaining =B remaining -b s ; if B remaining >0, then for the set S t ′ to further level the budget to ensure full utilization of the budget:

[0165] a. Calculate the budget adjustment coefficient k for each strategy s s , which is determined based on the strategy’s priorities and effect predictors;

[0166] b. For set S t For each strategy s in ″, execute it cyclically:

[0167] I. Calculate the budget increase Δb for strategy s s :

[0168]

[0169] Where K is the sum of all strategy budget adjustment coefficients.

[0170] II. Update the budget of strategy s to b s +Δb s ;

[0171] III. Calculate whether the updated budget exceeds the maximum budget limit of the strategy. If not, proceed to the next step; if yes, reduce the adjustment coefficient k in a stepwise or iterative manner. s ;

[0172] c. Return the sub-activity strategy set S after budget adjustment t ″.

[0173] S4013, return set S t ′ is the set of sub-activity strategies after budget allocation.

[0174] In this embodiment, regarding step S5, policy storage and issuance: the generated sub-activity policy is stored in the database for subsequent query and use. When the sub-activity is started, the sub-activity policy is issued to the corresponding driver. The driver can view and participate in the activity on the terminal, and then dispatch and complete the order according to the issued policy. For example, the driver can see an exclusive activity page that displays the detailed information of the activity, the reward amount, and the conditions for participation.

[0175] In this embodiment, the Python execution framework program of all the above technical solutions is as follows:

[0176]

[0177]

[0178]

[0179] remaining_budget=budget

[0180] for reward in rewards:

[0181] if reward>0and remaining_budget>=reward:

[0182] strategies.append(('Offer Reward',reward))

[0183] remaining_budget - = reward

[0184] else:

[0185] strategies.append(('No Reward',0))

[0186] return strategies

[0187] strategies=generate_strategies(rewards)

[0188] # Output results

[0189] for strategy in strategies:

[0190] print(strategy)

[0191] In the above program, the Pandas library is used to create a sample data set, including driver ID, number of completed orders, number of vehicle dispatches, and online time. The rolling window average method is used to generate statistical features of historical completed orders and vehicle dispatch time. The XGBoost regression model is used to train the feature data. The test set is predicted to obtain the predicted values ​​of the next day's completed orders and vehicle dispatch time.

[0192] Based on the predicted number of completed orders and the duration of the trip, as well as the set threshold, the driver is reviewed to see if he meets the activity conditions. Qualified drivers will receive rewards. Based on the reward amount and the set budget, a sub-activity strategy is generated to ensure that the budget is fully utilized and within the budget limit.

[0193] Finally, print out each driver's strategy, including whether he or she received a reward and the amount of the reward.

[0194] Embodiment 2: Based on Embodiment 1, this embodiment further provides an adjustment coefficient k s The intelligent assignment strategy is the particle swarm optimization (PSO):

[0195] In this embodiment, regarding S40120, the objective function is defined as:

[0196] f(x)=x 2 +10sin(x);

[0197] Among them, x is the target value, that is, the adjustment coefficient k s value.

[0198] It can be understood that the objective function is a nonlinear function, and its graph shows multiple local extreme points, so it is a complex function suitable for optimization using the particle swarm algorithm. The goal is to find the global minimum of this function in a given interval ([-10,10]) to dynamically minimize the priority of the adaptation strategy.

[0199] It should be pointed out that in a series of strategies given by the algorithm, by adjusting the adjustment coefficient k s The value of the objective function f(x) is made to reach the global minimum, thereby determining the optimal fitness or priority of each strategy. The "dynamic" here refers to the process of adjusting the strategy in real time according to the changes in the objective function. For each strategy s, the adjustment characteristics and their effects are:

[0200] (1) Adjustment characteristics:

[0201] Each strategy s has an associated adjustment factor k s value, which can be optimized by the particle swarm algorithm. During the optimization process, the particle swarm algorithm will continuously try different ks values ​​to find the adjustment coefficient k that can make the objective function f(x) reach the global minimum. s Once the optimal adjustment coefficient k is found s The fitness or priority of strategy s is determined as the value corresponding to the minimum value of the objective function.

[0202] (2) Effect:

[0203] By dynamically adjusting the adjustment factor k s Each strategy s can be assigned a clear fitness or priority according to its corresponding objective function value. This makes it more direct and efficient to select the best strategy among multiple strategies, because only their objective function values ​​need to be compared. In practical applications, this dynamic adjustment mechanism can help the system or algorithm better adapt to changing environments or conditions, thereby improving overall performance and effectiveness.

[0204] In this embodiment, regarding S40121, the particle swarm is initialized:

[0205] A group of particles are randomly generated in the interval [-10, 10], each particle has a position (representing the value of x) and a velocity. Then the objective function value at each particle position, i.e. f(x), is calculated.

[0206] In this embodiment, regarding S40122, the individual and global optimal solutions are updated:

[0207] For each particle, compare its current objective function value with the individual historical best objective function value, and update the individual best position (pbest). Among all particles, find the particle with the best objective function value and update the global best position (gbest).

[0208] In this embodiment, regarding S40123, the particle speed and position are updated:

[0209] Use the PSO velocity update formula to adjust the velocity of each particle. Use the updated velocity to adjust the particle's position to ensure that the position is still within the interval [-10,10]. Where:

[0210] (1) Speed ​​update formula:

[0211] v(t+1)=w*v(t)+c1*r1*(pbest-x(t))+c2*r2*(gbest-x(t));

[0212] Among them, v(t) and x(t) are the velocity and position of the particle at time t, w is the inertia weight, c1 and c2 are learning factors, and r1 and r2 are random numbers.

[0213] (2) Position update formula:

[0214] x(t+1)=x(t)+v(t+1);

[0215] It should be pointed out that in the algorithm, each particle represents a potential solution to the problem, and updates its own speed and position by tracking the individual optimal solution (pbest) and the global optimal solution (gbest). The inertia weight w balances the influence of the particle's historical speed and current speed, and the learning factors c1 and c2 control the influence of individual experience and group experience on the particle's speed, respectively. The introduction of random numbers r1 and r2 increases the randomness of the algorithm and helps avoid falling into a local optimal solution. Through continuous iteration, the particle swarm gradually approaches the optimal solution to the problem.

[0216] In this embodiment, regarding S40124, iterate until the termination condition is met:

[0217] Repeat steps S40122 to S40123 until a preset maximum number of iterations is reached or the objective function value of the global optimal solution meets a specific condition (for example, the change is less than a preset threshold).

[0218] In this embodiment, regarding S40125, the output result is:

[0219] Output the global optimal position (i.e. the value of x) and its corresponding objective function value, and obtain the adjustment coefficient k s value.

[0220] All the above embodiments only express the implementation methods of the relevant practical applications of the present invention, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the invention patent. It should be pointed out that, for ordinary technicians in this field, several variations and improvements can be made without departing from the concept of the present invention, which all belong to the protection scope of the present invention. Therefore, the protection scope of the patent of the present invention shall be based on the attached claims.

[0221] For those skilled in the art, it can be further appreciated that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of the two. In order to clearly illustrate the interchangeability of hardware and software, the composition and steps of each example have been generally described in the above description according to function. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of the present invention.

[0222] At the same time, those skilled in the art can understand that all or part of the processes in all the above-mentioned embodiments can be completed by instructing the relevant hardware through a computer program, and 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-mentioned methods. Among them, any reference to memory, storage, database or other media provided in this application and used in the embodiments may include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. As an illustration and not limitation, RAM is available in many forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (SSRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM).

Claims

1. An optimization method for building a marketing activity strategy based on vehicle dispatch data, characterized in that: The steps include: S1, collect the driver's order completion data a, vehicle dispatch data b and duration data c to form a data set D; S2, generating feature data D' based on data set D, including the driver's historical completed orders a' and driving time c' statistics; S3, configure the basic information of the main activity, including time period d, city e and region f requirements; based on the feature data D', use the treelite model to predict the driver's next day's order volume A and driving time C; generate additional reward information according to the main activity rules; S4: When all tenants have completed the review and the activity start time has arrived, a daily sub-activity strategy s is generated according to the rules of the main activity; based on the leveling strategy algorithm, budget allocation and leveling are performed on the generated sub-activity strategy s; S5, when the sub-activity is started, the sub-activity strategy s is sent to the corresponding driver.

2. The optimization method according to claim 1, characterized in that: In S2, the statistical information of the historical order completion amount a' and the vehicle dispatching time c' is calculated as follows: Among them, a j represents the number of orders completed by the driver in the jth time unit, and k is the size of the sliding window; c j It represents the driver’s driving time in the jth time unit, and l is the size of the sliding window.

3. The optimization method according to claim 1, characterized in that: In the S3, the treelite model is an XGBoost model: in is the predicted number of orders completed the next day. is the predicted trip time; θ represents the parameters of the model, including the structure of all trees and the weight of leaf nodes; K is the number of decision trees, q k is the structure data, x i is the i-th sample in the dataset D'; XGBoostModel is the XGBoost model.

4. The optimization method according to claim 3, characterized in that: The architecture of the XGBoost model XGBoostModel is as follows: 1) The XGBoost model consists of K decision trees, each of which is represented by f k (x); for each decision tree f k (x), which all include structural data q k and leaf node weight f(w) k ; 2) First, structure the data q k Map the input space to leaf nodes: q k =R m →{1,2,...,T k }; Among them, T k is the number of leaf nodes of the kth decision tree; → represents a mapping operation; R represents a set of real numbers, and m is a positive integer; 3) Then specify the leaf node weight is the score associated with each leaf node of the kth decision tree, and the prediction output of the XGBoost model is the sum of the predictions of all trees: in, is a two-dimensional vector, 5. The optimization method according to claim 3, characterized in that: In the S3, the method for generating additional reward information is as follows: The first review standard: predicted next day's order volume A i Must be greater than or equal to N orders; Second review standard: predicted dispatch time C i Must be greater than or equal to M hours; If both audit criteria 1 and audit criteria 2 are met, the activity passes; otherwise, the activity fails; The first additional reward condition: Drivers who pass the activity, if the predicted number of orders completed the next day is A i If the number of orders is greater than or equal to N, an additional reward of n yuan will be given; Second additional reward condition 2: Drivers who pass the activity, if the predicted driving time is C i If it is greater than or equal to M hours, an additional reward of m yuan will be given; For drivers who meet the additional reward conditions, reward information including the reward amount is generated and notified to the driver.

6. The optimization method according to any one of claims 1 to 3, characterized in that: In S4, generating a daily sub-activity strategy includes: S4000, initialize the empty set S t Used to store sub-activity strategies; S4001, for each rule r in the main active rule set R, execute the following in a loop: According to the rule r and the characteristic data F of the day t , calculate whether the conditions for satisfying the rule are established; If the condition is met, generate the corresponding sub-activity strategy s according to rule r and add it to the set S t middle; S4002, return set S t As a collection of sub-activity strategies for the day.

7. The optimization method according to any one of claims 1 to 3, characterized in that: In S4, the execution strategy of the leveling strategy algorithm is as follows: S4010, Initialize budget balance B remaining =B t ; S4011, initialize the empty set S t ′ is used to store the sub-activity strategy after budget allocation; S4012, for the set S t For each sub-activity strategy s in , execute it in a loop: Calculate the budget requirement b for strategy s s ; If b s ≤B remaining , then allocate the budget to strategy s and add it to the set S t '; Update budget balance B remaining =B remaining -b s ; if B remaining >0, then for the set S t ′ to further level the budget; S4013, return set S t ′ is the set of sub-activity strategies after budget allocation.

8. The optimization method according to claim 7, characterized in that: In S4012, the method for further budget leveling is as follows: a. Calculate the budget adjustment coefficient k for each strategy s s ; b. For set S t For each strategy s in ″, execute it cyclically: I. Calculate the budget increase Δb for strategy s s : Where K is the sum of all strategy budget adjustment coefficients; II. Update the budget of strategy s to b s +Δb s ; III. Calculate whether the updated budget exceeds the maximum budget limit of the strategy. If not, proceed to the next step; if yes, reduce the adjustment coefficient k in a stepwise or iterative manner. s ; c. Return the sub-activity strategy set S after budget adjustment t ″.

9. An optimization system for building marketing activity strategies based on vehicle dispatch data, characterized by: The system includes a processor and a memory connected to the processor, wherein program instructions are stored in the memory, and when the program instructions are executed by the processor, the processor executes the optimization method according to any one of claims 1 to 8.

10. The optimization system according to claim 9, characterized in that: The processor is connected to a data collection module responsible for collecting core operational information from a database of the online ride-hailing platform in real time; A feature generation and storage module for generating feature data; The main activity configuration and review module is responsible for configuring the basic information of the main activity; Start the strategy generation and budget allocation module for scheduled tasks to monitor the activity status; The policy storage and delivery module stores the generated sub-activity policies in the policy result table.

Images