Model training method, task processing method and network flow data prediction method

Abstract

The embodiment of the invention provides a model training method, a task processing method and a network flow data prediction method, and the model training method comprises the steps: determining historical sequence data and label sequence data; inputting the historical sequence data into a task processing model to obtain predicted sequence data output by the task processing model; determining prediction condition distribution corresponding to the prediction sequence data and label condition distribution corresponding to the label sequence data; and training the task processing model according to the prediction condition distribution and the label condition distribution, so that the prediction condition distribution is aligned with the label condition distribution.

Description

Technical Field

[0001] The embodiments in this specification relate to the field of artificial intelligence technology, and in particular to model training methods, task processing methods, and network traffic data prediction methods. Background Technology

[0002] Time series forecasting refers to predicting the values ​​or trends of one or more variables at one or more points in the future, based on historical data of one or more variables over a period of time. For example, in the transportation sector, time series data from historical periods can be used to predict road traffic flow, passenger flow, flight delays, and travel demand information at one or more points in the future.

[0003] In practical applications, multi-task learning is often used for multi-step time series forecasting. A shared model can be used to learn multiple related tasks simultaneously to achieve multi-step prediction. However, current training of shared models typically assumes that the predicted labels at each step are independent, ignoring the correlation between tasks. This makes it difficult for the shared model to effectively capture the relationships in the data, thus affecting its time series forecasting capabilities and reducing the accuracy of the prediction results. Therefore, an effective technical solution is urgently needed to address these issues. Summary of the Invention

[0004] In view of the above, embodiments of this specification provide a model training method. One or more embodiments of this specification also relate to a model training apparatus, a task processing method, a task processing apparatus, a network traffic data prediction method, a network traffic data prediction apparatus, a computing device, a computer-readable storage medium, and a computer program product, to address the technical deficiencies existing in the prior art.

[0005] According to a first aspect of the embodiments of this specification, a model training method is provided, comprising: Determine historical sequence data and tag sequence data, wherein the historical sequence data is data within a first time period, and the tag sequence data is data within a second time period, wherein the first time period is earlier than the second time period; The historical sequence data is input into the task processing model to obtain the predicted sequence data output by the task processing model. Determine the prediction condition distribution corresponding to the predicted sequence data and the label condition distribution corresponding to the label sequence data, wherein the prediction condition distribution is the uncertainty probability corresponding to the predicted sequence data under the premise of the historical sequence data, and the label condition distribution is the uncertainty probability corresponding to the label sequence data under the premise of the historical sequence data; The task processing model is trained based on the predicted conditional distribution and the label conditional distribution to align the predicted conditional distribution and the label conditional distribution.

[0006] According to a second aspect of the embodiments of this specification, a model training apparatus is provided, comprising: The first determining module is configured to determine historical sequence data and tag sequence data, wherein the historical sequence data is data within a first time period, and the tag sequence data is data within a second time period, wherein the first time period is earlier than the second time period; The input module is configured to input the historical sequence data into the task processing model to obtain the predicted sequence data output by the task processing model; The second determining module is configured to determine the prediction condition distribution corresponding to the predicted sequence data and the label condition distribution corresponding to the label sequence data, wherein the prediction condition distribution is the uncertainty probability corresponding to the predicted sequence data under the premise of the historical sequence data, and the label condition distribution is the uncertainty probability corresponding to the label sequence data under the premise of the historical sequence data; The training module is configured to train the task processing model based on the predicted conditional distribution and the label conditional distribution to align the predicted conditional distribution and the label conditional distribution.

[0007] According to a third aspect of the embodiments of this specification, a task processing method is provided, including: Determine the historical sequence data corresponding to the target task, wherein the historical sequence data is sequence data within a historical time period; The historical sequence data is input into the task processing model to obtain the predicted sequence data corresponding to the target task output by the task processing model. The predicted sequence data is the sequence data within the prediction time period. The task processing model is trained according to the model training method provided in the embodiments of this specification.

[0008] According to a fourth aspect of the embodiments of this specification, a task processing apparatus is provided, comprising: The determination module is configured to determine the historical sequence data corresponding to the target task, wherein the historical sequence data is sequence data within a historical time period; The input module is configured to input the historical sequence data into the task processing model to obtain the predicted sequence data corresponding to the target task output by the task processing model, wherein the predicted sequence data is the sequence data within the prediction time period, and the task processing model is trained according to the model training method provided in the embodiments of this specification.

[0009] According to a fifth aspect of the embodiments of this specification, a method for predicting network traffic data is provided, comprising: Obtain historical network traffic sequence data of the target platform within a historical time period; The historical network traffic sequence data is input into the task processing model to obtain the predicted network traffic sequence data of the target platform within the target time period output by the task processing model, wherein the task processing model is trained according to the model training method provided in the embodiments of this specification.

[0010] According to a sixth aspect of the embodiments of this specification, a network traffic data prediction apparatus is provided, comprising: The acquisition module is configured to acquire historical traffic flow sequence data of the target platform within a historical time period. The input module is configured to input the historical network traffic sequence data into the task processing model to obtain the predicted network traffic sequence data of the target platform within the target time period output by the task processing model, wherein the task processing model is trained according to the model training method provided in the embodiments of this specification.

[0011] According to a seventh aspect of the embodiments of this specification, a computing device is provided, comprising: Memory and processor; The memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions, which, when executed by the processor, implement the steps of the above method.

[0012] According to an eighth aspect of the embodiments of this specification, a computer-readable storage medium is provided that stores computer-executable instructions that, when executed by a processor, implement the steps of the above-described method.

[0013] According to a ninth aspect of the embodiments of this specification, a computer program product is provided, including a computer program or instructions that, when executed by a processor, implement the steps of the above-described method.

[0014] One embodiment of this specification provides a model training method, comprising: determining historical sequence data and label sequence data, wherein the historical sequence data is data within a first time period, and the label sequence data is data within a second time period, the first time period being earlier than the second time period; inputting the historical sequence data into a task processing model to obtain predicted sequence data output by the task processing model; determining the prediction conditional distribution corresponding to the predicted sequence data and the label conditional distribution corresponding to the label sequence data, wherein the prediction conditional distribution is the uncertainty probability corresponding to the predicted sequence data under the premise of the historical sequence data, and the label conditional distribution is the uncertainty probability corresponding to the label sequence data under the premise of the historical sequence data; and training the task processing model according to the prediction conditional distribution and the label conditional distribution to align the prediction conditional distribution and the label conditional distribution.

[0015] In the above method, when training the task processing model, after determining the historical sequence data and label sequence data, the historical sequence data can be input into the task processing model to obtain the predicted sequence data output by the task processing model. Furthermore, the prediction conditional distribution corresponding to the predicted sequence data and the label conditional distribution corresponding to the label sequence data are determined. Based on the prediction conditional distribution and the label conditional distribution, the task processing model is trained. The training objective is to align the prediction conditional distribution and the label conditional distribution, making the prediction conditional distribution output by the task processing model as consistent as possible with the label conditional distribution. By considering the overall nature of the conditional distributions, the task learning model can learn the overall information and correlations in the label conditional distribution, achieving holistic learning. This improves the time series prediction capability of the task learning model and further enhances the accuracy and precision of the time series prediction results of the task processing model. Attached Figure Description Figure 1 This is a schematic diagram illustrating an application scenario of a model training method provided in one embodiment of this specification; Figure 2 This is a flowchart illustrating a model training method provided in one embodiment of this specification; Figure 3 This is a flowchart illustrating the processing procedure of a model training method provided in one embodiment of this specification. Figure 4 This is a schematic diagram of the structure of a model training device provided in one embodiment of this specification; Figure 5 This is a flowchart illustrating a task processing method provided in one embodiment of this specification; Figure 6 This is a schematic diagram of the structure of a task processing device provided in one embodiment of this specification; Figure 7This is a flowchart illustrating a network traffic data prediction method provided in one embodiment of this specification; Figure 8 This is a schematic diagram of the structure of a network traffic data prediction device provided in one embodiment of this specification; Figure 9 This is a structural block diagram of a computing device provided in one embodiment of this specification. Detailed Implementation

[0016] Many specific details are set forth in the following description to provide a full understanding of this specification. However, this specification can be implemented in many other ways than those described herein, and those skilled in the art can make similar extensions without departing from the spirit of this specification. Therefore, this specification is not limited to the specific implementations disclosed below.

[0017] The terminology used in one or more embodiments of this specification is for the purpose of describing particular embodiments only and is not intended to be limiting of the one or more embodiments of this specification. The singular forms “a,” “described,” and “the” as used in one or more embodiments of this specification and the appended claims are also intended to include the plural forms unless the context clearly indicates otherwise. It should also be understood that the term “and / or” as used in one or more embodiments of this specification refers to and includes any or all possible combinations of one or more associated listed items.

[0018] It should be understood that although the terms first, second, etc., may be used to describe various information in one or more embodiments of this specification, such information should not be limited to these terms. These terms are only used to distinguish information of the same type from one another. For example, first may also be referred to as second without departing from the scope of one or more embodiments of this specification, and similarly, second may also be referred to as first. Depending on the context, the word "if" as used herein may be interpreted as "when," "when," or "in response to a determination."

[0019] Furthermore, 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, stored data, displayed data, etc.) involved in one or more embodiments of this specification are all information and data authorized by the user or fully authorized by all parties. Moreover, the collection, use and processing of related data must comply with the relevant laws, regulations and standards of the relevant countries and regions, and corresponding operation entry points are provided for users to choose to authorize or refuse.

[0020] First, the terms and concepts used in one or more embodiments of this specification will be explained.

[0021] Transformer: A deep learning model based on self-attention mechanism.

[0022] TCN: Temporal Convolutional Network, a model that uses one-dimensional convolution and dilated convolution to process time series data.

[0023] MLP: Multi-Layer Perceptron, a basic feedforward neural network composed of multiple fully connected layers.

[0024] Conditional KL divergence measures the information loss between two conditional distributions given an input.

[0025] Conditional JS Divergence measures the average difference between two conditional distributions given an input.

[0026] Conditional Maximum Mean Discrepancy (MMD) measures the maximum average difference between two conditional distributions in a feature space given an input.

[0027] Conditional Wasserstein divergence measures the Wasserstein difference between two conditional distributions given an input.

[0028] In practical applications, time series forecasting has wide applications in many industries, especially in energy and transportation. However, current multi-task learning methods are typically based on label-by-label matching loss functions. This approach often suffers from bias when there are correlations between labels, causing the model to fail to effectively capture the underlying structure and relationships in the data. In practice, multi-task learning methods often train models by minimizing the mean squared error, which has a fundamental flaw in time series forecasting, especially when there are correlations between labels. This point-by-point independent error estimation often fails to accurately reflect the prediction accuracy of the entire sequence.

[0029] This specification provides a model training method, and also relates to a model training device, a task processing method, a task processing device, a network traffic data prediction method, a network traffic data prediction device, a computing device, a computer-readable storage medium, and a computer program product, which will be described in detail in the following embodiments.

[0030] See Figure 1 , Figure 1 The diagram illustrates an application scenario of a model training method according to an embodiment of this specification. The model training method specifically includes the following steps.

[0031] Historical sequence data and tag sequence data are determined, wherein the historical sequence data is data within a first time period, and the tag sequence data is data within a second time period, wherein the first time period is earlier than the second time period.

[0032] The historical sequence data is input into the task processing model to obtain the predicted sequence data output by the task processing model.

[0033] Determine the prediction condition distribution corresponding to the predicted sequence data and the label condition distribution corresponding to the label sequence data, wherein the prediction condition distribution is the uncertainty probability corresponding to the predicted sequence data under the premise of the historical sequence data, and the label condition distribution is the uncertainty probability corresponding to the label sequence data under the premise of the historical sequence data.

[0034] The task processing model is trained based on the predicted conditional distribution and the label conditional distribution to align the predicted conditional distribution and the label conditional distribution.

[0035] like Figure 1 As shown, Figure 1 It includes end-side device 102 and cloud-side device 104.

[0036] In practice, users can send a model training request to the cloud-side device 104 via the edge device 102. This request may include model training requirements. Upon receiving the request, the cloud-side device 104 determines the task processing model to be trained based on these requirements, acquires historical sequence data and label sequence data, inputs the historical sequence data into the task processing model, obtains the predicted sequence data output by the model, calculates the prediction condition distribution corresponding to the predicted sequence data, and calculates the label condition distribution corresponding to the label sequence data. The task processing model is then trained based on these distributions to align them, resulting in a trained model. Afterward, the model call interface or model parameters of the task processing model can be sent to the edge device 102. If the model call interface is sent, the user can invoke the trained task processing model through it. If model parameters are sent, the task processing model can be deployed on the edge device 102 based on these parameters, allowing the user to directly invoke the model for processing via the edge device 102.

[0037] In practical applications, task processing models can be data prediction models, such as network traffic data prediction models, traffic flow data prediction models, meteorological data prediction models, or commodity sales prediction models.

[0038] The edge device 102 may include a browser, an app (application), or a web application such as an H5 (Hypertext Markup Language 5) application, a lightweight application (also known as a mini-program), or a cloud application. The edge device can be developed based on a software development kit (SDK) provided by the server, such as a real-time communication (RTC) SDK. The edge device can be deployed in an electronic device and depends on the device's operation or certain apps within the device to run. The electronic device may have a display screen and support information browsing, such as a personal mobile terminal like a mobile phone, tablet, or personal computer. Various other types of applications can also be configured in the electronic device, such as human-computer interaction applications, model training applications, text processing applications, web browser applications, shopping applications, search applications, instant messaging tools, email clients, and social media platform software.

[0039] Cloud-side device 104 can be understood as a server providing various services, including physical servers and cloud servers. Examples include servers providing communication services to multiple clients, servers supporting backend training of models used on clients, and servers processing data sent by clients. It should be noted that cloud-side device 104 can be implemented as a distributed server cluster composed of multiple servers, or as a single server. Cloud-side device 104 can also be a server for a distributed system, or a server integrated with blockchain. Cloud-side device 104 can also be a cloud server providing basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, content delivery networks (CDNs), and big data and artificial intelligence platforms, or an intelligent cloud computing server or intelligent cloud host with artificial intelligence technology.

[0040] See Figure 2 , Figure 2 A flowchart of a model training method according to an embodiment of this specification is shown, which specifically includes the following steps.

[0041] Step 202: Determine the historical sequence data and the tag sequence data.

[0042] The historical sequence data refers to data within a first time period, and the tag sequence data refers to data within a second time period, wherein the first time period is earlier than the second time period.

[0043] Historical sequence data and labeled sequence data can be understood as training data for training the task processing model. They can also be understood as time series data, which is a collection of data arranged chronologically, with each data point associated with a specific timestamp. Time series data has a time dimension, sequentiality, and can be continuous or periodic. For example, a weather station records the temperature once a day at noon; the temperature values ​​for each day over the past year can constitute a time series of daily average temperatures. Similarly, a website's backend counts online users every hour; 24 hours in a day correspond to 24 online users, and these 24 online user counts can be arranged chronologically to form a time series. The training objective of the task processing model is to enable it to predict future time series data for a predetermined time period based on the input historical time series data. For example, to enable the task processing model to predict the electricity consumption data for the following week based on the input electricity consumption data for the previous week.

[0044] Historical sequence data can be sequence data within a historical time period, while labeled sequence data can be real sequence data for a future preset time period relative to that historical time period. For example, if the historical sequence data is the daily electricity consumption data for January 2024, then the labeled sequence data could be the daily electricity consumption data for February 2024. Subsequently, the task processing model is trained based on this historical sequence data and the labeled sequence data, enabling the task processing model to learn the ability to predict the labeled sequence data based on the given sample sequence data. In other words, the daily electricity consumption data for February 2024 predicted by the task processing model based on the given daily electricity consumption data for January 2024 can be consistent with the actual daily electricity consumption data for February 2024.

[0045] Based on this, sequence data within a historical time period can be obtained as historical sequence data, and real sequence data within a future preset time period relative to that historical time period can be obtained as label sequence data, which facilitates subsequent training of the task processing model based on the historical sequence data and label sequence data.

[0046] In practical applications, historical data can be obtained in chronological order, such as the daily sales volume of a product during a historical period, the hourly temperature of a region during a historical period, and the daily traffic flow data of a road during a historical period. The obtained historical data can be cleaned, specifically by handling missing data and filtering out abnormal data, to obtain preprocessed historical data as historical sequence data and label sequence data, which is convenient for subsequent model training of task processing models.

[0047] In specific implementation, determining the historical sequence data and tag sequence data includes: Determine the sample sequence data corresponding to the sample task, wherein the sample sequence data is sequence data within a historical time period; The sample sequence data is divided into a first sequence data within a first time period and a second sequence data within a second time period, wherein the historical time period includes the first time period and the second time period, and the first time period is earlier than the second time period; The first sequence data is determined as the historical sequence data, and the second sequence data is determined as the tag sequence data.

[0048] Here, a sample task can be understood as a task processed by the task processing model. For example, to train the task processing model's ability to predict electricity consumption data, the sample task could be an electricity consumption data prediction task, and the corresponding sample sequence data could be electricity consumption data within a historical time period. Alternatively, to train the task processing model's ability to predict traffic flow data, the sample task could be a traffic flow data prediction task, and the corresponding sample sequence data could be traffic flow data within a historical time period.

[0049] Understandably, the sample task can be determined according to actual needs, and the embodiments in this specification do not limit it. For example, if an e-commerce platform wants to train a task processing model to predict product sales, then the corresponding sample task is a product sales prediction task, and the sample sequence data corresponding to this sample task can be the sales data of the product in a historical time period, which can be obtained from the e-commerce platform's database.

[0050] The first and second time periods can be understood as two time periods within a historical time period. Optionally, the first and second time periods can be consecutive or discontinuous. The first time period is earlier than the second time period; for example, if the historical time period is from January to March 2024, then the first time period could be January 2024, and the second time period could be February 2024, or even March 2024. The purpose of dividing the data into first and second time periods is to separate historical sequence data and labeled sequence data from the sample sequence data within the historical time period. Historical sequence data is considered historical relative to labeled sequence data, while labeled sequence data is considered future sequence data relative to historical sequence data. This enables the training task processing model to predict future sequence data based on historical sequence data.

[0051] Based on this, sample sequence data corresponding to the sample task and within the historical time period can be obtained, and the first sequence data within the first time period and the second sequence data within the second time period can be determined from the sample sequence data. The first sequence data is used as historical sequence data, and the second sequence data is used as label sequence data.

[0052] For example, if a power company wants to predict daily electricity consumption for the next week in order to rationally allocate power generation resources, then the sample task is an electricity consumption prediction task. The company can obtain the daily electricity consumption data corresponding to this prediction task from its database for the past year, such as the daily electricity consumption for 2024. This past year is the historical time period. To achieve the ability to predict future sequence data based on historical sequence data, the daily electricity consumption for January 2024 can be used as the first sequence data, and the daily electricity consumption for February 2024 as the second sequence data. The first sequence data can be identified as historical sequence data, and the second sequence data as labeled sequence data. Alternatively, the daily electricity consumption from January to June 2024 can be used as the first sequence data, and the daily electricity consumption from July to December 2024 as the second sequence data. This specification does not limit the division of sample sequence data in the embodiments; it can be divided according to actual needs.

[0053] In practical applications, the first sequence data within a first time period and the second sequence data within a second time period can be determined from the sample sequence data based on the timestamp corresponding to each data point. This specification does not limit this aspect in the embodiments. In summary, by acquiring the sample sequence data corresponding to the sample task, and extracting the first sequence data within the first time period and the second sequence data within the second time period from the sample sequence data, the historical sequence data and label sequence data can be determined, providing a data foundation for model training of the subsequent task processing model.

[0054] Step 204: Input the historical sequence data into the task processing model to obtain the predicted sequence data output by the task processing model.

[0055] Specifically, after obtaining historical sequence data and label sequence data for model training, the historical sequence data can be input into the task processing model to obtain the predicted sequence data output by the task processing model.

[0056] Here, the task processing model can be understood as the task processing model to be trained. The predicted sequence data can be understood as the predicted sequence data for a future preset time period, which the task processing model predicts based on the given historical sequence data. For example, if the historical sequence data is the traffic flow data of a certain road in the previous month (i.e., the historical time period), the predicted sequence data could be the traffic flow data of that road in the next week or next month (i.e., the future preset time period).

[0057] Based on this, historical sequence data can be input into the task processing model to be trained, and the task processing model can be used to predict the predicted sequence data based on the given historical sequence data.

[0058] In practical applications, task processing models can be statistical models, machine learning models, deep learning models, etc. Statistical models can include autoregressive integral moving average (ARM) models, seasonal autoregressive ARM models, exponential smoothing models, etc. ARM models are applicable to linear, stationary, or differenceably stationary time series; seasonal ARM models are applicable to seasonal time series; and exponential smoothing models are applicable to trending and / or seasonal time series. Machine learning models can include linear regression models, random forest models, etc., which can transform time series problems into supervised learning problems. Deep learning models can include recurrent neural network models, long short-term memory networks, gated recurrent units, temporal convolutional networks, attention mechanism processing models, etc. This specification does not limit these specific models.

[0059] Specifically, historical sequence data and prompt words corresponding to sample tasks can be input into the task processing model to obtain the predicted sequence data output by the task processing model. These prompt words can be used to instruct the task processing model to predict sequence data for a future preset time period based on historical sequence data.

[0060] In one or more optional embodiments of this specification, the task processing model can achieve distribution-enhanced direct prediction. That is, the task processing model can not only output predicted values, but also directly output probability distributions as prediction conditional distributions to quantify uncertainty. Furthermore, the task processing model can output predicted values ​​for multiple future time steps at once. In practical applications, direct prediction structures (such as Transformer, TCN, or MLP models) can be used, or conditional distribution modeling can be achieved based on normal distribution (e.g., model training based on negative log-likelihood loss function), quantile regression (e.g., model training based on quantile loss function), and mixture density networks outputting mixture Gaussian parameters.

[0061] Furthermore, the historical sequence data includes multiple sample values; The step of inputting the historical sequence data into the task processing model to obtain the predicted sequence data output by the task processing model includes: The historical sequence data is input into the task processing model to obtain multiple predicted values ​​corresponding to each time point output by the task processing model. The multiple predicted values ​​corresponding to each time point are determined as the predicted sequence data.

[0062] Historical sequence data can include multiple sample values, which can be understood as sequence data of different dimensions corresponding to the sample task. For example, for a meteorological data prediction task, historical sequence data could be the meteorological data of a city for the previous week. This meteorological data could include daily average temperature, daily average humidity, and whether it rained, etc. The daily average temperature, daily average humidity, and whether it rained could all be sample values ​​in the historical sequence data. Similarly, the multiple predicted values ​​corresponding to each time point can also be understood as sequence data of different dimensions corresponding to the sample task. For example, it could be the predicted daily temperature, daily humidity, and whether it rained for a city in the coming week. Each time point could be, for example, each day of the coming week. Therefore, the multiple predicted values ​​corresponding to each time point could be the predicted temperature, humidity, and whether it rained for each day of the coming week.

[0063] Therefore, the task processing model can be a multi-task processing model. This model can predict predicted sequence data in multiple dimensions based on historical sequence data in multiple dimensions as input. These multiple dimensions can be understood as multiple tasks. A shared multi-task processing model can simultaneously learn multiple related tasks, that is, simultaneously learn the ability to predict predicted data in multiple dimensions. This allows for the sharing of the underlying feature representations of the multi-task processing model, improving generalization ability, reducing model redundancy, and increasing training efficiency. In practical applications, when predicting predicted data in multiple dimensions, the multi-task processing model can share an encoder and use multiple task heads to predict data in different dimensions separately. For example, one task head can predict temperature, and another can predict humidity.

[0064] Based on this, historical sequence data containing sample values ​​of multiple dimensions can be input into the task processing model to obtain multiple predicted values ​​corresponding to each time point within a preset future time period, and the multiple predicted values ​​corresponding to each time point can be determined as the predicted sequence data.

[0065] In summary, by using a multi-task processing model to simultaneously predict values ​​across multiple dimensions, prediction efficiency is further improved. Furthermore, since there may be correlations between multiple dimensions during the prediction process, simultaneous prediction based on a shared multi-task processing model can take into account these correlations, thereby further improving the accuracy of the prediction results.

[0066] Step 206: Determine the prediction conditional distribution corresponding to the prediction sequence data and the label conditional distribution corresponding to the label sequence data.

[0067] Wherein, the prediction condition distribution is the uncertainty probability corresponding to the predicted sequence data under the premise of the historical sequence data, and the label condition distribution is the uncertainty probability corresponding to the label sequence data under the premise of the historical sequence data.

[0068] Specifically, after obtaining the label sequence data, the label conditional distribution corresponding to the label sequence data can be determined. Furthermore, after obtaining the prediction sequence data output by the task processing model, the prediction conditional distribution corresponding to the prediction sequence data can be determined.

[0069] The prediction conditional distribution can be understood as the conditional distribution corresponding to the predicted sequence data. Given the input historical sequence data, this distribution models the uncertainty of the output predicted sequence data by the task processing model. It describes the predicted value for each dimension (e.g., predicted temperature and humidity), the correlation between each dimension (e.g., high temperature with low humidity), and the uncertainty of the prediction (e.g., prediction uncertainty under extreme weather conditions). Correspondingly, the label conditional distribution models the uncertainty of the label conditional distribution given the input historical sequence data.

[0070] Based on this, it can be determined that, given the input historical sequence data, the task processing model models the uncertainty of the output predicted sequence data, and also determines the uncertainty of the label conditional distribution, given the input historical sequence data.

[0071] For example, if the historical sequence data input to the task processing model is the electricity consumption, temperature, and date of the past 24 hours, and the predicted sequence data output by the task processing model includes the electricity consumption during tomorrow's morning peak, the electricity consumption during tomorrow's evening peak, and the total electricity consumption for tomorrow, then the predicted conditional distribution corresponding to this predicted sequence data can be used to describe the uncertainty (i.e., standard deviation) of the morning peak, whether the evening peak is positively correlated with the morning peak, and the 90% confidence interval of the total electricity consumption.

[0072] In specific implementation, determining the prediction conditional distribution corresponding to the prediction sequence data and the label conditional distribution corresponding to the label sequence data includes: Given the historical sequence data, determine the first uncertainty probability corresponding to the predicted sequence data, and define the first uncertainty probability as the prediction conditional distribution corresponding to the predicted sequence data; Given the historical sequence data, determine the second uncertainty probability corresponding to the label sequence data, and define the second uncertainty probability as the label conditional distribution corresponding to the label sequence data.

[0073] The first uncertainty probability can be understood as the probability distribution of future predicted sequence data output by the task processing model, given the historical sequence data. For example, if the input historical sequence data is temperature data, the first uncertainty probability could be "tomorrow's temperature follows a normal distribution, with a 95% probability of being between 21 and 29°C". Correspondingly, the second uncertainty probability can be understood as the probability distribution of the labeled sequence data, given the historical sequence data.

[0074] Specifically, it is possible to obtain the first uncertainty probability of the predicted sequence data within a future preset time period output by the task processing model under the premise that the given input is historical sequence data, and to obtain the second uncertainty probability of the label sequence data under the premise that the given input is historical sequence data. The first uncertainty probability can be the prediction conditional distribution corresponding to the predicted sequence data, and the second uncertainty probability can be the label conditional distribution corresponding to the label sequence data.

[0075] In practical applications, the mean and standard deviation of the predicted sequence data can be calculated based on the predicted sequence data, and the prediction conditional distribution corresponding to the predicted sequence data can be determined based on the mean and standard deviation.

[0076] Furthermore, based on the predicted values ​​of each dimension included in the predicted sequence data, the conditional distribution corresponding to each dimension can be calculated, and the joint conditional distribution corresponding to the predicted sequence data can be calculated based on the conditional distributions corresponding to each dimension. This joint conditional distribution can be used as the prediction conditional distribution. The joint conditional distribution can be understood as the joint probability distribution of multiple random variables taking values ​​simultaneously under certain conditions. Therefore, the joint conditional distribution corresponding to the predicted sequence data can be understood as the joint probability distribution of the predicted values ​​of each dimension occurring simultaneously under the condition of known historical sequence data. For example, the input sample sequence data includes the product sales, advertising investment, weather information, and holiday information of an e-commerce platform over the past 30 days. The predicted sequence data output by the task processing model includes the predicted values ​​of three dimensions: tomorrow's product sales, tomorrow's website visits, and tomorrow's average user time. In addition to the predicted values ​​of each dimension, it is also necessary to understand the relationship between the predicted values ​​of multiple dimensions. For example, does high traffic usually accompany high sales, and are there situations where traffic is high but sales are low? Based on this, the joint probability distribution between the predicted values ​​of multiple dimensions can be calculated.

[0077] For example, if the predicted sequence data consists of daily product sales, daily product visits, and daily promotional information for the next week, then the mean and standard deviation of the daily product sales can be calculated. Based on these mean and standard deviation, the conditional distribution corresponding to the daily product sales can be calculated. Similarly, the mean and standard deviation of the daily product visits can be calculated. Based on these mean and standard deviation, the conditional distribution corresponding to the daily product visits can be determined. Finally, based on the conditional distributions of daily product sales and daily product visits, the joint conditional distribution (i.e., the predicted conditional distribution) corresponding to the predicted sequence data can be determined. Correspondingly, the mean and standard deviation of the tag sequence data can also be calculated based on the tag sequence data, and the tag conditional distribution corresponding to the tag sequence data can be calculated based on these mean and standard deviation. This will not be repeated in the embodiments of this specification.

[0078] Furthermore, the prediction conditional distribution and label conditional distribution can be calculated using methods such as parametric joint distribution, dependency structure modeling, autoregressive decomposition, and deep generative models. In parametric joint distribution, it can be assumed that the predicted sequence data follows a multivariate normal distribution. Based on this, the mean vector and covariance matrix output by the task processing model can be obtained, with the covariance matrix encoding the correlation between different dimensions. In dependency structure modeling, the marginal distribution of the predicted values ​​for each dimension can be modeled separately, and then the dependency structure between the marginal distributions of the predicted values ​​for each dimension can be modeled, offering high flexibility. In autoregressive decomposition, the joint probability distribution can be decomposed into conditional chains, allowing the task processing model to predict the predicted values ​​for each dimension sequentially, with the predicted value of a later dimension depending on the predicted value of the previous dimension. In deep generative models, the deep generative model can directly learn the prediction from historical sequence data to the predicted conditional distribution. The embodiments in this specification do not limit this.

[0079] It is understandable that the process of calculating the label conditional distribution is the same as the process of calculating the prediction conditional distribution. For the specific implementation of calculating the label conditional distribution, please refer to the process of calculating the prediction conditional distribution in the embodiments of this specification. This specification will not repeat the details in the embodiments.

[0080] In summary, by determining the conditional distribution, we can not only provide predicted values ​​but also quantify uncertainty, further improving the prediction accuracy of predicted sequence data. Furthermore, by modeling the joint probability distribution, we can model the positive correlations between multiple dimensions, capture the correlations between multiple tasks, improve prediction consistency, support complex queries, and identify low-probability but high-risk joint events.

[0081] Step 208: Train the task processing model based on the predicted condition distribution and the label condition distribution to align the predicted condition distribution and the label condition distribution.

[0082] Specifically, after obtaining the prediction condition distribution and the label condition distribution, the task processing model can be trained based on the prediction condition distribution and the label condition distribution to align the prediction condition distribution and thus obtain the trained task processing model.

[0083] The training objective of the task processing model can be to align the prediction conditional distribution with the label conditional distribution to achieve unbiased training.

[0084] Based on this, a loss function can be calculated according to the prediction condition distribution and the label condition distribution. The task processing model can then be trained according to the loss function to align the prediction condition distribution and the label condition distribution, thus obtaining a trained task processing model.

[0085] Further, training the task processing model based on the predicted conditional distribution and the label conditional distribution to align the predicted conditional distribution and the label conditional distribution includes: Calculate the conditional distribution divergence between the predicted conditional distribution and the label conditional distribution; The task processing model is trained based on the conditional distribution divergence to align the predicted conditional distribution with the label conditional distribution.

[0086] Among them, the conditional distribution divergence can be used to measure the degree of difference between the predicted conditional distribution and the label conditional distribution. The smaller the conditional distribution divergence, the more similar the predicted conditional distribution and the label conditional distribution are. When the conditional distribution divergence is 0, it means that the predicted conditional distribution and the label conditional distribution are exactly the same.

[0087] Specifically, the conditional distribution divergence between the predicted conditional distribution and the label conditional distribution can be calculated, and this divergence can be used as a loss function to train the task processing model so that the predicted conditional distribution and the label conditional distribution are aligned.

[0088] In practical applications, the alignment of the prediction conditional distribution and the label conditional distribution can be achieved by minimizing the conditional distribution divergence.

[0089] In one or more optional embodiments of this specification, the conditional distribution divergence can be conditional KL divergence, conditional JS divergence, conditional MMD divergence, etc. Conditional KL divergence can be used to measure, on average, how much information is lost when approximating the label conditional distribution with the predicted conditional distribution. Conditional JS divergence can be used to measure the average difference between the predicted conditional distribution and the label conditional distribution. Conditional MMD divergence can be used to measure the difference between the predicted conditional distribution and the label conditional distribution in the reproducing kernel Hilbert space. The embodiments in this specification do not limit this.

[0090] In summary, the embodiments of this specification propose that minimizing the conditional distribution divergence between the predicted conditional distribution of the task processing model and the true label conditional distribution is an unbiased training objective. This means that the training objective of the task processing model is to learn to make the overall probability distribution of the predicted sequence data it generates as consistent as possible with the true probability distribution, given historical sequence data. By using the conditional distribution divergence as the loss function for multi-task learning, an unbiased training objective is still provided when label correlation exists in multiple dimensions, thus avoiding prediction bias.

[0091] Further, calculating the conditional distribution divergence between the predicted conditional distribution and the label conditional distribution includes: Calculate the distribution distance information between the predicted conditional distribution and the label conditional distribution; Based on the distribution distance information, the conditional distribution divergence between the predicted conditional distribution and the label conditional distribution is determined.

[0092] The distribution distance information can be understood as the distance between the predicted conditional distribution and the label conditional distribution.

[0093] Based on this, the distribution distance information between the predicted conditional distribution and the label conditional distribution can be calculated, and the distribution distance information can be determined as the conditional distribution divergence between the predicted conditional distribution and the label conditional distribution.

[0094] In practical applications, conditional distribution divergence can be measured using joint Wassstein divergence. Wassstein distance, a mathematical measure of the difference between two probability distributions, originates from optimal transport theory. Its core idea is to calculate the minimum amount of work required to "transfer" one distribution into another. Pile A can represent the labeled conditional distribution, and pile B can represent the predicted conditional distribution. Suppose we need to use a bulldozer to completely transfer pile A into the shape of pile B. Then, the cost per shovel of soil equals the weight of the soil multiplied by the transport distance. The Wassstein distance represents the total cost of the transport scheme with the minimum total cost among all possible transport schemes. Determining joint Wassstein divergence based on Wassstein distance considers not only the similarity of the shapes of the two distributions but also their spatial relationship. Even if the two distributions do not overlap, a smaller distance between them will result in a smaller Wassstein distance.

[0095] In one or more optional embodiments of this specification, the joint Wasserstein divergence can be used as a loss function to estimate the difference between conditional distributions. The joint Wasserstein divergence is equivalent to the conditional distribution divergence, which can be effectively estimated based on a finite number of samples, and the joint Wasserstein divergence is differentiable, which can be used for model training.

[0096] The formula for calculating the joint Wasserstein divergence is shown below.

[0097]

[0098] in, This can be understood as the joint Wassstein divergence. This can be understood as a set of feasible transportation schemes, where D represents the pairwise distance matrix between the predicted conditional distribution and the label conditional distribution, and matrix P defines the transportation scheme. It can be used to represent the inner product of matrices. It can be a prediction conditional distribution or a label conditional distribution.

[0099] In summary, by using the joint Wasserstein divergence to estimate the difference between the prediction conditional distribution and the label conditional distribution, the bias of point-by-point error estimation is avoided, the prediction effect is improved, and the prediction results are more accurate.

[0100] In summary, the above method, when training the task processing model, after determining the historical sequence data and label sequence data, can input the historical sequence data into the task processing model to obtain the predicted sequence data output by the task processing model. Furthermore, it determines the prediction conditional distribution corresponding to the predicted sequence data and the label conditional distribution corresponding to the label sequence data. Based on the prediction conditional distribution and the label conditional distribution, the task processing model is trained. The training objective is to align the prediction conditional distribution and the label conditional distribution, making the prediction conditional distribution output by the task processing model as consistent as possible with the label conditional distribution. By considering the overall nature of the conditional distributions, the task learning model can learn the overall information and correlations in the label conditional distribution, achieving holistic learning and thus improving the time series prediction capability of the task learning model, further enhancing the accuracy and precision of the time series prediction results of the task processing model. The following is in conjunction with the appendix Figure 3 Taking the model training method provided in this specification as an example in the training of a multi-task model, the model training method will be further explained. Among other things, Figure 3 The flowchart of a model training method provided in one embodiment of this specification is shown, which specifically includes the following steps.

[0101] Step 302: Obtain historical sequence data and label sequence data.

[0102] Step 304: Input the historical sequence data into the task processing model to obtain the predicted sequence data output by the task processing model.

[0103] Step 306: Calculate the prediction conditional distribution corresponding to the prediction sequence data, and calculate the label conditional distribution corresponding to the label sequence data.

[0104] Step 308: Calculate the joint Wasserstein divergence between the predicted conditional distribution and the label conditional distribution.

[0105] Step 310: Train the task processing model based on the joint Wasserstein divergence to align the predicted conditional distribution with the label conditional distribution.

[0106] In the above method, when training the task processing model, after determining the historical sequence data and label sequence data, the historical sequence data can be input into the task processing model to obtain the predicted sequence data output by the task processing model. Furthermore, the prediction conditional distribution corresponding to the predicted sequence data and the label conditional distribution corresponding to the label sequence data are determined. Based on the prediction conditional distribution and the label conditional distribution, the task processing model is trained. The training objective is to align the prediction conditional distribution and the label conditional distribution, making the prediction conditional distribution output by the task processing model as consistent as possible with the label conditional distribution. By considering the overall nature of the conditional distributions, the task learning model can learn the overall information and correlations in the label conditional distribution, achieving holistic learning. This improves the time series prediction capability of the task learning model and further enhances the accuracy and precision of the time series prediction results of the task processing model. Corresponding to the above method embodiments, this specification also provides embodiments of a model training device. Figure 4 A schematic diagram of a model training apparatus according to one embodiment of this specification is shown. Figure 4 As shown, the device includes: The first determining module 402 is configured to determine historical sequence data and tag sequence data, wherein the historical sequence data is data within a first time period, and the tag sequence data is data within a second time period, wherein the first time period is earlier than the second time period; The input module 404 is configured to input the historical sequence data into the task processing model to obtain the predicted sequence data output by the task processing model; The second determining module 406 is configured to determine the prediction condition distribution corresponding to the prediction sequence data and the label condition distribution corresponding to the label sequence data, wherein the prediction condition distribution is the uncertainty probability corresponding to the prediction sequence data under the premise of the historical sequence data, and the label condition distribution is the uncertainty probability corresponding to the label sequence data under the premise of the historical sequence data; Training module 408 is configured to train the task processing model based on the prediction condition distribution and the label condition distribution to align the prediction condition distribution and the label condition distribution.

[0107] In an optional embodiment, the training module 408 is further configured to: Calculate the conditional distribution divergence between the predicted conditional distribution and the label conditional distribution; The task processing model is trained based on the conditional distribution divergence to align the predicted conditional distribution with the label conditional distribution.

[0108] In an optional embodiment, the training module 408 is further configured to: Calculate the distribution distance information between the predicted conditional distribution and the label conditional distribution; Based on the distribution distance information, the conditional distribution divergence between the predicted conditional distribution and the label conditional distribution is determined.

[0109] In an optional embodiment, the second determining module 406 is further configured to: Given the historical sequence data, determine the first uncertainty probability corresponding to the predicted sequence data, and define the first uncertainty probability as the prediction conditional distribution corresponding to the predicted sequence data; Given the historical sequence data, determine the second uncertainty probability corresponding to the label sequence data, and define the second uncertainty probability as the label conditional distribution corresponding to the label sequence data.

[0110] In an optional embodiment, the first determining module 402 is further configured to: Determine the sample sequence data corresponding to the sample task, wherein the sample sequence data is sequence data within a historical time period; The sample sequence data is divided into a first sequence data within a first time period and a second sequence data within a second time period, wherein the historical time period includes the first time period and the second time period, and the first time period is earlier than the second time period; The first sequence data is determined as the historical sequence data, and the second sequence data is determined as the tag sequence data.

[0111] In one optional embodiment, the historical sequence data includes multiple sample values; The input module 404 is further configured as follows: The historical sequence data is input into the task processing model to obtain multiple predicted values ​​corresponding to each time point output by the task processing model. The multiple predicted values ​​corresponding to each time point are determined as the predicted sequence data.

[0112] In the aforementioned apparatus, when training the task processing model, after determining the historical sequence data and label sequence data, the historical sequence data can be input into the task processing model to obtain the predicted sequence data output by the task processing model. Furthermore, the predicted conditional distribution corresponding to the predicted sequence data and the label conditional distribution corresponding to the label sequence data are determined. Based on the predicted conditional distribution and the label conditional distribution, the task processing model is trained. The training objective is to align the predicted conditional distribution and the label conditional distribution, making the predicted conditional distribution output by the task processing model as consistent as possible with the label conditional distribution. By considering the overall nature of the conditional distributions, the task learning model can learn the overall information and correlations in the label conditional distribution, achieving holistic learning. This improves the time series prediction capability of the task learning model and further enhances the accuracy and precision of the time series prediction results of the task processing model.

[0113] The above is an illustrative scheme of a model training device according to this embodiment. It should be noted that the technical solution of this model training device and the technical solution of the model training method described above belong to the same concept. For details not described in detail in the technical solution of the model training device, please refer to the description of the technical solution of the model training method described above.

[0114] Corresponding to the above method embodiments, this specification also provides a task processing method, see [link to documentation]. Figure 5 , Figure 5 A flowchart of a task processing method according to an embodiment of this specification is shown, and the specific steps are as follows. Step 502: Determine the historical sequence data corresponding to the target task, wherein the historical sequence data is sequence data within a historical time period; Step 504: Input the historical sequence data into the task processing model to obtain the predicted sequence data corresponding to the target task output by the task processing model.

[0115] The predicted sequence data is the sequence data within the predicted time period, and the task processing model is trained according to the model training method provided in the embodiments of this specification.

[0116] Specifically, after training the task processing model, the trained task processing model can be used to process historical sequence data to obtain target sequence data.

[0117] The above is an illustrative scheme of a task processing method according to this embodiment. It should be noted that the technical solution of this task processing method and the technical solution of the model training method described above belong to the same concept. For details not described in detail in the technical solution of the task processing method, please refer to the description of the technical solution of the model training method described above.

[0118] Corresponding to the above method embodiments, this specification also provides embodiments of a task processing device. Figure 6 A schematic diagram of a task processing apparatus according to one embodiment of this specification is shown. Figure 6 As shown, the device includes: The determining module 602 is configured to determine the historical sequence data corresponding to the target task, wherein the historical sequence data is sequence data within a historical time period; The input module 604 is configured to input the historical sequence data into the task processing model to obtain the predicted sequence data corresponding to the target task output by the task processing model, wherein the predicted sequence data is the sequence data within the prediction time period, and the task processing model is trained according to the model training method provided in the embodiments of this specification.

[0119] The above is an illustrative scheme of a task processing device according to this embodiment. It should be noted that the technical solution of this task processing device and the technical solution of the model training method described above belong to the same concept. For details not described in detail in the technical solution of the task processing device, please refer to the description of the technical solution of the model training method described above.

[0120] Corresponding to the above method embodiments, this specification also provides a network traffic data prediction method, see [link to documentation]. Figure 7 , Figure 7 A flowchart of a network traffic data prediction method according to an embodiment of this specification is shown, and the specific steps are as follows. Step 702: Obtain historical network traffic sequence data of the target platform within a historical time period; Step 704: Input the historical network traffic sequence data into the task processing model to obtain the predicted network traffic sequence data of the target platform within the target time period output by the task processing model.

[0121] The task processing model is trained according to the model training method provided in the embodiments of this specification.

[0122] Specifically, the target platform can be understood as a network platform, and the network traffic data can be the access traffic data of the target platform. The trained task processing model can be used to predict the network traffic data. The historical network traffic sequence data of the target platform within a historical time period can be input into the trained task processing model to obtain the predicted network traffic sequence data of the target platform within a future target time period.

[0123] The above is an illustrative scheme of a network traffic data prediction method according to this embodiment. It should be noted that the technical solution of this network traffic data prediction method and the technical solution of the model training method described above belong to the same concept. For details not described in detail in the technical solution of the network traffic data prediction method, please refer to the description of the technical solution of the model training method described above.

[0124] Corresponding to the above method embodiments, this specification also provides embodiments of a network traffic data prediction device. Figure 8 A schematic diagram of a network traffic data prediction device according to one embodiment of this specification is shown. Figure 8 As shown, the device includes: The acquisition module 802 is configured to acquire historical network traffic sequence data of the target platform within a historical time period; The input module 804 is configured to input the historical network traffic sequence data into the task processing model to obtain the predicted network traffic sequence data of the target platform within the target time period output by the task processing model, wherein the task processing model is trained according to the model training method provided in the embodiments of this specification.

[0125] The above is an illustrative scheme of a network traffic data prediction device according to this embodiment. It should be noted that the technical solution of this network traffic data prediction device and the technical solution of the model training method described above belong to the same concept. For details not described in detail in the technical solution of the network traffic data prediction device, please refer to the description of the technical solution of the model training method described above.

[0126] Figure 9 A structural block diagram of a computing device 900 according to one embodiment of this specification is shown. The components of the computing device 900 include, but are not limited to, a memory 910 and a processor 920. The processor 920 is connected to the memory 910 via a bus 930, and a database 950 is used to store data.

[0127] The computing device 900 also includes an access device 940, which enables the computing device 900 to communicate via one or more networks 960. Examples of these networks include Public Switched Telephone Network (PSTN), Local Area Network (LAN), Wide Area Network (WAN), Personal Area Network (PAN), or combinations of communication networks such as the Internet. The access device 940 may include one or more of any type of wired or wireless network interface (e.g., a network interface card (NIC)), such as an IEEE 802.11 Wireless Local Area Network (WLAN) wireless interface, a Wi-MAX (Worldwide Interoperability for Microwave Access) interface, an Ethernet interface, a Universal Serial Bus (USB) interface, a cellular network interface, a Bluetooth interface, or a Near Field Communication (NFC) interface.

[0128] In one embodiment of this specification, the above-described components of the computing device 900 and Figure 9 Other components, not shown, can also be connected to each other, for example, via a bus. It should be understood that... Figure 9 The block diagram of the computing device shown is for illustrative purposes only and is not intended to limit the scope of this specification. Those skilled in the art can add or replace other components as needed.

[0129] The computing device 900 can be any type of stationary or mobile computing device, including mobile computers or mobile computing devices (e.g., tablet computers, personal digital assistants, laptop computers, notebook computers, netbooks, etc.), mobile phones (e.g., smartphones), wearable computing devices (e.g., smartwatches, smart glasses, etc.) or other types of mobile devices, or stationary computing devices such as desktop computers or personal computers (PCs). The computing device 900 can also be a mobile or stationary server.

[0130] The processor 920 is configured to execute the following computer-executable instructions, which, when executed by the processor, implement the steps of the above method.

[0131] The above is an illustrative scheme of a computing device according to this embodiment. It should be noted that the technical solution of this computing device and the technical solution of the above method belong to the same concept, and all details not described in detail in the technical solution of the computing device can be referred to the description of the technical solution of the above method.

[0132] An embodiment of this specification also provides a computer-readable storage medium storing computer-executable instructions that, when executed by a processor, implement the steps of the above-described method.

[0133] The above is an illustrative scheme of a computer-readable storage medium according to this embodiment. It should be noted that the technical solution of this storage medium and the technical solution of the method described above belong to the same concept, and all details not described in detail in the technical solution of the storage medium can be referred to the description of the technical solution of the method described above.

[0134] An embodiment of this specification also provides a computer program product, including a computer program or instructions that, when executed by a processor, implement the steps of the above-described method.

[0135] The above is an illustrative scheme of a computer program product according to this embodiment. It should be noted that the technical solution of this computer program product and the technical solution of the above method belong to the same concept, and all details not described in detail in the technical solution of the computer program product can be referred to the description of the technical solution of the above method.

[0136] The foregoing has described specific embodiments of this specification. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recited in the claims may be performed in a different order than that shown in the embodiments and may still achieve the desired result. Furthermore, the processes depicted in the drawings do not necessarily require the specific or sequential order shown to achieve the desired result. In some embodiments, multitasking and parallel processing are possible or may be advantageous.

[0137] The computer instructions include computer program code, which may be in the form of source code, object code, executable file, or certain intermediate forms. The computer-readable medium may include: any entity or device capable of carrying the computer program code, recording media, USB flash drive, portable hard drive, magnetic disk, optical disk, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signals, telecommunication signals, and software distribution media, etc. It should be noted that the content included in the computer-readable medium may be appropriately added or removed according to the requirements of patent practice. For example, in some regions, according to patent practice, computer-readable media may not include electrical carrier signals and telecommunication signals.

[0138] It should be noted that, for the sake of simplicity, the foregoing method embodiments are all described as a series of actions. However, those skilled in the art should understand that the embodiments in this specification are not limited to the described order of actions, because according to the embodiments in this specification, some steps can be performed in other orders or simultaneously. Furthermore, those skilled in the art should also understand that the embodiments described in this specification are all preferred embodiments, and the actions and modules involved are not necessarily essential to the embodiments in this specification.

[0139] In the above embodiments, the descriptions of each embodiment have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0140] The preferred embodiments disclosed above are merely illustrative of this specification. Optional embodiments do not exhaustively describe all details, nor do they limit the invention to the specific implementations described. Clearly, many modifications and variations can be made based on the embodiments described in this specification. These embodiments are selected and specifically described in this specification to better explain the principles and practical applications of the embodiments, thereby enabling those skilled in the art to better understand and utilize this specification.

Claims

1. A model training method, characterized in that, include: Determine historical sequence data and tag sequence data, wherein the historical sequence data is data within a first time period, and the tag sequence data is data within a second time period, wherein the first time period is earlier than the second time period; The historical sequence data is input into the task processing model to obtain the predicted sequence data output by the task processing model. Determine the prediction condition distribution corresponding to the predicted sequence data and the label condition distribution corresponding to the label sequence data, wherein the prediction condition distribution is the uncertainty probability corresponding to the predicted sequence data under the premise of the historical sequence data, and the label condition distribution is the uncertainty probability corresponding to the label sequence data under the premise of the historical sequence data; The task processing model is trained based on the predicted conditional distribution and the label conditional distribution to align the predicted conditional distribution and the label conditional distribution.

2. The method according to claim 1, characterized in that, The step of training the task processing model based on the predicted conditional distribution and the label conditional distribution to align the predicted conditional distribution and the label conditional distribution includes: Calculate the conditional distribution divergence between the predicted conditional distribution and the label conditional distribution; The task processing model is trained based on the conditional distribution divergence to align the predicted conditional distribution with the label conditional distribution.

3. The method according to claim 1, characterized in that, Determining the prediction conditional distribution corresponding to the prediction sequence data and the label conditional distribution corresponding to the label sequence data includes: Given the historical sequence data, determine the first uncertainty probability corresponding to the predicted sequence data, and define the first uncertainty probability as the prediction conditional distribution corresponding to the predicted sequence data; Given the historical sequence data, determine the second uncertainty probability corresponding to the label sequence data, and define the second uncertainty probability as the label conditional distribution corresponding to the label sequence data.

4. The method according to any one of claims 1-3, characterized in that, The determination of historical sequence data and tag sequence data includes: Determine the sample sequence data corresponding to the sample task, wherein the sample sequence data is sequence data within a historical time period; The sample sequence data is divided into a first sequence data within a first time period and a second sequence data within a second time period, wherein the historical time period includes the first time period and the second time period, and the first time period is earlier than the second time period; The first sequence data is determined as the historical sequence data, and the second sequence data is determined as the tag sequence data.

5. The method according to any one of claims 1-3, characterized in that, The historical sequence data includes multiple sample values; The step of inputting the historical sequence data into the task processing model to obtain the predicted sequence data output by the task processing model includes: The historical sequence data is input into the task processing model to obtain multiple predicted values ​​corresponding to each time point output by the task processing model. The multiple predicted values ​​corresponding to each time point are determined as the predicted sequence data.

6. A task processing method, characterized in that, include: Determine the historical sequence data corresponding to the target task, wherein the historical sequence data is sequence data within a historical time period; The historical sequence data is input into the task processing model to obtain the predicted sequence data corresponding to the target task output by the task processing model, wherein the predicted sequence data is the sequence data within the prediction time period, and the task processing model is trained according to the method of any one of claims 1-5.

7. A method for predicting network traffic data, characterized in that, include: Obtain historical network traffic sequence data of the target platform within a historical time period; The historical network traffic sequence data is input into the task processing model to obtain the predicted network traffic sequence data of the target platform within the target time period output by the task processing model, wherein the task processing model is trained according to any one of claims 1-5.

8. A computer-readable storage medium, characterized in that, It stores computer-executable instructions that, when executed by a processor, implement the steps of the method according to any one of claims 1 to 7.

9. A computer program product, characterized in that, It includes a computer program or instructions that, when executed by a processor, implement the steps of the method according to any one of claims 1 to 7.

Images