User object number prediction method, computer equipment, storage medium and computer program product

Abstract

The invention relates to a user object number prediction method, computer equipment, a storage medium and a computer program product, and relates to the technical field of logistics information. The method comprises the following steps: acquiring an actual object number corresponding to a to-be-predicted user in a preset time period and to-be-predicted time of the to-be-predicted user; processing the actual object number and the to-be-predicted time based on the effect parameter and the hyper-parameter of the trained number prediction model to obtain a target prediction result; the effect parameter is obtained by updating an initial parameter of the number prediction model based on test data and evaluation indexes of each user; the initial parameters are obtained through analysis based on the actual object number corresponding to each user in the preset time period. By adopting the method, the accuracy of predicting the user demand quantity can be improved.

Description

Technical Field

[0001] This application relates to the field of logistics information processing technology, and in particular to a method for predicting the number of user objects, a computer device, a storage medium, and a computer program product. Background Technology

[0002] In the process of operating various types of users in the express delivery industry, there is a group of large-scale, high-demand top users whose daily demand is difficult to predict due to their own marketing activities and business conditions.

[0003] Currently, the prediction of daily demand from top users relies solely on manual forecasting, with no other data input or verification to validate the biases of these forecasts. Furthermore, the daily demand figures provided by top users are subject to multiple rounds of human manipulation, resulting in insufficient data reliability and consequently, inaccurate manual forecasts of user demand. Summary of the Invention

[0004] Therefore, it is necessary to provide a method, apparatus, computer device, computer-readable storage medium, and computer program product for predicting the number of user objects that can improve the accuracy of predicting the number of objects, in response to the above-mentioned technical problems.

[0005] Firstly, this application provides a method for predicting the number of user objects. The method includes: obtaining the actual number of objects corresponding to the user to be predicted within a preset time period, and the prediction time for the user to be predicted; processing the actual number of objects and the prediction time based on the effect parameters and hyperparameters of a trained number prediction model to obtain a target prediction result; the effect parameters are obtained by updating the initial parameters of the number prediction model based on the test data and evaluation indicators of each user; the initial parameters are obtained by analyzing the actual number of objects corresponding to each user within the preset time period.

[0006] In one embodiment, before processing the actual number of objects and the time to be predicted based on the effect parameters and hyperparameters of the trained number prediction model, the method further includes: determining multiple effect parameters based on the business attributes of each user and the actual number of objects of the user in each preset time period; the effect parameters include at least one of weekly effect parameters, annual effect parameters, holiday effect parameters and special event effect parameters; and training the number prediction model based on each effect parameter, the preset hyperparameters and the actual number of objects of each user to obtain the trained number prediction model.

[0007] In one embodiment, the business attributes include user type and user object number share; based on the business attributes of each user and the actual number of objects of the user in each preset time period, multiple effect parameters are determined, including: based on the user type of each user, determining the corresponding special event and obtaining the number of objects corresponding to the special event; based on the user object number share of each user, determining the target user, and based on the user object number share, determining the first distribution data corresponding to the number of objects of the target user;

[0008] Based on the user object share of each user, determine the second distribution data corresponding to the object number of all users; based on the object number of special events, analyze the first and second distribution data to determine the special event effect parameters; analyze the peaks and troughs of the first and second distribution data to configure holiday effect parameters, annual effect parameters, and weekly effect parameters.

[0009] In one embodiment, the test data includes a test time period and the actual number of objects; the training of the number prediction model based on each effect parameter, preset hyperparameters, and the actual number of objects for each user to obtain a trained number prediction model includes:

[0010] Based on the effect parameters, preset hyperparameters, and the actual number of objects for each user, the number prediction model is trained to obtain the initial number prediction model.

[0011] Based on the test period and the initial number prediction model, the number of predicted objects for the test period is obtained.

[0012] Based on the number of predicted objects and the actual number of objects for each user, the evaluation index of the initial number prediction model is determined.

[0013] If the evaluation metric does not meet the preset training stopping condition, at least one of the hyperparameters and effect parameters is updated to obtain the updated parameters. Based on the updated parameters, the training of the number prediction model is re-executed based on each effect parameter, the preset hyperparameters, and the actual number of objects for each user to obtain the initial number prediction model. This process continues until the evaluation metric meets the training stopping condition, resulting in a well-trained number prediction model.

[0014] In one embodiment, after processing the actual number of objects and the prediction time based on the effect parameters and hyperparameters of the trained number prediction model to obtain the target prediction result, the method further includes:

[0015] Based on a pre-configured number update strategy, the number update results corresponding to the time to be predicted are generated. The number update strategy is determined based on the actual number of objects in multiple preset time periods corresponding to the time to be predicted, as well as the actual number of objects on the peak date corresponding to the time to be predicted.

[0016] The target prediction results are updated based on the number update results to obtain the updated target prediction results.

[0017] In one embodiment, generating the number update result corresponding to the time to be predicted based on a pre-configured number update strategy includes:

[0018] Determine the first actual number of objects corresponding to the time to be predicted; the first actual number of objects is the average number of objects in the months adjacent to the time to be predicted in the same year.

[0019] Determine the ratio of the peak object number to the second actual object number corresponding to the time to be predicted; the second actual object number is the average object number of the months adjacent to the time to be predicted in past years; the peak object number is the actual object number of the peak date corresponding to the time to be predicted in past years.

[0020] The product of the ratio and the first actual number of objects is determined as the updated number for the time to be predicted.

[0021] Secondly, this application also provides a user object number prediction device. The device includes:

[0022] The data acquisition module is used to acquire the actual number of objects corresponding to the user to be predicted within a preset time period, as well as the prediction time of the user to be predicted.

[0023] The prediction module is used to process the actual number of objects and the time to be predicted based on the effect parameters and hyperparameters of the trained number prediction model to obtain the target prediction result. The effect parameters are obtained by updating the initial parameters of the number prediction model based on the test data and evaluation indicators of each user. The initial parameters are obtained by analyzing the actual number of objects corresponding to each user in a preset time period.

[0024] Thirdly, this application also provides a computer device. The computer device includes a memory and a processor, the memory storing a computer program, and the processor executing the computer program to implement the steps of the method described in the first aspect.

[0025] Fourthly, this application also provides a computer-readable storage medium. The computer-readable storage medium stores a computer program thereon, which, when executed by a processor, implements the steps of the method described in the first aspect.

[0026] Fifthly, this application also provides a computer program product. The computer program product includes a computer program that, when executed by a processor, implements the steps of the method as described in the first aspect.

[0027] The aforementioned user object number prediction method, apparatus, computer equipment, storage medium, and computer program product first acquire the prediction time of the user to be predicted and the actual number of objects corresponding to the user in a preset time period. Then, the actual number of objects and the prediction time are processed using the effect parameters and hyperparameters of a trained number prediction model to obtain the target prediction result corresponding to the prediction time. The effect parameters are obtained by continuously updating the initial parameters of the number prediction model based on test data and evaluation indicators for each user. The initial parameters are obtained after analyzing the actual number of objects for each user in the preset time period. The initial parameters are obtained by the server based on the analysis of the actual number of objects for multiple users in the preset time period. These initial parameters add the object number prediction features of the actual number of objects to the number prediction model, thereby improving the accuracy of the number prediction model. Furthermore, the effect parameters of the trained number prediction model are continuously updated with test data and evaluation indicators to continuously improve the accuracy of the number prediction model in predicting the number of user objects, ultimately improving the accuracy of the number prediction model in predicting the number of user objects. Attached Figure Description

[0028] To more clearly illustrate the technical solutions in the embodiments or related technologies of this application, the drawings used in the description of the embodiments or related technologies will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0029] Figure 1 This is a diagram illustrating the application environment of a user object number prediction method in one embodiment.

[0030] Figure 2 This is a flowchart illustrating a method for predicting the number of user objects in one embodiment;

[0031] Figure 3 This is a flowchart illustrating the steps for determining effect parameters in one embodiment;

[0032] Figure 4 This is a flowchart illustrating the steps of training a number prediction model in one embodiment;

[0033] Figure 5 This is a flowchart illustrating the steps for determining evaluation indicators in one embodiment;

[0034] Figure 6 This is a flowchart illustrating the steps for determining the number update result in one embodiment;

[0035] Figure 7This is a flowchart illustrating the user object number prediction method in another embodiment;

[0036] Figure 8 This is a structural block diagram of a user object number prediction device in one embodiment;

[0037] Figure 9 This is an internal structural diagram of a computer device in one embodiment. Detailed Implementation

[0038] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0039] The user object number prediction method provided in this application embodiment can be applied to, for example, Figure 1 In the application environment shown, terminal 102 communicates with server 104 via a network. A data storage system can store the data that server 104 needs to process. The data storage system can be integrated onto server 104 or located in the cloud or on other network servers. Terminal 102 can continuously send real-time actual object counts to server 104, and server 104 can store these actual object counts in the data storage system. As time progresses, the actual object counts for a preset time period are stored in the data storage system. Alternatively, the data storage system can receive the actual object counts for each user within a preset time period. Terminal 102 can send the prediction time for the user to be predicted to server 104. Server 104 can obtain the actual object counts for the user to be predicted within the preset time period from the data storage system and input the actual object counts and prediction times into a trained number prediction model. Server 104 can analyze the input actual object counts based on the effect parameters and hyperparameters of the trained number prediction model and output the target prediction result for the user to be predicted within the prediction time. The server first analyzes the actual number of objects from multiple users within a preset time period to obtain the initial parameters of the number prediction model, and then updates the initial parameters using test data and evaluation metrics to obtain the trained number prediction model.

[0040] The terminal 102 can be, but is not limited to, various personal computers, laptops, smartphones, and tablets. The server 104 can be implemented using a standalone server or a server cluster consisting of multiple servers.

[0041] In one exemplary embodiment, such as Figure 2 As shown, a method for predicting the number of user objects is provided, which is then applied to... Figure 1Taking server 104 as an example, the explanation includes the following steps S202 to S204. Wherein:

[0042] Step S202: Obtain the actual number of objects corresponding to the user to be predicted within a preset time period, and the prediction time of the user to be predicted.

[0043] The preset time period refers to one or more historical time periods; the number of objects represents the actual or upcoming demand for express delivery orders generated by each user under the influence of marketing activities and business conditions; the actual number of objects represents the actual demand generated by users within a certain unit of time. The time to be predicted is a time period after the preset time period.

[0044] Specifically, the server can obtain the number of actual objects corresponding to each unit of time of the user to be predicted in the preset time period from the database, and the server can obtain the prediction time corresponding to the user to be predicted from the terminal.

[0045] Step S204: Based on the effect parameters and hyperparameters of the trained number prediction model, process the actual number of objects and the time to be predicted to obtain the target prediction result.

[0046] The effect parameters are obtained by updating the initial parameters of the number prediction model based on the test data and evaluation indicators of each user; the initial parameters are obtained by analyzing the actual number of objects for each user within a preset time period. The target prediction result is used to represent the number of predicted demands for each user within each unit of time during the prediction period.

[0047] Specifically, the server inputs the time to be predicted and the actual number of objects corresponding to the user to be predicted within a preset time period into the trained number prediction model. In the trained number prediction model, the server processes the effect parameters, hyperparameters, and the actual number of objects to obtain the object number prediction features of the user to be predicted within the preset time period, and determines the target prediction result of the user to be predicted within the predicted time period based on the obtained object number prediction features and the time to be predicted.

[0048] Before making a prediction, the server can first analyze the actual number of objects for each user in a preset time period to obtain the initial parameters of the number prediction model. It can also obtain the evaluation index of the number prediction model through test data and continuously update the initial parameters through the evaluation index until the evaluation index of the updated number prediction model meets the training requirements, and finally obtain the trained number prediction model.

[0049] In the aforementioned method for predicting the number of user objects, the prediction time for each user to be predicted and the actual number of objects corresponding to that user within a preset time period are first obtained. Then, the effect parameters and hyperparameters of a trained prediction model are used to process these actual number of objects and the prediction time to obtain the target prediction result corresponding to the prediction time. The effect parameters are obtained by continuously updating the initial parameters of the prediction model based on test data and evaluation metrics for each user. The initial parameters are obtained after analyzing the actual number of objects for each user within the preset time period. These initial parameters are derived from server analysis of the actual number of objects for multiple users within the preset time period. The initial parameters add the object number prediction features of the actual number of objects to the prediction model, thereby improving the accuracy of the prediction model. Furthermore, the effect parameters of the trained prediction model are continuously updated with test data and evaluation metrics to continuously improve the accuracy of the prediction model in predicting the number of user objects, ultimately enhancing the overall accuracy of the prediction model in predicting the number of user objects.

[0050] In an exemplary embodiment, before the step "processing the actual number of objects and the time to be predicted based on the effect parameters and hyperparameters of the trained number prediction model", the method further includes:

[0051] Based on the business attributes of each user and the actual number of objects for each user in each preset time period, multiple effect parameters are determined; based on each effect parameter, the preset hyperparameters and the actual number of objects for each user, the number prediction model is trained to obtain a trained number prediction model.

[0052] The effect parameters include at least one of the following: weekly effect parameters, annual effect parameters, holiday effect parameters, and special event effect parameters. The weekly effect parameter represents the characteristic of the actual number of objects each week within a preset time period; the annual effect parameter represents the characteristic of the actual number of objects each year within a preset time period; and the holiday effect parameter represents the characteristic of the actual number of objects similar to holidays within a preset time period. Business attributes represent the statistical value of each user's demand quantity and the user's business type.

[0053] Specifically, the server acquires each user's business attributes and determines the time-demand quantity correspondence between users based on the business type and demand quantity statistics within those attributes. It then updates this correspondence by combining this data with the actual number of objects for each user within each preset time period. The server analyzes this time-demand quantity correspondence to generate special event effect parameters. Furthermore, the server analyzes the actual number of objects within each preset time period and generates weekly, annual, and holiday effect parameters for each user, based on the characteristics of weeks, years, and holidays.

[0054] The time-demand quantity mapping is a pre-configured mapping on the server, while the time-demand quantity mapping is a pre-configured mapping for the distribution of demand quantities for users of different sizes and types. For example, the demand quantity of e-commerce users will increase significantly during specific marketing holidays, and the time corresponding to those holidays will be set as the time corresponding to special events or holidays. Or, the demand quantity of e-commerce users will decrease significantly during the Chinese New Year, and the time corresponding to the Chinese New Year will be set as the time corresponding to special events or holidays.

[0055] Based on this, the server can use the aforementioned multiple effect parameters, along with preset hyperparameters and the actual number of objects for each user, as training data to train the number prediction model. By adjusting the hyperparameters and effect parameters, the server continuously trains the number prediction model based on the actual number of objects for each user, thus obtaining a well-trained number prediction model.

[0056] In this embodiment, training the number prediction model with the actual number of objects of multiple users enables the number prediction model to learn the characteristics of the historical demand quantity of multiple users. Furthermore, training the number prediction model with multiple effect parameters of users enables the number prediction model to quickly learn the characteristics of the historical demand quantity of individual users, thereby improving the accuracy and training efficiency of the number prediction model.

[0057] In one exemplary embodiment, such as Figure 3 As shown, the business attributes include user type and user object number share. The specific implementation process of the step "determine multiple effect parameters based on the business attributes of each user and the actual number of objects of each user in each preset time period" includes steps S302 to S310. Among them:

[0058] Step S302: Based on the user type of each user, determine the corresponding special event and obtain the number of objects corresponding to the special event.

[0059] The business attributes include user type and user group share. User type can be various types, such as enterprise type and individual type. Enterprise type can include e-commerce enterprise, general enterprise, etc. User group share represents the proportion of each user's demand in the total demand.

[0060] Specifically, the server can determine the specific events corresponding to a user's user type and then retrieve the corresponding number of objects within the time period specified by the specific event. For example, if the specific event is a marketing holiday, the server can retrieve the actual number of objects corresponding to each marketing holiday within a preset time period.

[0061] Step S304: Based on the share of user object counts for each user, determine the target user, and based on the share of user object counts, determine the first distribution data corresponding to the number of objects of the target user.

[0062] Among them, target users are used to represent important users among all users, and important users can be determined by the number of users' needs.

[0063] Specifically, the server obtains the share of user objects for each user, sorts the users according to their share of user objects from largest to smallest, obtains a user sequence, and identifies the first preset number of users in the user sequence as target users. Alternatively, users whose share of user objects is greater than a preset value are identified as target users. For each target user, the server counts the actual number of objects for that target user within a preset time period, obtaining distribution data corresponding to that target user. This results in a first distribution data set composed of the distribution data of multiple target users, containing the number of actual objects for each target user within each unit of time within the preset time period, with each unit of time corresponding to one number of actual objects.

[0064] Step S306: Based on the share of user object counts for each user, determine the second distribution data corresponding to the number of objects for all users.

[0065] Specifically, the server calculates the share of user objects for each user, obtaining the distribution data of the actual number of objects for each user within a preset time period. It then merges the distribution data from multiple users to determine the second distribution data corresponding to the number of objects for all users. For example, given users A, B, and C, the server determines the distribution data of user A's demand quantity over the past year, user B's demand quantity over the past year, and user C's demand quantity over the past year, respectively, based on their user object share. Finally, the server sums these three demand quantity distribution data to obtain the second distribution data.

[0066] Step S308: Based on the number of objects involved in the special event, analyze the first distribution data and the second distribution data to determine the special event effect parameters.

[0067] Specifically, for each user, the server obtains the number of objects corresponding to the special event at the specific time for that user, and obtains the actual number of objects at the specific time in the first distribution data and the actual number of objects at the specific time in the second distribution data. The number of objects is then compared with these two actual object counts to determine the difference between the number of objects for that user at the specific time and the corresponding actual object count. Based on these two differences, the special event effect parameter for that user is determined. For example, if the special event is New Year's Day and the specific time is from December 31st to January 3rd, the number of objects for that user is 'l', while the actual number of objects at the specific time in the first distribution data is 'n', and the actual number of objects at the specific time in the second distribution data is 'm'. The difference between 'n' and 'l', and the difference between the average value of 'm' and 'l', are obtained. Based on the sign of the difference and the ratio of the difference to 'l', the special event effect parameter for the special event in the first distribution data and the special event effect parameter for the special event in the second distribution data are determined. The two special event effect parameters are then combined to obtain the special event effect parameter for that user.

[0068] Step S310: Analyze the peaks and troughs of the first and second distribution data, and configure the holiday effect parameters, annual effect parameters, and weekly effect parameters.

[0069] The first and second distribution data can be converted into a line graph with time on the horizontal axis and the number of actual objects on the vertical axis.

[0070] Specifically, the server analyzes the line graph of the first distribution data to obtain the peaks and troughs corresponding to the target users, and configures the holiday effect parameters, annual utility parameters, and weekly effect parameters for the target users based on the timing of the peaks and troughs. The server can also analyze the line graph of the second distribution data to obtain the peaks and troughs corresponding to all users, and configure the holiday effect parameters, annual utility parameters, and weekly effect parameters for all users based on the timing of the peaks and troughs, which are then used as effect parameters for non-target users.

[0071] Optionally, the server can perform a weighted summation of the target user's holiday effect parameters, annual utility parameters, and weekly effect parameters with those of all users to obtain the final holiday effect parameters, annual utility parameters, and weekly effect parameters. The weights can be determined based on the number and share of user objects.

[0072] In this embodiment, the number of objects corresponding to a special event is determined by user type, and the first distribution data corresponding to the target user and the second distribution data corresponding to all users are determined. The number of objects at a special time is analyzed against the first and second distribution data to obtain the special event effect parameters. The peaks and troughs of the first and second distribution data are analyzed to obtain the holiday effect parameters, annual effect parameters, and weekly effect parameters. This can accurately determine multiple effect parameters corresponding to the number prediction model, improving the efficiency and accuracy of obtaining effect parameters.

[0073] In one exemplary embodiment, such as Figure 4 As shown, the test data includes the test time period and the actual number of objects. The specific implementation process of the step "training the number prediction model based on each effect parameter, preset hyperparameters, and the actual number of objects for each user to obtain the trained number prediction model" includes steps S402 to S408. Among them:

[0074] Step S402: Based on each effect parameter, the preset hyperparameters, and the actual number of objects for each user, the number prediction model is trained to obtain the initial number prediction model.

[0075] Specifically, the server can use multiple effect parameters, as well as preset hyperparameters and the actual number of objects for each user, as training data and input them into the number prediction model for training. After one round of training, an initial number prediction model can be obtained.

[0076] Step S404: Based on the test time period and the initial number prediction model, make a prediction to obtain the number of predicted objects for the test time period.

[0077] The test time period refers to multiple time periods corresponding to the test data, and each test time period includes the number of actual objects per unit of time.

[0078] Specifically, the server takes the test time period and the actual number of objects corresponding to the user in the preset time period as input, and feeds them into the initial number prediction model. The initial number prediction model outputs the predicted number of objects for the test time period.

[0079] Step S406: Determine the evaluation index of the initial number prediction model based on the number of predicted objects and the actual number of objects for each user.

[0080] The evaluation metrics can be determined by evaluating the accuracy of a single user, or they can be determined based on the accuracy of multiple users.

[0081] Specifically, the server can calculate the difference between the number of predicted objects and the actual number of objects in the test data for each unit of time within the test period. If the difference is less than a preset threshold, the prediction result is considered successful. If the difference is greater than or equal to the preset threshold, the prediction result is considered unsuccessful. For a specific user, the difference can be used to determine the prediction result for each unit of time and the percentage of successful predictions within the test period, which represents the user's accuracy.

[0082] In one example, accuracy can also be calculated by first determining the absolute value of the difference between the number of predicted objects per unit time and the number of actual objects per unit time, and then determining the quotient of the absolute value and the number of actual objects per unit time as the accuracy per unit time. Finally, the average of the accuracy per unit time during the test period is calculated as the overall accuracy.

[0083] Step S408: If the evaluation index does not meet the preset training stopping condition, then at least one of the hyperparameters and effect parameters is updated to obtain the updated parameters. Based on the updated parameters, the training of the number prediction model based on each effect parameter, the preset hyperparameters, and the actual number of objects for each user is re-executed to obtain the initial number prediction model, until the evaluation index meets the training stopping condition, and the trained number prediction model is obtained.

[0084] Specifically, the server can determine the evaluation metric corresponding to the initial number prediction model and whether the evaluation metric meets the preset training stopping condition. For example, the training stopping condition could be that the accuracy of a certain user is greater than a preset training threshold, or that the average accuracy of multiple users is greater than a preset training threshold. If the evaluation metric does not meet the preset training stopping condition, the server can adjust one or more parameters, such as hyperparameters and effect parameters, according to a preset training adjustment strategy to obtain updated parameters. Based on the updated parameters, the server can re-execute the above steps for training the number prediction model, thereby completing the update of the initial number prediction model, and continue to iteratively determine the evaluation metric of the initial number prediction model until the evaluation metric meets the preset training stopping condition, thus obtaining the trained number prediction model.

[0085] In this embodiment, by updating at least one of the effect parameters and hyperparameters, the initial number prediction model is continuously updated based on the updated parameters, and the evaluation index of the initial number prediction model is determined until the evaluation index meets the preset training stopping condition, thereby obtaining a trained number prediction model. This can improve the training speed of the number prediction model and the accuracy of the trained number prediction model.

[0086] In an exemplary embodiment, the number of predicted objects includes the number of predicted objects per unit time corresponding to the test period; the number of actual objects includes the number of actual objects per unit time corresponding to the test period. For example... Figure 5 As shown, the specific implementation process of the step "determine the evaluation index of the initial number prediction model based on the number of predicted objects and the actual number of objects for each user" includes steps S502 to S506. Wherein:

[0087] Step S502: For each unit of time in the test period, the unit time accuracy is obtained based on the difference between the predicted number of objects per unit time and the actual number of objects per unit time, as well as the actual number of objects per unit time.

[0088] Specifically, the server iterates through each unit of time in the test period, determines the difference between the predicted number of objects per unit of time and the actual number of objects per unit of time, and the ratio of this difference to the actual number of objects per unit of time, which is the accuracy per unit of time.

[0089] Step S504: Obtain the unit time accuracy of each user in the same unit time, and determine the unit time compliance rate based on the number of users whose unit time accuracy is greater than a preset threshold.

[0090] Specifically, the server determines the unit-time accuracy of multiple users within the same unit of time, and determines the number of users whose unit-time accuracy is greater than a preset threshold. The ratio of this number to the total number of users is then defined as the unit-time compliance rate. For example, the preset threshold can be set to 80%, and a unit-time accuracy rate greater than 80% indicates that the user has met the target.

[0091] Step S506: Determine the statistical values ​​of the unit time accuracy and unit time achievement rate for each user, which are used as evaluation indicators for the initial number prediction model.

[0092] Specifically, evaluation metrics can be determined based on the first average of the accuracy per unit time for each user, and the second average of the achievement rates per unit time for multiple users. Evaluation metrics can also be determined by a third average corresponding to multiple first averages over the test period. For example, if user A's two sequentially ordered unit-time accuracies are 100% and 100%, and user B's two sequentially ordered unit-time accuracies are 60% and 80%, and the test period contains two units of time, the server can determine the first average as 80% and 90%, and the third average as 85%. For example, if the initial number prediction model achieves an achievement rate of 60% and 40% per unit time over two units of time during the test period, the second average can be determined to be 50%.

[0093] It should be understood that evaluation metrics are not limited to being determined solely by accuracy per unit time, or solely by compliance per unit time, or by combining the accuracy and compliance rates of multiple users. Evaluation metrics can be configured according to user needs.

[0094] In this embodiment, the unit-time accuracy rate is calculated for each time unit within the test period, and the unit-time achievement rate is determined based on the unit-time accuracy rates of multiple users. Finally, based on the unit-time accuracy rates of multiple users across multiple time units and the unit-time achievement rate, the evaluation index of the initial number prediction model can be obtained, thereby improving the data dimension of the evaluation index and enhancing its evaluation accuracy.

[0095] In an exemplary embodiment, after the step "processing the actual number of objects and the time to be predicted based on the effect parameters and hyperparameters of the trained number prediction model to obtain the target prediction result", the method further includes:

[0096] Based on a pre-configured number update strategy, the number update results corresponding to the time to be predicted are generated; the target prediction results are updated based on the number update results to obtain the updated target prediction results.

[0097] The number update strategy is used to update the prediction results for certain units of time in the target prediction results. The number update strategy is determined based on the actual number of objects in multiple preset time periods corresponding to the time to be predicted, and the actual number of objects on the peak date corresponding to the time to be predicted.

[0098] Specifically, the server generates updated numbers for each unit of time within the predicted time period for the user to be predicted, based on a pre-configured number update strategy. The server then updates one or more of these updated numbers to the target prediction result according to actual needs. In one example, the target prediction result contains predictions for three units of time, denoted as l, m, and n. The number update strategy determines these updated numbers for three units of time, denoted as a, b, and c. The server determines that the unit of time requiring updating is the second unit of time. Therefore, after updating the number update results to the target prediction result, the updated target prediction result becomes l, b, and n.

[0099] In this embodiment, by updating some of the results in the target prediction results through a number update strategy, the updated target prediction results can be obtained, which can further improve the accuracy of the target prediction results.

[0100] In one exemplary embodiment, such as Figure 6 As shown, the specific implementation process of the step "generating the number update result corresponding to the time to be predicted based on the pre-configured number update strategy" includes steps S602 to S606. Wherein:

[0101] Step S602: Determine the number of the first actual objects corresponding to the time to be predicted.

[0102] The first actual number of objects is the average number of objects in the months adjacent to the time of prediction within the same year.

[0103] Specifically, for a specific unit of time within the timeframe to be predicted, the server can obtain the average number of objects per day in the previous month corresponding to that unit of time. Alternatively, the server can obtain the average number of objects per day in the previous two months; however, this embodiment is not limited to this.

[0104] Optionally, when determining the average demand quantity, the peak demand quantity for the month can be removed. The peak demand quantity can be the highest number of objects corresponding to each unit of time in the month.

[0105] Step S604: Determine the ratio of the number of peak objects to the number of second actual objects corresponding to the time to be predicted.

[0106] The second actual object number is the average number of objects in the months adjacent to the time to be predicted in the past years; the peak object number is the actual object number on the peak date corresponding to the time to be predicted in the past years.

[0107] Specifically, for a given unit of time in the time to be predicted, the server can obtain the average number of objects in the adjacent months corresponding to that unit of time in past years (e.g., last year, the year before, etc.). It should be understood that these adjacent months should be the same as the adjacent months in step S602. The server can obtain the actual number of objects on the peak date in the time to be predicted in past years and determine it as the peak object number. The server determines the ratio of the peak object number to the second actual object number corresponding to that unit of time.

[0108] Step S606: Determine the product of the ratio and the number of the first actual objects, which is the updated number result corresponding to the time to be predicted.

[0109] Specifically, for a given unit of time within the timeframe to be predicted, the server calculates the product of the ratio and the first actual number of objects, which becomes the updated number for that unit of time. Once all unit-time intervals within the timeframe to be predicted have been iterated, multiple updated number results for each unit of time are obtained and determined as the updated number for the timeframe to be predicted.

[0110] In this embodiment, by determining the ratio of the peak number of objects in each unit of time within the time to be predicted to the second actual number of objects corresponding to the time to be predicted, and the product of the first actual number of objects and the ratio, the number update result per unit of time is obtained. Based on the past number of objects, the number update result can be accurately obtained, thereby improving the prediction accuracy of the time to be predicted.

[0111] In an exemplary embodiment, the specific implementation process of the step "updating the target prediction result based on the number update result to obtain the updated target prediction result" includes:

[0112] If it is determined that the evaluation index of the user to be predicted in the number prediction model does not meet the first preset condition, or if it is determined that the target prediction result does not meet the second preset condition, then an abnormal time in the time to be predicted is determined; based on the number update result corresponding to the abnormal time, the target prediction result is updated to obtain the updated target prediction result.

[0113] The first preset condition can be that the accuracy of the number prediction model for the users to be predicted meets a first preset value, or the achievement rate of the number prediction model meets a second preset value. The second preset condition can be that the target prediction result for each unit of time in the time to be predicted is not 0, or is greater than or equal to a third preset value.

[0114] Specifically, the server assesses each unit of time within the predicted time period to determine whether the evaluation metrics of the user to be predicted in the number prediction model meet the first preset condition, and whether the target prediction result output by the number prediction model meets the second prediction condition. If both the first and second preset conditions are met, there is no need to update the target prediction result. If either the first or second preset condition is not met, the corresponding unit of time within the predicted time period is determined to be an abnormal time, and the number update result corresponding to the abnormal time is updated to the target prediction result.

[0115] In this embodiment, the first preset condition and the second preset condition determine whether the target prediction result needs to be updated for certain units of time during the prediction time. If an update is required, certain units of time are identified as abnormal times, and the target prediction result corresponding to the abnormal times is updated to obtain the updated target prediction result, thereby improving the accuracy of the target prediction result.

[0116] like Figure 7 As shown, the following describes the specific execution process of the above-mentioned user object number prediction method in detail with reference to a specific embodiment, including the following steps:

[0117] Step 1, Data Preparation. The server, combining the business attributes of each user, analyzes the actual number of objects within a preset time period, compares the demand of key user group A over the past n years, and determines the weekly effect parameter, annual effect parameter, holiday effect parameter, special event effect parameter, and other hyperparameter settings.

[0118] The business attributes can include: whether the user is an e-commerce user (to determine if the user is eligible for an e-commerce promotional event) and the user's demand share. If a user is identified as one of the top N users based on their demand share, they are considered a large user. Because large users have a large demand share and a significant impact on the model, holiday effect parameters need to be customized separately.

[0119] Specifically, by judging business attributes and the historical demand volume over the past N years (the actual number of objects that won the bet within a preset time period), the server can determine the historical line graph of the total demand volume of all users, as well as the historical line graph of the demand volume of the top N users. By comparing the peaks and troughs during various statutory holidays, e-commerce festivals, and non-statutory holidays in historical years, the server can determine the holiday effect parameters, annual effect parameters, and weekly effect parameters.

[0120] For example, user A analyzes the demand trend charts for 2022 and 2023 and determines that June 1st to June 20th is the peak demand period. The server can be configured with holiday parameters as follows: ID = 'A', ds = '20220601', holiday = '618_01'; ID = 'A', ds = '20220602', holiday = '618_02'; ...; ID = 'A', ds = '20230620', holiday = '618_20'. Furthermore, the demand trend charts show sharp increases during periods such as Singles' Day (November 11th), 99 Shopping Festival (September 9th), Double Twelve (December 12th), Spring Festival, and company anniversaries; these dates can be configured as holiday parameters. If the user's demand does not show a regular increase or decrease throughout the 7 days of the week, the weekly effect parameter can be turned off. If the user's demand decreases during statutory holidays (365 days a year), configure the annual effect parameters ds = '20220501', ds = '20230501', ds = '20240501'. If the user's demand is not affected by urban control measures during 2022, do not configure the special event effect.

[0121] Step 2: Modeling the basic prediction model for the quantity of large user demand. Feature engineering is performed on data from the past n years. Combined with the effect parameters analyzed above, code is developed, the model is built, debugged (updated), and the initial predicted demand quantity information is output through the quantity prediction model.

[0122] Feature engineering includes: dirty data processing, such as standardizing the time format to "yyyy-MM-dd". If some data has a time format of "20240801", it needs to be changed to "2024-08-01". Feature engineering also includes data removal, such as removing users with short cooperation periods who do not meet the modeling training time requirements. For example, user C's cooperation period with the logistics company is only one month, and there is no longer historical data, so user C needs to be removed.

[0123] Furthermore, the quantity forecasting model outputs predicted demand information including: the forecast date and the predicted demand quantity. For example, on August 1st, it can predict the demand quantity for each day from August 2nd to August 15th. It should be noted that the forecast predicts the demand quantity for days that have not yet occurred. For instance, a forecast for August 1st will always predict the demand quantity for that day or some day thereafter; data before August 1st is historical data. The server will not predict the demand quantity for July 31st on August 1st. Test data is not applicable to the above conclusions; test data can be historical data.

[0124] When determining the evaluation metrics for the model, the test data includes the test period and the corresponding actual demand quantity. For example, if the model predicts the demand quantity from August 1st to August 15th on August 1st, the test period for the test data is from August 1st to August 15th. It should be understood that the model's accuracy and success rate cannot be calculated on August 1st itself, because the actual demand quantity corresponding to that test period is not yet available. The model's accuracy and success rate for August 1st can be determined on August 2nd, and the accuracy and success rate for August 2nd can be determined on August 3rd.

[0125] Step 3, Model Evaluation. The model predicts daily demand over the past m months, obtaining the difference between predicted and actual demand. It then calculates the overall accuracy rate for large users, the accuracy rate for individual users, and the achievement rate, among other metrics, to preliminarily evaluate the model's effectiveness.

[0126] The accuracy rate is calculated as follows: The difference between the user's predicted demand for the day and the user's actual demand for the day is determined, and the ratio of this difference to the user's actual demand for the day is the accuracy rate. The achievement rate is calculated as follows: On a given day, the ratio of the number of users with a daily accuracy rate of 80% or higher (the specific threshold is confirmed by the business team, generally between 70% and 90%) to the total number of users is the achievement rate. For example, on August 1st, if there are 100 users in total, and 70 of them have a daily accuracy rate of 80% or higher, then the achievement rate for that day is determined to be 70%.

[0127] Specifically, if the model's evaluation metrics are unsatisfactory, the server can optimize the model by adjusting the Prophet parameters (hyperparameter tuning), including adjusting hyperparameters such as growth, changepoint, and n_changepoint. Furthermore, for users with low accuracy and a large share of user demand, the demand trend over the past two years can be analyzed separately, and holiday effect parameters can be configured separately (for example, user A experiences a sharp increase in demand on February 14th every year, while all other users show no demand fluctuation on February 14th; in this case, user A's holiday effect needs to be configured separately on February 14th, for example, by specifying ID = A, holiday = 'Valentine's Day', and ds = '20240214'). After configuration, when the model learns from the data, the demand volume on February 14th of each historical year for user A can be marked as a holiday. Each subsequent February 14th will then have a corresponding demand increase coefficient, thus more closely approximating the actual sharp increase in demand.

[0128] Furthermore, model predictions cannot be improved indefinitely in terms of accuracy. For example, User B's demand for the 618 shopping festival was around 100,000 units annually for the previous three years, and the model predicted the same for 2024. However, due to poor management, User B was unable to participate in the 618 event in 2024, causing the demand to drop to 10,000 units. This scenario would lower the accuracy, and the model would be unable to predict based on historical demand. In one example, an accuracy rate of 90% and a compliance rate of 40% could be set as thresholds for the model to meet its performance evaluation criteria.

[0129] Step 4: Development of Base Period Mean Cusp Rule. For users whose accuracy and compliance rates remain consistently low over a period of time—i.e., those difficult to predict—a base period mean fallback rule can be used. Additionally, for users whose predicted demand quantity is 0 and clearly does not match actual business needs, a base period mean fallback rule can also be used. The server uses the base period mean fallback rule to obtain the fallback demand quantity P, which replaces the predicted demand quantity in the quantity prediction model, and outputs the final overall predicted demand quantity.

[0130] For example, User A has consistently had a daily demand of at least 500 items over the past few years. However, the model predicts that the company will have a forecast of zero demand for 15 days in the next month, indicating that the forecast demand clearly does not match the actual business situation.

[0131] In one example, for user A, the quantity prediction model repeatedly predicts that the user's demand quantity is 0. The business side, based on the actual business situation, determines that the user's actual demand quantity will not be 0. Based on this, the server uses the base period average as a fallback rule for user A. According to the fallback rule, the predicted demand quantity for August 5th is 439 items. The subsequent demand quantity for user A with ID in the prediction table will be the predicted demand quantity generated according to the base period average as a fallback rule. For other users who do not use the base period average as a fallback rule, the prediction result of the quantity prediction model will be the final result.

[0132] It should be understood that the formula for the base period mean catch-all rule is as follows:

[0133] The number of users with guaranteed demand during the peak period of 24 years is P = A * (C / B).

[0134] A: The average monthly demand for this user over the past 24 years (excluding peak business days);

[0135] B: The average monthly demand for this user over the past 23 years (excluding peak business days);

[0136] C: The peak daily demand for this user in 2023.

[0137] In this embodiment of the application, there are no restrictions on the year and month. The year can be 24 years or any year, and the month can be one month or several months.

[0138] In one example, to predict the demand for August 2nd on August 1st, A represents the demand from July 1st to August 1st, 2024 (excluding the peak date of July 13th), B represents the demand from July 1st to August 1st, 2023 (excluding the peak date of July 13th), and C represents the demand for August 2nd, 2023.

[0139] Step 5: Deploy the above integrated model (the trained numerical prediction model) and update the prediction data daily.

[0140] In this embodiment, the Prophet time series machine learning model is used to analyze the historical demand volume of large users. Combined with logistics characteristics, various feature effect parameters are configured individually. Simultaneously, with the addition of fallback business rules, the final predicted user demand volume is comprehensively output. This solves the problems of relying on user managers for prediction and assessment in reality, which are characterized by strong subjectivity, poor interpretability, inaccuracy, and instability, thus improving the accuracy of user demand volume prediction.

[0141] It should be understood that although the steps in the flowcharts of the above embodiments are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the above embodiments may include multiple steps or multiple stages. These steps or stages are not necessarily completed 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 performed alternately or in turn with other steps or at least some of the steps or stages of other steps.

[0142] Based on the same inventive concept, this application also provides a user object number prediction device for implementing the user object number prediction method described above. The solution provided by this device is similar to the implementation described in the above method; therefore, the specific limitations in one or more user object number prediction device embodiments provided below can be found in the limitations of the user object number prediction method described above, and will not be repeated here.

[0143] In one exemplary embodiment, such as Figure 8 As shown, a user object number prediction device 800 is provided, including: a data acquisition module 801 and a number prediction module 802, wherein:

[0144] The data acquisition module 801 is used to acquire the number of actual objects corresponding to the user to be predicted in a preset time period, as well as the prediction time of the user to be predicted.

[0145] The number prediction module 802 is used to process the actual number of objects and the time to be predicted based on the effect parameters and hyperparameters of the trained number prediction model to obtain the target prediction result. The effect parameters are obtained by updating the initial parameters of the number prediction model based on the test data and evaluation indicators of each user. The initial parameters are obtained by analyzing the actual number of objects corresponding to each user in the preset time period.

[0146] Furthermore, the device also includes a model training module, specifically used for: determining multiple effect parameters based on the business attributes of each user and the actual number of objects of each user in each preset time period; the effect parameters include at least one of weekly effect parameters, annual effect parameters, holiday effect parameters and special event effect parameters; and training the number prediction model based on each effect parameter, preset hyperparameters and the actual number of objects of each user to obtain a trained number prediction model.

[0147] Furthermore, the business attributes include user type and user object number share; the model training module is specifically used for: determining the corresponding special events based on the user type of each user, and obtaining the number of objects corresponding to the special events; determining the target users based on the user object number share of each user, and determining the first distribution data corresponding to the number of objects of the target users based on the user object number share; determining the second distribution data corresponding to the number of objects of all users based on the user object number share of each user; analyzing the first and second distribution data based on the number of objects of the special events to determine the special event effect parameters; and analyzing the peaks and troughs of the first and second distribution data to configure holiday effect parameters, annual effect parameters, and weekly effect parameters.

[0148] Furthermore, the test data includes the test time period and the actual number of objects; the model training module is specifically used for: training the number prediction model based on each effect parameter, preset hyperparameters, and the actual number of objects for each user to obtain an initial number prediction model; making predictions based on the test time period and the initial number prediction model to obtain the predicted number of objects for the test time period; determining the evaluation index of the initial number prediction model based on the predicted number of objects and the actual number of objects for each user; if the evaluation index does not meet the preset training stopping condition, updating at least one of the hyperparameters and effect parameters to obtain updated parameters, and re-executing the steps of training the number prediction model based on each effect parameter, preset hyperparameters, and the actual number of objects for each user to obtain the initial number prediction model based on the updated parameters, until the evaluation index meets the training stopping condition, and obtaining the trained number prediction model.

[0149] Furthermore, the number of predicted objects includes the number of predicted objects per unit time corresponding to the test period; the number of actual objects includes the number of actual objects per unit time corresponding to the test period. The model training module is specifically used for: obtaining the unit time accuracy for each unit time in the test period based on the difference between the number of predicted objects per unit time and the number of actual objects per unit time, as well as the number of actual objects per unit time; obtaining the unit time accuracy for each user in the same unit time, and determining the unit time achievement rate based on the number of users whose unit time accuracy is greater than a preset threshold; and determining the statistical values ​​of the unit time accuracy and unit time achievement rate for each user as evaluation indicators for the initial number prediction model.

[0150] Furthermore, the device also includes a result update module, specifically used for: generating a number update result corresponding to the time to be predicted based on a pre-configured number update strategy; the number update strategy is determined based on the actual number of objects in multiple preset time periods corresponding to the time to be predicted, and the actual number of objects on the peak date corresponding to the time to be predicted; updating the target prediction result based on the number update result to obtain the updated target prediction result.

[0151] Furthermore, the result update module is specifically used for: determining the first actual number of objects corresponding to the time to be predicted; the first actual number of objects is the average number of objects in the months adjacent to the time to be predicted in the same year; determining the ratio of the peak number of objects to the second actual number of objects corresponding to the time to be predicted; the second actual number of objects is the average number of objects in the months adjacent to the time to be predicted in previous years; the peak number of objects is the actual number of objects on the peak date corresponding to the time to be predicted in previous years; and determining the product of the ratio and the first actual number of objects as the update result of the number corresponding to the time to be predicted.

[0152] Furthermore, the result update module is specifically used to: if it is determined that the evaluation index of the user to be predicted in the number prediction model does not meet the first preset condition, or if it is determined that the target prediction result does not meet the second preset condition, then determine the abnormal time in the time to be predicted; based on the number update result corresponding to the abnormal time, update the target prediction result to obtain the updated target prediction result.

[0153] Each module in the aforementioned user object number prediction device can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device, or stored in the memory of a computer device as software, so that the processor can call and execute the operations corresponding to each module.

[0154] In one exemplary embodiment, a computer device is provided, which may be a server, and its internal structure diagram may be as follows: Figure 9As shown, the computer device includes a processor, memory, input / output (I / O) interfaces, and a communication interface. The processor, memory, and I / O interfaces are connected via a system bus, and the communication interface is also connected to the system bus via the I / O interfaces. The processor provides computational and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and a database. The internal memory provides the environment for the operating system and computer programs in the non-volatile storage media to run. The database stores the actual number of objects to be predicted for a user within a preset time period. The I / O interfaces are used for exchanging information between the processor and external devices. The communication interface is used for communicating with external terminals via a network connection. When the computer program is executed by the processor, it implements a method for predicting the number of user objects.

[0155] Those skilled in the art will understand that Figure 9 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0156] In one exemplary embodiment, a computer device is also provided, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps in the above-described method embodiments.

[0157] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon that, when executed by a processor, implements the steps in the above method embodiments.

[0158] In one embodiment, a computer program product is provided, including a computer program that, when executed by a processor, implements the steps in the above method embodiments.

[0159] 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 used for analysis, data stored, data displayed, 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 the relevant data must comply with relevant regulations.

[0160] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments described above. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. 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), magnetic 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 take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM). The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, etc., and are not limited to these.

[0161] The technical features of the above embodiments can be combined in any way. For the sake of brevity, 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, they should be considered to be within the scope of this specification.

[0162] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of this patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this application should be determined by the appended claims.

Claims

1. A method of predicting a number of user objects, characterized by, The method comprises: acquiring an actual object number corresponding to a preset time period of a user to be predicted, and a to-be-predicted time of the user to be predicted; processing the actual object number and the to-be-predicted time based on effect parameters and hyperparameters of a trained number prediction model, to obtain a target prediction result; the effect parameters are obtained by updating initial parameters of the number prediction model based on test data and evaluation indexes of each user; the initial parameters are obtained by analyzing actual object numbers of each user in the preset time period.

2. The method of claim 1, wherein, Before processing the actual object number and the to-be-predicted time based on the effect parameters and the hyperparameters of the trained number prediction model, the method further comprises: determining a plurality of effect parameters based on business attributes of each user and actual object numbers of the user in each preset time period; the effect parameters include at least one of a week effect parameter, a year effect parameter, a holiday effect parameter, and a special event effect parameter; training a number prediction model based on each effect parameter, preset hyperparameters, and actual object numbers of each user, to obtain a trained number prediction model.

3. The method of claim 2, wherein, The business attributes include user types and user object number shares; determining a plurality of effect parameters based on business attributes of each user and actual object numbers of the user in each preset time period comprises: determining a corresponding special event based on the user type of each user, and acquiring an object number corresponding to the special event; determining a target user based on the user object number share of each user, and determining first distribution data corresponding to an object number of the target user based on the user object number share; determining second distribution data corresponding to object numbers of all users based on the user object number share of each user; analyzing the first distribution data and the second distribution data based on the object number of the special event, to determine a special event effect parameter; analyzing wave crests and wave troughs of the first distribution data and the second distribution data, to configure a holiday effect parameter, a year effect parameter, and a week effect parameter.

4. The method of claim 2, wherein, The test data includes a test time period and an actual object number; training a number prediction model based on each effect parameter, preset hyperparameters, and actual object numbers of each user, to obtain a trained number prediction model, comprises: training a number prediction model based on each effect parameter, preset hyperparameters, and actual object numbers of each user, to obtain an initial number prediction model; performing prediction based on the test time period and the initial number prediction model, to obtain a predicted object number of the test time period; determining an evaluation index of the initial number prediction model based on the predicted object number and the actual object number of each user. If the evaluation index does not satisfy the preset training stop condition, at least one of the hyperparameters and the effect parameters is updated to obtain updated parameters, and the step of training the number prediction model based on each of the effect parameters, the preset hyperparameters and the actual number of objects of each of the users is re-executed based on the updated parameters until the evaluation index satisfies the training stop condition, and a trained number prediction model is obtained.

5. The method of claim 4, wherein, The predicted object number includes a plurality of unit time predicted object numbers corresponding to the test time period; the actual object number includes a plurality of unit time actual object numbers corresponding to the test time period; and the evaluation index of the initial number prediction model is determined based on the predicted object number and the actual object number of each of the users, including: For each unit time in the test time period, a unit time accuracy is obtained based on a difference value between the unit time predicted object number and the unit time actual object number, and the unit time actual object number; A unit time accuracy of each of the users in the same unit time is obtained, and a unit time compliance rate is determined based on a number of users whose unit time accuracy is greater than a preset threshold. The evaluation index of the initial number prediction model is determined based on the statistical values of the unit time accuracy and the unit time compliance rate corresponding to each of the users.

6. The method of claim 1, wherein, After the target prediction result is obtained by processing the actual object number and the to-be-predicted time based on the trained number prediction model, the effect parameters and the hyperparameters, the method further includes: Based on the preconfigured number update strategy, a number update result corresponding to the to-be-predicted time is generated; the number update strategy is determined based on the actual object numbers of a plurality of preset time periods corresponding to the to-be-predicted time, and the actual object number of a peak date corresponding to the to-be-predicted time; The target prediction result is updated based on the number update result to obtain an updated target prediction result.

7. The method of claim 6, wherein, The number update result corresponding to the to-be-predicted time is generated based on the preconfigured number update strategy, including: A first actual object number corresponding to the to-be-predicted time is determined; the first actual object number is an average object number of adjacent months in the same year as the to-be-predicted time; A ratio of a peak object number to a second actual object number corresponding to the to-be-predicted time is determined; the second actual object number is an average object number of adjacent months in a past year as the to-be-predicted time; and the peak object number is an actual object number of a peak date corresponding to the to-be-predicted time in the past year; A product value of the ratio and the first actual object number is determined as the number update result corresponding to the to-be-predicted time.

8. A computer device comprising a memory and a processor, the memory storing a computer program, characterized in that, The processor executes the computer program to implement the steps of the method of any one of claims 1 to 7.

9. A computer-readable storage medium having stored thereon a computer program, characterized in that The computer program is executed by the processor to implement the steps of the method of any one of claims 1 to 7.

10. A computer program product comprising a computer program, characterized in that, The computer program, which when executed by the processor, implements the steps of the method of any one of claims 1 to 7.

Images