Time sequence prediction model generation method, data processing method and device

Abstract

The invention discloses a time sequence prediction model generation method and device and a data processing method and device. The method comprises the steps that historical target variable data and a covariable data set corresponding to a preset prediction task are acquired, the historical target variable data are historical prediction time sequence data, and the covariable data set is a set of data related to prediction of the historical target variable data; target covariable information is determined from the covariable data set, and the target covariable information is information influencing prediction of historical target variable data; performing feature embedding and fusion processing on the historical target variable data and the target covariable information to obtain a fusion feature vector; and training a pre-training model according to the fusion feature vector to obtain a target time sequence prediction model corresponding to the preset prediction task. According to the method and the device, the technical problem of relatively low time sequence data prediction accuracy due to the fact that a time sequence prediction model is not combined with covariable information in related technologies is solved.

Description

Technical Field

[0001] The present application relates to the field of artificial intelligence, and specifically, to a method for generating a time series prediction model, a data processing method and a device. Background Art

[0002] With the rapid development of deep learning technology, time series forecasting models, as an important branch of deep learning technology, have been widely used in many fields such as industrial discrete manufacturing, retail sales forecasting, power load forecasting, etc. However, the existing time series forecasting models do not incorporate covariate information, resulting in low accuracy in time series data prediction.

[0003] To address the above-mentioned problems, no effective solution has been proposed yet. Summary of the invention

[0004] The embodiments of the present application provide a method for generating a time series prediction model, a data processing method and a device, so as to at least solve the technical problem in the related art that the time series prediction model does not combine covariate information and has a relatively low accuracy in time series data prediction.

[0005] According to one aspect of an embodiment of the present application, a method for generating a time series prediction model is provided, including: obtaining historical target variable data and a covariate data set corresponding to a preset prediction task, wherein the preset prediction task includes at least a time series prediction task in the power field, the historical target variable data is historical prediction time series data, and the covariate data set is a collection of data related to the prediction of the historical target variable data; determining target covariate information from the covariate data set, wherein the target covariate information is information that affects the prediction of the historical target variable data; performing feature embedding and fusion processing on the historical target variable data and the target covariate information to obtain a fused feature vector; and training a pre-trained model based on the fused feature vector to obtain a target time series prediction model corresponding to the preset prediction task.

[0006] Furthermore, the modality of the target covariate information is at least one of the following: text, number, enumeration, and determining the target covariate information from the covariate data set includes: if the modality of the target covariate information is number, respectively calculating the correlation coefficient between the time series covariate data and the historical target variable data in the covariate data set to obtain a calculation result; based on the calculation result, determining the target covariate information from the time series covariate data.

[0007] Furthermore, determining the target covariate information from the covariate data set includes: if the mode of the target covariate information is text and / or enumeration, performing a deviation analysis on the timestamp information and event information in the covariate data set based on the historical target variable data to obtain an analysis result; based on the analysis result, determining the target covariate information from the timestamp information and event information.

[0008] Furthermore, feature embedding and fusion processing are performed on the historical target variable data and the target covariate information to obtain a fused feature vector, including: feature embedding processing is performed on the historical target variable data and the target covariate information respectively through a preset embedding module to obtain a feature vector corresponding to the historical target variable data and a feature vector corresponding to the target covariate information; and fusion processing is performed on the feature vector corresponding to the historical target variable data and the feature vector corresponding to the target covariate information to obtain a fused feature vector.

[0009] Furthermore, the preset embedding module includes a first embedding module, a second embedding module, a third embedding module and a fourth embedding module. The first embedding module, the second embedding module, the third embedding module and the fourth embedding module process different types of information. If the modality of the target covariate information includes text, numerical value and enumeration, the preset embedding module performs feature embedding processing on the historical target variable data and the target covariate information respectively to obtain the feature vector corresponding to the historical target variable data and the feature vector corresponding to the target covariate information, including: extracting and embedding information on the historical target variable data through the first embedding module to obtain a first feature vector; extracting and embedding information on the timestamp information whose modality is numerical value in the target covariate information through the second embedding module to obtain a second feature vector; extracting and embedding information on the time series covariate data whose modality is numerical value in the target covariate information through the third embedding module to obtain a third feature vector, and extracting and embedding information on the event information whose modality is enumeration in the target covariate information through the third embedding module to obtain a fourth feature vector; extracting and embedding information on the event information whose modality is text in the target covariate information through the fourth embedding module to obtain a fifth feature vector.

[0010] Furthermore, the eigenvectors corresponding to the historical target variable data and the eigenvectors corresponding to the target covariate information are fused to obtain a fused eigenvector, including: fusing the first eigenvector, the second eigenvector, the third eigenvector, the fourth eigenvector and the fifth eigenvector to obtain a fused eigenvector.

[0011] Furthermore, the pre-trained model is trained based on the fused feature vector to obtain a target time series prediction model corresponding to the preset prediction task, including: inputting the fused feature vector into the pre-trained model, and adjusting the model parameters of the pre-trained model based on a preset loss function to obtain the target time series prediction model.

[0012] According to another aspect of an embodiment of the present application, a data processing method is also provided, including: obtaining historical time series data and predetermined target covariate information corresponding to a preset prediction task, wherein the preset prediction task includes at least a time series prediction task in the power field, and the target covariate information is information that affects the prediction of the target prediction time series data; fusing the historical time series data and the target covariate information through a target time series prediction model to obtain a target feature vector, wherein the target time series prediction model is obtained by using any of the above-mentioned time series prediction model generation methods; and predicting the target prediction time series data corresponding to the preset prediction task based on the target feature vector.

[0013] According to another aspect of an embodiment of the present application, a data processing method is also provided, including: obtaining historical time series data and predetermined target covariate information corresponding to a preset prediction task uploaded by a client, wherein the preset prediction task includes at least a time series prediction task in the power field, and the target covariate information is information that affects the prediction of the target prediction time series data; fusing the historical time series data and the target covariate information through a target time series prediction model in a cloud server to obtain a target feature vector, wherein the target time series prediction model is obtained by using any of the above-mentioned time series prediction model generation methods; predicting the target prediction time series data corresponding to the preset prediction task based on the target feature vector; and feeding back the target prediction time series data to the client.

[0014] According to another aspect of an embodiment of the present application, a device for generating a time series prediction model is also provided, including: a first acquisition unit, used to acquire historical target variable data and a covariate data set corresponding to a preset prediction task, wherein the preset prediction task includes at least a time series prediction task in the power field, the historical target variable data is the historical prediction time series data, and the covariate data set is a collection of data related to the prediction of the historical target variable data; a first determination unit, used to determine target covariate information from the covariate data set, wherein the target covariate information is information that affects the prediction of the historical target variable data; a first processing unit, used to perform feature embedding and fusion processing on the historical target variable data and the target covariate information to obtain a fused feature vector; a second processing unit, used to train a pre-trained model based on the fused feature vector to obtain a target time series prediction model corresponding to the preset prediction task.

[0015] Furthermore, the first determination unit includes: a calculation subunit, which is used to calculate the correlation coefficient between the time series covariate data and the historical target variable data in the covariate data set respectively to obtain a calculation result if the mode of the target covariate information is a numerical value; and a first determination subunit, which is used to determine the target covariate information from the time series covariate data based on the calculation result.

[0016] Furthermore, the first determination unit also includes: an analysis subunit, which is used to perform a deviation analysis on the timestamp information and event information in the covariate data set based on the historical target variable data to obtain an analysis result if the mode of the target covariate information is text and / or enumeration; and a second determination subunit, which is used to determine the target covariate information from the timestamp information and event information based on the analysis result.

[0017] Furthermore, the first processing unit includes: a first processing sub-unit, used to perform feature embedding processing on the historical target variable data and the target covariate information respectively through a preset embedding module to obtain the feature vector corresponding to the historical target variable data and the feature vector corresponding to the target covariate information; a second processing sub-unit, used to perform fusion processing on the feature vector corresponding to the historical target variable data and the feature vector corresponding to the target covariate information to obtain a fused feature vector.

[0018] Furthermore, the first processing subunit includes: a first processing module, which is used to perform information extraction and embedding processing on historical target variable data through a first embedding module to obtain a first feature vector; a second processing module, which is used to perform information extraction and embedding processing on timestamp information whose modality is a numerical value in the target covariate information through a second embedding module to obtain a second feature vector; a third processing module, which is used to perform information extraction and embedding processing on time series covariate data whose modality is a numerical value in the target covariate information through a third embedding module to obtain a third feature vector, and to perform information extraction and embedding processing on event information whose modality is an enumeration in the target covariate information through the third embedding module to obtain a fourth feature vector; a fourth processing module, which is used to perform information extraction and embedding processing on event information whose modality is text in the target covariate information through a fourth embedding module to obtain a fifth feature vector.

[0019] Furthermore, the second processing subunit includes: a fifth processing module, configured to perform fusion processing on the first feature vector, the second feature vector, the third feature vector, the fourth feature vector and the fifth feature vector to obtain a fused feature vector.

[0020] Furthermore, the second processing unit includes: an adjustment subunit, which is used to input the fused feature vector into the pre-trained model, and adjust the model parameters of the pre-trained model according to a preset loss function to obtain a target time series prediction model.

[0021] According to another aspect of an embodiment of the present application, a data processing device is also provided, including: a second acquisition unit, used to acquire historical time series data and predetermined target covariate information corresponding to a preset prediction task, wherein the preset prediction task at least includes a time series prediction task in the power field, and the target covariate information is information that affects the prediction of the target prediction time series data; a third processing unit, used to fuse the historical time series data and the target covariate information through a target time series prediction model to obtain a target feature vector, wherein the target time series prediction model is obtained by using any of the above-mentioned time series prediction model generation methods; a fourth processing unit, used to predict the target prediction time series data corresponding to the preset prediction task based on the target feature vector.

[0022] According to another aspect of an embodiment of the present invention, there is also provided an electronic device, comprising: a memory storing an executable program; and a processor for running the program, wherein when the program is running, any one of the above-mentioned methods for generating a time series prediction model is executed.

[0023] According to another aspect of an embodiment of the present invention, a computer-readable storage medium is provided, wherein the storage medium stores a program, wherein when the program is running, the device where the storage medium is located is controlled to execute any one of the above-mentioned methods for generating a time series prediction model.

[0024] According to another aspect of an embodiment of the present invention, a computer program product is provided, including a computer program or instructions, which implements any of the above-mentioned methods for generating a time series prediction model when executed by a processor.

[0025] In an embodiment of the present application, by obtaining historical target variable data and covariate data sets corresponding to a preset prediction task, wherein the preset prediction task includes at least a time series prediction task in the power field, the historical target variable data is the historical prediction time series data, and the covariate data set is a collection of data related to the prediction of the historical target variable data; determining target covariate information from the covariate data set, wherein the target covariate information is information that affects the prediction of the historical target variable data; performing feature embedding and fusion processing on the historical target variable data and the target covariate information to obtain a fused feature vector; training a pre-trained model based on the fused feature vector to obtain a target time series prediction corresponding to the preset prediction task; The method of testing the model effectively utilizes the covariate information. By fusing the multimodal covariate information to train the pre-trained model, the model can better understand and utilize the covariate information when dealing with complex prediction tasks involving multiple types of covariates, and make more accurate predictions, thereby improving the accuracy and robustness of the time series prediction model, improving the prediction accuracy and efficiency, and providing strong support for decision-making in related fields, achieving the purpose of achieving more accurate time series prediction, thereby achieving the technical effect of improving the prediction accuracy of the time series prediction model for time series data, and further solving the technical problem that the time series prediction model in the related technology does not combine the covariate information and has a relatively low prediction accuracy of time series data. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] The drawings described herein are used to provide a further understanding of the present application and constitute a part of the present application. The illustrative embodiments of the present application and their descriptions are used to explain the present application and do not constitute an improper limitation on the present application. In the drawings:

[0027] Figure 1 is a schematic diagram of a computer terminal provided according to Embodiment 1 of the present application;

[0028] Figure 2 is a flowchart of a method for generating a time series prediction model according to the first embodiment of the present application;

[0029] Figure 3 is a schematic diagram of an optional multimodal covariate fusion processing framework provided according to the first embodiment of the present application;

[0030] Figure 4 is a flow chart of a data processing method provided according to Embodiment 2 of the present application;

[0031] Figure 5 is a flow chart of a data processing method provided according to Embodiment 3 of the present application;

[0032] Figure 6 is a schematic diagram of a device for generating a time series prediction model according to Embodiment 4 of the present application;

[0033] Figure 7 is a schematic diagram of a data processing device provided according to Embodiment 5 of the present application;

[0034] Figure 8 This is a structural block diagram of an electronic device provided according to Embodiment 6 of the present application. DETAILED DESCRIPTION

[0035] In order to enable those skilled in the art to better understand the solution of the present application, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of the present application.

[0036] It should be noted that the terms "first", "second", etc. in the specification and claims of the present application and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchangeable where appropriate, so that the embodiments of the present application described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any of their variations are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device comprising a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0037] 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 this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of relevant data must comply with relevant laws, regulations and standards in the relevant regions, and provide corresponding operation entrances for users to choose to authorize or refuse.

[0038] First, some nouns or terms that appear in the description of the embodiments of the present application are subject to the following explanations:

[0039] Embedding: Convert high-dimensional sparse vectors into dense vectors to facilitate downstream model processing. For example, time series data, images, audio, etc. are all processed by the model after embedding.

[0040] Description information (prompt): The prompt word of the large model is used to make it easier for the large model to understand the intention.

[0041] Example 1

[0042] According to an embodiment of the present application, a method for generating a time series prediction model is also provided. It should be noted that the steps shown in the flowchart of the accompanying drawings can be executed in a computer system such as a set of computer executable instructions, and although the logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0043] The method embodiment provided in the first embodiment of the present application can be executed in a mobile terminal, a computer terminal or a similar computing device. Figure 1 The hardware structure block diagram of a computer terminal (or mobile device) for implementing a method for generating a time series prediction model is shown. Figure 1 As shown, the computer terminal (or mobile device) 10 may include a processor set 102 (the processor set 102 may include but is not limited to a processing device such as a microprocessor (Microcontroller Unit, referred to as MCU) or a programmable logic device (Field Programmable Gate Array, referred to as FPGA), and the processor set 102 may include a processor set, Figure 1 102a, 102b, ..., 102n are used to illustrate), a memory 104 for storing data, and a transmission device 106 for communication functions. In addition, it may also include: a display, an input / output interface (I / O interface), a Universal Serial Bus (USB) port (which may be included as one of the ports of the BUS bus), a network interface, a power supply and / or a camera. Those skilled in the art can understand that Figure 1 The structure shown is only for illustration and does not limit the structure of the above electronic device. Figure 1 More or fewer components as shown, or with Figure 1 Different configurations are shown.

[0044] It should be noted that the one or more processors 102 and / or other data processing circuits described above may generally be referred to herein as "data processing circuits". The data processing circuits may be embodied in whole or in part as software, hardware, firmware, or any other combination thereof. In addition, the data processing circuits may be a single independent processing module, or may be incorporated in whole or in part into any of the other components in the computer terminal 10 (or mobile device).

[0045] The memory 104 can be used to store software programs and modules of application software, such as the program instructions / data storage device corresponding to the method for generating a time series prediction model in the embodiment of the present application. The processor 102 executes various functional applications and data processing by running the software programs and modules stored in the memory 104, that is, the method for generating a time series prediction model described above is realized. The memory 104 may include a high-speed random access memory, and may also include a non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory. In some instances, the memory 104 may further include a memory remotely arranged relative to the processor 102, and these remote memories may be connected to the computer terminal 10 via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0046] The transmission device 106 is used to receive or send data via a network. The specific example of the above network may include a wireless network provided by a communication provider of the computer terminal 10. In one example, the transmission device 106 includes a network adapter (Network Interface Controller, referred to as NIC), which can be connected to other network devices through a base station so as to communicate with the Internet. In one example, the transmission device 106 can be a radio frequency (RF) module, which is used to communicate with the Internet wirelessly.

[0047] The display may be a touch screen type liquid crystal display, which may enable a user to interact with a user interface of the computer terminal 10 (or mobile device).

[0048] Accurately predicting the future state of time series variables is crucial in many fields such as industrial discrete manufacturing, retail sales forecasting, and power load forecasting. Although traditional forecasting methods, such as statistical analysis and classical machine learning algorithms, can provide a certain degree of forecasting capabilities based on historical data, their forecasting accuracy is often limited by the inherent limitations of the model and insufficient understanding of complex time series patterns.

[0049] With the rapid development of deep learning technology, deep learning models based on self-attention mechanism have quickly become a research hotspot in the field of time series prediction due to their powerful sequence processing capabilities and self-attention mechanism, which can effectively capture the long-term and short-term dependencies in time series data. However, despite the wide application of this model, it still has some problems: high training cost and dependence on a large number of training samples, especially when training a randomly initialized model from scratch, it is limited to data training in a single field, resulting in insufficient generalization ability of the model when processing cross-domain time series data; it ignores the covariate information that is highly correlated with the target variable, resulting in relatively low accuracy in predicting time series data.

[0050] In the above technical background, the present application provides Figure 2 The method for generating the time series prediction model shown. Figure 2 1 is a flow chart of a method for generating a time series prediction model according to the first embodiment of the present application. The method includes:

[0051] Step S201, obtain historical target variable data and covariate data sets corresponding to preset prediction tasks, wherein the preset prediction tasks at least include time series prediction tasks in the power field, the historical target variable data are historical prediction time series data, and the covariate data set is a collection of data related to the prediction of the historical target variable data.

[0052] Optionally, the preset prediction task can be a time series prediction task in multiple fields such as manufacturing, retail, electricity, and transportation, such as solid waste furnace temperature prediction, power load prediction, etc. For example, after determining the vertical scenario (taking solid waste furnace temperature prediction as an example), the historical target variable data and covariate data set corresponding to the solid waste furnace temperature prediction task are obtained through the generation system of the time series prediction model, wherein the historical target variable data is the historical predicted furnace temperature, and the covariate data set includes oxygen concentration, ambient temperature, time, and other data related to the historical predicted furnace temperature.

[0053] Step S202 , determining target covariate information from the covariate data set, wherein the target covariate information is information that has an impact on the prediction of historical target variable data.

[0054] In order to enable the time series large model to have a better fine-tuning effect in the vertical field, multimodal time series covariate mining is performed, that is, the target covariate information is determined from the covariate data set. The mode of the target covariate information is at least one of the following: text type, numerical type, enumeration type. The target covariate information is information that affects the prediction of the target time series variable (such as furnace temperature), including the following types:

[0055] Time series covariates: time series variables with the same frequency as the target time series variable, for example, oxygen concentration (numeric value) and weather (numeric value) in solid waste furnace temperature prediction;

[0056] Timestamp information: time information (text) with the same frequency as the target time series variable, or calendar information (numeric value) where the timestamp is located;

[0057] Event information: discrete variables, for example, a failure occurred at a certain time (text), holiday information (text), whether it is a holiday (enumeration), whether it is winter (enumeration).

[0058] Step S203, feature embedding and fusion processing are performed on the historical target variable data and the target covariate information to obtain a fused feature vector.

[0059] In order to enable the time series large model to have a better fine-tuning effect in the vertical field, multimodal covariate fusion is performed, that is, feature embedding and fusion processing is performed on the historical target variable data and the target covariate information to obtain a fused feature vector. For example, the feature embedding processing is performed on the historical predicted furnace temperature and its corresponding target covariate information (such as oxygen concentration, etc.) through the preset embedding module, and the feature vector corresponding to the historical predicted furnace temperature and the feature vector corresponding to the target covariate information are obtained. The feature vector corresponding to the historical predicted furnace temperature and the feature vector corresponding to the target covariate information are fused to obtain a fused feature vector.

[0060] Step S204, training the pre-trained model based on the fused feature vector to obtain a target time series prediction model corresponding to the preset prediction task.

[0061] Optionally, the pre-trained model is fine-tuned based on the fused feature vector to obtain a target time series prediction model corresponding to a preset prediction task. For example, the target time series prediction model is a time series prediction model for solid waste furnace temperature prediction.

[0062] Optionally, the pre-trained model can be a large language model or a time series pre-trained large model. For example, the time series data of multiple fields such as manufacturing, retail, electricity, and transportation are used to pre-train the deep learning model based on the self-attention mechanism to obtain a time series pre-trained large model. The pre-trained model can be determined by the following method: the training set is processed with a large language model and a time series pre-trained large model respectively, and the model with the smallest error between the model output result and the true label of the training set is selected as the pre-trained model.

[0063] In this scheme, the pre-trained model is trained by fusing multimodal covariate information, so that the model can better understand and utilize these covariate information when dealing with complex prediction tasks involving multiple types of covariates, make more accurate predictions, improve the accuracy and robustness of the time series prediction model, improve the prediction accuracy and efficiency, provide strong support for decision-making in related fields, and achieve the purpose of more accurate time series prediction, thereby achieving the technical effect of improving the prediction accuracy of the time series prediction model for time series data, and further solving the technical problem that the time series prediction model in the related technology does not combine covariate information and has relatively low prediction accuracy of time series data.

[0064] In order to mine suitable covariate information, in the method for generating a time series prediction model provided in Example 1 of the present application, the modality of the target covariate information is at least one of the following: text, number, enumeration, and determining the target covariate information from the covariate data set includes: if the modality of the target covariate information is a number, then respectively calculating the correlation coefficient between the time series covariate data and the historical target variable data in the covariate data set to obtain a calculation result; based on the calculation result, determining the target covariate information from the time series covariate data.

[0065] In order to mine suitable covariate information, in the method for generating a time series prediction model provided in Example 1 of the present application, determining target covariate information from a covariate data set includes: if the modality of the target covariate information is text and / or enumeration, performing a deviation analysis on the timestamp information and event information in the covariate data set based on the historical target variable data to obtain an analysis result; based on the analysis result, determining the target covariate information from the timestamp information and event information.

[0066] Blindly combining potential covariates may reduce the prediction effect. Therefore, for specific vertical scenarios, how to mine appropriate covariate information will become the key. Optionally, the mining of target covariate information is divided into correlation mining and deviation mining. Among them, correlation mining is to judge the validity of covariates based on the correlation coefficient between covariates and target variables. It is suitable for information mining of time series covariates, which are numerical modes. For example, in the prediction of solid waste furnace temperature, the correlation coefficients between all process variables (i.e., time series covariate data in the covariate data set) and furnace temperature (i.e., historical target variable data) are calculated respectively to obtain multiple correlation coefficients (i.e., calculation results). Then, the target covariate information can be determined from the time series covariate data based on the calculation results. For example, multiple correlation coefficients are inverted indexed and the Top K covariates are selected as target covariate information, where K is a positive integer. Among them, the calculation of the correlation coefficient can adopt the widely used Pearson correlation coefficient, Spearman rank correlation coefficient or Kendall rank correlation coefficient, etc., which are not limited here.

[0067] Deviation mining is to mine the difference in the distribution of target variable values ​​when the covariate takes different values. It is suitable for information mining with text and / or enumeration as the modality. First, data preprocessing such as timestamp alignment and event information encoding can be performed. Timestamp alignment ensures that the timestamp information of all sequence data is consistent, which is convenient for subsequent analysis. Event information (such as holidays, abnormal alarms, etc.) is converted into numerical or category encoding so that the model can understand and process it. For timestamp information, time windows can be defined based on the time scale of the prediction task, such as hours, days, weeks, or months, and features can be extracted from timestamps, such as whether it is a weekday or weekend, summer or winter, etc. Then, historical target variable data can be grouped according to time features, and the groups can be compared to see if there are significant differences in the distribution of target variables under different feature values. For event information, event information can be classified into different categories, such as "holidays", "abnormal alarms", etc., and the time windows before and after the event can be defined, such as a few hours or days before and after the event. The target variable data before and after the event are used for comparative analysis to see if there are significant differences. Deviation analysis can be implemented through a variety of statistical test methods, such as variance test, chi-square test or non-parametric test. For example, the variance test method is used to compare whether the mean values ​​of the target variable under different time characteristics or events are significantly different, and the p-value (i.e., the mean value) is calculated. If the p-value is less than the preset significance level (such as 0.05), it is considered that the distribution of the target variable under this condition is significantly different, thereby identifying those timestamp information and event information that statistically significantly affect the target variable. For example, in the power load forecasting task, the variance test method is used to analyze the power load series in summer and the power load series in winter, and it is found that the electricity consumption in summer is significantly higher than that in winter. Therefore, whether it is summer or winter can be used as the target covariate information.

[0068] Through correlation coefficient calculation and deviation analysis, appropriate covariate information can be selected and redundant information can be filtered out.

[0069] In order to effectively combine the covariate information, in the method for generating a time series prediction model provided in Example 1 of the present application, feature embedding and fusion processing are performed on the historical target variable data and the target covariate information to obtain a fused feature vector, including: feature embedding processing is performed on the historical target variable data and the target covariate information respectively through a preset embedding module to obtain a feature vector corresponding to the historical target variable data and a feature vector corresponding to the target covariate information; and the feature vector corresponding to the historical target variable data and the feature vector corresponding to the target covariate information are fused to obtain a fused feature vector.

[0070] After determining the target covariate information, how to effectively combine the covariate information is also crucial. Optionally, by using a preset embedding module to perform feature embedding processing on the historical target variable data and the target covariate information respectively, the feature vector corresponding to the historical target variable data and the feature vector corresponding to the target covariate information can be obtained, and then the feature vector corresponding to the historical target variable data and the feature vector corresponding to the target covariate information are fused to obtain a fused feature vector. Among them, the preset embedding module includes at least a positional embedding module (positional embedding), a temporal embedding module (temporal embedding), a covariate embedding module (covariate embedding) and a description information embedding module (prompt embedding). Positional embedding belongs to the basic embedding, which is used to process the target variable and can be plugged in and out with other embeddings. Temporal embedding is used to extract information such as the year, month, day, and hour of the timestamp to form a time series variable of the same length, and then the obtained time series variable is mapped to a vector of the same dimension as the time series embedding of the target variable with the help of sine, cosine trigonometric functions or learnable embedding or fully connected layers, and then point addition or point multiplication operations can be performed. Covariate embedding is used to process time series covariates and enumeration type event information to obtain time series variables, which are then connected with the target variable and sent to the model for training and reasoning. Prompt embedding is used to process text type event information, such as holidays, abnormal alarms, etc., which are non-continuous events. Prompt words and text embedding can be used to perform dot addition or dot multiplication operations with the target variable.

[0071] The feature vectors corresponding to the historical target variable data and the feature vectors corresponding to the target covariate information are fused to construct a fused feature vector that can fully capture various information dimensions in the time series prediction task, and provide a richer and more comprehensive feature representation for subsequent model training and prediction. Feature fusion can adopt widely used technologies such as early fusion, late fusion, or deep fusion. Early fusion directly connects all feature vectors in a certain order to form a long vector as a fused feature vector. Late fusion processes each feature vector separately, and then connects or averages the processed results as a fused feature vector. Deep fusion fuses feature vectors through neural network layers (such as fully connected layers or attention mechanism layers), allowing the model to learn how to combine features from different information sources.

[0072] By performing feature embedding processing on the historical target variable data and target covariate information respectively and fusing them to obtain a fused feature vector, the effective combination of covariate information is achieved, providing a data basis for model fine-tuning training.

[0073] In order to effectively combine the covariate information, in the method for generating a time series prediction model provided in the first embodiment of the present application, the preset embedding module includes a first embedding module, a second embedding module, a third embedding module and a fourth embedding module. The first embedding module, the second embedding module, the third embedding module and the fourth embedding module process different types of information. If the modality of the target covariate information includes text, numerical value and enumeration, the historical target variable data and the target covariate information are respectively subjected to feature embedding processing by the preset embedding module to obtain the feature vector corresponding to the historical target variable data and the feature vector corresponding to the target covariate information, including: the historical target variable data is processed by the first embedding module. The variable data is processed by information extraction and embedding to obtain a first eigenvector; the timestamp information whose modality is numerical value in the target covariate information is processed by a second embedding module to obtain a second eigenvector; the time series covariate data whose modality is numerical value in the target covariate information is processed by a third embedding module to obtain a third eigenvector, and the event information whose modality is enumeration in the target covariate information is processed by the third embedding module to obtain a fourth eigenvector; the event information whose modality is text in the target covariate information is processed by the fourth embedding module to obtain a fifth eigenvector.

[0074] In order to effectively combine the covariate information, in the method for generating a time series prediction model provided in Example 1 of the present application, the feature vector corresponding to the historical target variable data and the feature vector corresponding to the target covariate information are fused to obtain a fused feature vector, including: fusing the first feature vector, the second feature vector, the third feature vector, the fourth feature vector and the fifth feature vector to obtain a fused feature vector.

[0075] Optionally, the first embedding module is the aforementioned position embedding module, the second embedding module is the aforementioned time embedding module, the third embedding module is the aforementioned covariate embedding module, and the fourth embedding module is the aforementioned description information embedding module. For example, by extracting and embedding the position information of the historical target variable data through positional embedding, a feature vector (i.e., a first feature vector) corresponding to the historical target variable data can be obtained; by extracting and embedding the timestamp information of the target covariate information whose modality is a numerical value through temporal embedding, a second feature vector can be obtained; by extracting and embedding the time series covariate data of the target covariate information whose modality is a numerical value through covariate embedding, a third feature vector can be obtained; by extracting and embedding the event information of the target covariate information whose modality is an enumeration through covariate embedding, a fourth feature vector can be obtained; by extracting and embedding the event information of the target covariate information whose modality is a text through prompttembedding, a fifth feature vector can be obtained.

[0076] Optionally, the eigenvector corresponding to the target covariate information includes a second eigenvector, a third eigenvector, a fourth eigenvector, and a fifth eigenvector. The first eigenvector, the second eigenvector, the third eigenvector, the fourth eigenvector, and the fifth eigenvector are fused to obtain a fused eigenvector. The fusion process can be performed according to actual needs using operations such as splicing fusion (splicing in a specific order), element-level fusion (such as point addition or point multiplication), attention mechanism fusion (weighted fusion based on weights between different eigenvectors), and neural network fusion (by designing a specific neural network structure, such as a fully connected layer, a convolutional layer, or a recurrent neural network layer, to fuse eigenvectors) to form a comprehensive fused eigenvector.

[0077] By fusing the first eigenvector, the second eigenvector, the third eigenvector, the fourth eigenvector and the fifth eigenvector, the effective combination of covariate information is achieved, providing a data basis for model fine-tuning training.

[0078] In order to improve the prediction accuracy of the model, in the method for generating a time series prediction model provided in Example 1 of the present application, a pre-trained model is trained based on a fused feature vector to obtain a target time series prediction model corresponding to a preset prediction task, including: inputting the fused feature vector into the pre-trained model, and adjusting the model parameters of the pre-trained model based on a preset loss function to obtain a target time series prediction model.

[0079] Optionally, after obtaining the fused feature vector, the target time series prediction model can be obtained by inputting the fused feature vector into a pre-trained model and adjusting the model parameters of the pre-trained model according to a preset loss function. For example, by inputting the fused feature vector into a large language model or a time series pre-trained large model for fine-tuning training, a time series prediction model for solid waste furnace temperature prediction can be obtained.

[0080] By fusing multimodal covariate information to train the pre-trained model, the model can better understand and utilize the covariate information when dealing with complex prediction tasks involving multiple types of covariates, make more accurate predictions, and improve the accuracy and robustness of the time series prediction model, thereby improving the prediction accuracy and efficiency.

[0081] In an optional embodiment, it is possible to use Figure 3 The schematic diagram shown realizes multimodal covariate fusion processing. Figure 3 is a schematic diagram of an optional multimodal covariate fusion processing framework provided according to the first embodiment of the present application, such as Figure 3 As shown in , the information type processed by embedding can be multimodal, for example, time information such as timestamps and seasons, and text information such as holidays, events, and weather. The embedding modules include location embedding module, time embedding module, covariate embedding module, and description information embedding module, which are used to process different information types respectively. The embedding level is divided into column dimension and row dimension, where the column dimension means that each column of the table data has its own encoding, and the row dimension means that the table data has a unified encoding. Figure 3 As shown, examples of application scenarios corresponding to row dimensions and column dimensions and types of covariates in the scenarios are given.

[0082] In an embodiment of the present application, by obtaining historical target variable data and covariate data sets corresponding to a preset prediction task, wherein the preset prediction task includes at least a time series prediction task in the power field, the historical target variable data is the historical prediction time series data, and the covariate data set is a collection of data related to the prediction of the historical target variable data; determining target covariate information from the covariate data set, wherein the target covariate information is information that affects the prediction of the historical target variable data; performing feature embedding and fusion processing on the historical target variable data and the target covariate information to obtain a fused feature vector; training a pre-trained model based on the fused feature vector to obtain a target time series prediction corresponding to the preset prediction task; The method of testing the model effectively utilizes the covariate information. By fusing the multimodal covariate information to train the pre-trained model, the model can better understand and utilize the covariate information when dealing with complex prediction tasks involving multiple types of covariates, and make more accurate predictions, thereby improving the accuracy and robustness of the time series prediction model, improving the prediction accuracy and efficiency, and providing strong support for decision-making in related fields, achieving the purpose of achieving more accurate time series prediction, thereby achieving the technical effect of improving the prediction accuracy of the time series prediction model for time series data, and further solving the technical problem that the time series prediction model in the related technology does not combine the covariate information and has a relatively low prediction accuracy of time series data.

[0083] It should be noted that, for the aforementioned method embodiments, for the sake of simplicity, they are all expressed as a series of action combinations, but those skilled in the art should be aware that the present application is not limited by the described order of actions, because according to the present application, certain steps can be performed in other orders or simultaneously. Secondly, those skilled in the art should also be aware that the embodiments described in the specification are all preferred embodiments, and the actions and modules involved are not necessarily required by the present application.

[0084] Through the description of the above implementation methods, those skilled in the art can clearly understand that the method according to the above embodiment can be implemented by means of software plus a necessary general hardware platform, and of course it can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of the present application, or the part that contributes to the prior art, can be embodied in the form of a software product, which is stored in a storage medium (such as ROM / RAM, disk, CD), and includes a number of instructions for a terminal device (which can be a mobile phone, computer, server, or network device, etc.) to execute the methods of each embodiment of the present application.

[0085] Example 2

[0086] According to an embodiment of the present application, a data processing method is also provided, such as Figure 4 As shown, the method includes:

[0087] Step S401, obtaining historical time series data corresponding to a preset prediction task and predetermined target covariate information, wherein the preset prediction task at least includes a time series prediction task in the electric power field, and the target covariate information is information that affects the prediction of the target prediction time series data;

[0088] Step S402, fusing the historical time series data and the target covariate information through a target time series prediction model to obtain a target feature vector, wherein the target time series prediction model is obtained by using any of the above-mentioned time series prediction model generation methods;

[0089] Step S403, obtaining target prediction time series data corresponding to the preset prediction task according to the target feature vector prediction.

[0090] Optionally, after obtaining the target time series prediction model, corresponding data processing can be performed. For example, the preset prediction task is solid waste furnace temperature prediction, and a time series prediction model for solid waste furnace temperature prediction is obtained by using any of the above-mentioned time series prediction model generation methods. The solid waste furnace temperature in the past period of time (i.e., historical time series data) and the target covariate information (such as oxygen concentration, weather, etc.) mined during the model training process are input into the time series prediction model. The model can be fused to obtain the target feature vector, and then the solid waste furnace temperature in the future period of time (i.e., the target prediction time series data) is predicted based on the target feature vector.

[0091] By fusing multimodal covariate information to train the pre-trained model, the model can better understand and utilize the covariate information when dealing with complex prediction tasks involving multiple types of covariates, make more accurate predictions, improve the accuracy and robustness of the time series prediction model, and enhance the prediction accuracy and efficiency, providing strong support for decision-making in related fields, achieving the goal of more accurate time series prediction, thereby achieving the technical effect of improving the prediction accuracy of the time series prediction model for time series data, and further solving the technical problem in related technologies that the time series prediction model does not combine covariate information and has relatively low prediction accuracy for time series data.

[0092] Example 3

[0093] According to an embodiment of the present application, a data processing method is also provided, such as Figure 5 As shown, the method includes:

[0094] Step S501, obtaining historical time series data corresponding to a preset prediction task uploaded by a client and predetermined target covariate information, wherein the preset prediction task at least includes a time series prediction task in the electric power field, and the target covariate information is information that affects the prediction of the target prediction time series data;

[0095] Step S502: In the cloud server, the historical time series data and the target covariate information are fused and processed by the target time series prediction model to obtain a target feature vector, wherein the target time series prediction model is obtained by using any of the above-mentioned time series prediction model generation methods; and the target prediction time series data corresponding to the preset prediction task is obtained according to the target feature vector prediction;

[0096] Step S503: Feedback the target prediction time series data to the client.

[0097] In the cloud server, the specific method for data processing is the same as that in the second embodiment, and will not be repeated here.

[0098] It should be noted that, for the aforementioned method embodiments, for the sake of simplicity, they are all expressed as a series of action combinations, but those skilled in the art should be aware that the present application is not limited by the described order of actions, because according to the present application, certain steps can be performed in other orders or simultaneously. Secondly, those skilled in the art should also be aware that the embodiments described in the specification are all preferred embodiments, and the actions and modules involved are not necessarily required by the present application.

[0099] Through the description of the above implementation methods, those skilled in the art can clearly understand that the method according to the above embodiment can be implemented by means of software plus a necessary general hardware platform, and of course it can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of the present application, or the part that contributes to the prior art, can be embodied in the form of a software product, which is stored in a storage medium (such as ROM / RAM, disk, CD), and includes a number of instructions for a terminal device (which can be a mobile phone, computer, server, or network device, etc.) to execute the methods of each embodiment of the present application.

[0100] Example 4

[0101] According to an embodiment of the present application, a device for generating a time series prediction model for implementing the above-mentioned method for generating a time series prediction model is also provided. Figure 6 As shown, the device includes: a first acquisition unit 601, a first determination unit 602, a first processing unit 603, and a second processing unit 604.

[0102] A first acquisition unit 601 is used to acquire historical target variable data and a covariate data set corresponding to a preset prediction task, wherein the preset prediction task at least includes a time series prediction task in the power field, the historical target variable data is the historical prediction time series data, and the covariate data set is a collection of data related to the prediction of the historical target variable data;

[0103] A first determining unit 602 is used to determine target covariate information from the covariate data set, wherein the target covariate information is information that has an impact on the prediction of historical target variable data;

[0104] The first processing unit 603 is used to perform feature embedding and fusion processing on the historical target variable data and the target covariate information to obtain a fused feature vector;

[0105] The second processing unit 604 is used to train the pre-trained model according to the fused feature vector to obtain a target time series prediction model corresponding to a preset prediction task.

[0106] In the device for generating a time series prediction model provided in Example 4 of the present application, historical target variable data and a covariate data set corresponding to a preset prediction task are obtained through a first acquisition unit 601, wherein the preset prediction task includes at least a time series prediction task in the power field, the historical target variable data is the historical prediction time series data, and the covariate data set is a collection of data related to the prediction of the historical target variable data; the first determination unit 602 determines the target covariate information from the covariate data set, wherein the target covariate information is information that affects the prediction of the historical target variable data; the first processing unit 603 performs feature embedding and fusion processing on the historical target variable data and the target covariate information to obtain a fused feature vector; the second processing unit 604 trains the pre-trained model based on the fused feature vector to obtain a target time series prediction model corresponding to the preset prediction task. In this scheme, the pre-trained model is trained by fusing multimodal covariate information, so that the model can better understand and utilize these covariate information when dealing with complex prediction tasks involving multiple types of covariates, make more accurate predictions, improve the accuracy and robustness of the time series prediction model, improve the prediction accuracy and efficiency, provide strong support for decision-making in related fields, and achieve the purpose of more accurate time series prediction, thereby achieving the technical effect of improving the prediction accuracy of the time series prediction model for time series data, and further solving the technical problem that the time series prediction model in the related technology does not combine covariate information and has relatively low prediction accuracy of time series data.

[0107] Optionally, in the generation device of the time series prediction model provided in Example 4 of the present application, the first determination unit 602 includes: a calculation subunit, which is used to calculate the correlation coefficient between the time series covariate data in the covariate data set and the historical target variable data if the mode of the target covariate information is a numerical value, and obtain a calculation result; a first determination subunit, which is used to determine the target covariate information from the time series covariate data based on the calculation result.

[0108] Optionally, in the generation device of the time series prediction model provided in Example 4 of the present application, the first determination unit 602 also includes: an analysis subunit, which is used to perform a deviation analysis on the timestamp information and event information in the covariate data set based on the historical target variable data to obtain an analysis result if the mode of the target covariate information is text and / or enumeration; and a second determination subunit, which is used to determine the target covariate information from the timestamp information and event information based on the analysis result.

[0109] Optionally, in the generation device of the time series prediction model provided in Example 4 of the present application, the first processing unit 603 includes: a first processing sub-unit, used to perform feature embedding processing on the historical target variable data and the target covariate information respectively through a preset embedding module to obtain the feature vector corresponding to the historical target variable data and the feature vector corresponding to the target covariate information; a second processing sub-unit, used to perform fusion processing on the feature vector corresponding to the historical target variable data and the feature vector corresponding to the target covariate information to obtain a fused feature vector.

[0110] Optionally, in the generation device of the time series prediction model provided in Example 4 of the present application, the first processing sub-unit includes: a first processing module, which is used to perform information extraction and embedding processing on historical target variable data through a first embedding module to obtain a first feature vector; a second processing module, which is used to perform information extraction and embedding processing on timestamp information whose modality is a numerical value in the target covariate information through a second embedding module to obtain a second feature vector; a third processing module, which is used to perform information extraction and embedding processing on time series covariate data whose modality is a numerical value in the target covariate information through a third embedding module to obtain a third feature vector, and to perform information extraction and embedding processing on event information whose modality is an enumeration in the target covariate information through the third embedding module to obtain a fourth feature vector; a fourth processing module, which is used to perform information extraction and embedding processing on event information whose modality is text in the target covariate information through a fourth embedding module to obtain a fifth feature vector.

[0111] Optionally, in the generation device of the time series prediction model provided in Example 4 of the present application, the second processing sub-unit includes: a fifth processing module, which is used to fuse the first eigenvector, the second eigenvector, the third eigenvector, the fourth eigenvector and the fifth eigenvector to obtain a fused eigenvector.

[0112] Optionally, in the device for generating a time series prediction model provided in Example 4 of the present application, the second processing unit 604 includes: an adjustment subunit, used to input the fused feature vector into a pre-trained model, and adjust the model parameters of the pre-trained model according to a preset loss function to obtain a target time series prediction model.

[0113] It should be noted that the first acquisition unit 601, the first determination unit 602, the first processing unit 603, and the second processing unit 604 correspond to steps S201 to S204 in Example 1, and the examples and application scenarios implemented by the above units and the corresponding steps are the same, but are not limited to the contents disclosed in the above Example 1. It should be noted that the above modules as part of the device can be run in the computer terminal 10 provided in Example 1.

[0114] It should be noted that the preferred implementation scheme involved in the above embodiments of the present application is the same as the scheme provided in Example 1, as well as the application scenario and implementation process, but is not limited to the scheme provided in Example 1.

[0115] Example 5

[0116] According to an embodiment of the present application, a data processing device for implementing the above data processing method is also provided. Figure 7 As shown, the device includes: a second acquisition unit 701, a third processing unit 702, and a fourth processing unit 703.

[0117] The second acquisition unit 701 is used to acquire historical time series data corresponding to a preset prediction task and predetermined target covariate information, wherein the preset prediction task at least includes a time series prediction task in the power field, and the target covariate information is information that affects the prediction of the target prediction time series data;

[0118] The third processing unit 702 is used to fuse the historical time series data and the target covariate information through a target time series prediction model to obtain a target feature vector, wherein the target time series prediction model is obtained by using any of the above-mentioned time series prediction model generation methods;

[0119] The fourth processing unit 703 is used to obtain target prediction time series data corresponding to a preset prediction task according to the target feature vector prediction.

[0120] In the data processing device provided in Example 5 of the present application, the historical time series data and predetermined target covariate information corresponding to the preset prediction task are acquired through the second acquisition unit 701, wherein the preset prediction task at least includes the time series prediction task in the power field, and the target covariate information is the information that affects the prediction of the target prediction time series data; the third processing unit 702 fuses the historical time series data and the target covariate information through the target time series prediction model to obtain the target feature vector, wherein the target time series prediction model is obtained by using any of the above-mentioned time series prediction model generation methods; the fourth processing unit 703 predicts the target prediction time series data corresponding to the preset prediction task based on the target feature vector.

[0121] It should be noted here that the above-mentioned second acquisition unit 701, third processing unit 702, and fourth processing unit 703 correspond to steps S401 to S403 in Example 2, and the instances and application scenarios implemented by the above-mentioned units and corresponding steps are the same, but are not limited to the contents disclosed in the above-mentioned Example 2.

[0122] It should be noted that the preferred implementation scheme involved in the above embodiments of the present application is the same as the scheme provided in Example 2 as well as the application scenario and implementation process, but is not limited to the scheme provided in Example 2.

[0123] Example 6

[0124] The embodiment of the present application may provide an electronic device, which may be any electronic device in a group of electronic devices. Optionally, in this embodiment, the electronic device may also be replaced by a terminal device such as a mobile terminal.

[0125] Optionally, in this embodiment, the electronic device may be located in at least one network device among a plurality of network devices of a computer network.

[0126] In this embodiment, the above-mentioned electronic device can execute the program code of the following steps in the method for generating a time series prediction model: obtaining historical target variable data and covariate data sets corresponding to a preset prediction task, wherein the preset prediction task at least includes a time series prediction task in the power field, the historical target variable data is the historical prediction time series data, and the covariate data set is a collection of data related to the prediction of the historical target variable data; determining the target covariate information from the covariate data set, wherein the target covariate information is information that affects the prediction of the historical target variable data; performing feature embedding and fusion processing on the historical target variable data and the target covariate information to obtain a fused feature vector; training the pre-trained model based on the fused feature vector to obtain a target time series prediction model corresponding to the preset prediction task.

[0127] The above-mentioned electronic device can also execute the program code of the following steps in the method for generating a time series prediction model: if the mode of the target covariate information is a numerical value, then respectively calculate the correlation coefficient between the time series covariate data in the covariate data set and the historical target variable data to obtain a calculation result; based on the calculation result, determine the target covariate information from the time series covariate data.

[0128] The above-mentioned electronic device can also execute the program code of the following steps in the method for generating a time series prediction model: if the mode of the target covariate information is text and / or enumeration, then based on the historical target variable data, a deviation analysis is performed on the timestamp information and event information in the covariate data set to obtain an analysis result; based on the analysis result, the target covariate information is determined from the timestamp information and event information.

[0129] The above-mentioned electronic device can also execute the program code of the following steps in the method for generating a time series prediction model: performing feature embedding processing on the historical target variable data and the target covariate information respectively through a preset embedding module to obtain the feature vector corresponding to the historical target variable data and the feature vector corresponding to the target covariate information; performing fusion processing on the feature vector corresponding to the historical target variable data and the feature vector corresponding to the target covariate information to obtain a fused feature vector.

[0130] The above-mentioned electronic device can also execute the program code of the following steps in the method for generating a time series prediction model: extracting and embedding historical target variable data through the first embedding module to obtain a first feature vector; extracting and embedding timestamp information with a numerical mode in the target covariate information through the second embedding module to obtain a second feature vector; extracting and embedding time series covariate data with a numerical mode in the target covariate information through the third embedding module to obtain a third feature vector, and extracting and embedding event information with an enumeration mode in the target covariate information through the third embedding module to obtain a fourth feature vector; extracting and embedding event information with a text mode in the target covariate information through the fourth embedding module to obtain a fifth feature vector.

[0131] The electronic device may also execute the program code of the following steps in the method for generating a time series prediction model: fusing the first feature vector, the second feature vector, the third feature vector, the fourth feature vector and the fifth feature vector to obtain a fused feature vector.

[0132] The electronic device can also execute the program code of the following steps in the method for generating a time series prediction model: inputting the fused feature vector into a pre-trained model, and adjusting the model parameters of the pre-trained model according to a preset loss function to obtain a target time series prediction model.

[0133] Optionally, Figure 8 is a structural block diagram of an electronic device according to an embodiment of the present application. Figure 8 As shown, the electronic device 80 may include: one or more ( Figure 8 Only one is shown in the figure) processor 802, memory 804. The electronic device 80 may also include a storage controller, through which the memory 804 is controlled and managed; the electronic device 80 may also include a peripheral interface, through which the radio frequency module, audio module and display screen are connected.

[0134] Among them, the memory can be used to store software programs and modules, such as the program instructions / modules corresponding to the method and device for generating the time series prediction model in the embodiment of the present application. The processor executes various functional applications and data processing by running the software programs and modules stored in the memory, that is, the method for generating the time series prediction model described above is realized. The memory may include a high-speed random access memory, and may also include a non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory. In some instances, the memory may further include a memory remotely arranged relative to the processor, and these remote memories may be connected to the terminal 10 via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0135] The processor can call the information and application programs stored in the memory through the transmission device to execute the following steps: obtain the historical target variable data and covariate data set corresponding to the preset prediction task, wherein the preset prediction task at least includes the time series prediction task in the power field, the historical target variable data is the historical prediction time series data, and the covariate data set is a collection of data related to the prediction of the historical target variable data; determine the target covariate information from the covariate data set, wherein the target covariate information is the information that affects the prediction of the historical target variable data; perform feature embedding and fusion processing on the historical target variable data and the target covariate information to obtain a fused feature vector; train the pre-trained model based on the fused feature vector to obtain the target time series prediction model corresponding to the preset prediction task.

[0136] Optionally, the processor may also execute the following program code: if the mode of the target covariate information is numerical, then respectively calculating the correlation coefficients between the time series covariate data in the covariate data set and the historical target variable data to obtain calculation results; based on the calculation results, determining the target covariate information from the time series covariate data.

[0137] Optionally, the processor may also execute the program code of the following steps: if the mode of the target covariate information is text and / or enumeration, a deviation analysis is performed on the timestamp information and event information in the covariate data set based on the historical target variable data to obtain an analysis result; based on the analysis result, the target covariate information is determined from the timestamp information and event information.

[0138] Optionally, the processor may also execute the program code of the following steps: performing feature embedding processing on the historical target variable data and the target covariate information respectively through a preset embedding module to obtain feature vectors corresponding to the historical target variable data and feature vectors corresponding to the target covariate information; performing fusion processing on the feature vectors corresponding to the historical target variable data and the feature vectors corresponding to the target covariate information to obtain fused feature vectors.

[0139] Optionally, the processor may also execute the program code of the following steps: extracting and embedding historical target variable data through a first embedding module to obtain a first feature vector; extracting and embedding timestamp information whose modality is numerical in target covariate information through a second embedding module to obtain a second feature vector; extracting and embedding time series covariate data whose modality is numerical in target covariate information through a third embedding module to obtain a third feature vector, and extracting and embedding event information whose modality is enumeration in target covariate information through the third embedding module to obtain a fourth feature vector; extracting and embedding event information whose modality is text in target covariate information through a fourth embedding module to obtain a fifth feature vector.

[0140] Optionally, the processor may also execute a program code of the following steps: fusing the first feature vector, the second feature vector, the third feature vector, the fourth feature vector and the fifth feature vector to obtain a fused feature vector.

[0141] Optionally, the processor may also execute the following program code: input the fused feature vector into a pre-trained model, and adjust the model parameters of the pre-trained model according to a preset loss function to obtain a target time series prediction model.

[0142] It can be understood by those skilled in the art that Figure 8 The structure shown is for illustration only, and the electronic device may also be a smart phone (such as an Android phone, an iOS phone, etc.), a tablet computer, a PDA, a mobile Internet device (Mobile Internet Devices, MID), a PAD, and other terminal devices. Figure 8 The structure of the electronic device is not limited. Figure 8 More or fewer components (such as network interfaces, display devices, etc.) shown in, or with Figure 8 Different configurations are shown.

[0143] A person of ordinary skill in the art can understand that all or part of the steps in the various methods of the above embodiments can be completed by instructing the hardware related to the terminal device through a program, and the program can be stored in a computer-readable storage medium, and the storage medium may include: a flash drive, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, etc.

[0144] Example 7

[0145] The embodiment of the present application also provides a computer-readable storage medium. Optionally, in this embodiment, the storage medium can be used to store the program code executed by the method for generating a time series prediction model provided in the first embodiment.

[0146] Optionally, in this embodiment, the above storage medium may be located in any electronic device in a group of electronic devices in a computer network, or in any mobile terminal in a group of mobile terminals.

[0147] Example 8

[0148] The embodiment of the present application further provides a computer program product. Optionally, in this embodiment, the computer program product may include a computer program, and when the computer program is executed by a processor, the method for generating a time series prediction model provided in the first embodiment above is implemented.

[0149] The serial numbers of the above-mentioned embodiments of the present application are for description only and do not represent the advantages or disadvantages of the embodiments.

[0150] In the above embodiments of the present application, the description of each embodiment has its own emphasis. For parts that are not described in detail in a certain embodiment, please refer to the relevant description of other embodiments.

[0151] In the several embodiments provided in this application, it should be understood that the disclosed technical content can be implemented in other ways. Among them, the device embodiments described above are only schematic, for example, the division of units is only a logical function division, and there may be other division methods in actual implementation, for example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of units or modules, which can be electrical or other forms.

[0152] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0153] In addition, each functional unit in each embodiment of the present application may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.

[0154] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application, or the part that contributes to the prior art, or all or part of the technical solution can be embodied in the form of a software product. The computer software product is stored in a storage medium, including a number of instructions to enable a computer device (which can be a personal computer, server or network device, etc.) to execute all or part of the steps of the various embodiments of the present application. The aforementioned storage medium includes: U disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), mobile hard disk, disk or optical disk and other media that can store program codes.

[0155] The above are only preferred implementations of the present application. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present application. These improvements and modifications should also be regarded as the scope of protection of the present application.

Claims

1. A method for generating a time series prediction model, characterized in that: include: Acquire historical target variable data and covariate data sets corresponding to a preset prediction task, wherein the preset prediction task at least includes a time series prediction task in the power field, the historical target variable data is historical prediction time series data, and the covariate data set is a collection of data related to the prediction of the historical target variable data; Determining target covariate information from the covariate data set, wherein the target covariate information is information that has an impact on the prediction of the historical target variable data; Performing feature embedding and fusion processing on the historical target variable data and the target covariate information to obtain a fused feature vector; The pre-trained model is trained according to the fused feature vector to obtain a target time series prediction model corresponding to the preset prediction task.

2. The method according to claim 1, characterized in that: The modality of the target covariate information is at least one of the following: text, numerical value, and enumeration. Determining the target covariate information from the covariate data set includes: If the mode of the target covariate information is the numerical value, then respectively calculating the correlation coefficient between the time series covariate data in the covariate data set and the historical target variable data to obtain a calculation result; Based on the calculation result, the target covariate information is determined from the time series covariate data.

3. The method according to claim 2, characterized in that Determining target covariate information from the covariate data set includes: If the modality of the target covariate information is the text and / or the enumeration, performing a deviation analysis on the timestamp information and event information in the covariate data set based on the historical target variable data to obtain an analysis result; Based on the analysis result, the target covariate information is determined from the timestamp information and the event information.

4. The method according to claim 1, characterized in that: The historical target variable data and the target covariate information are subjected to feature embedding and fusion processing to obtain a fused feature vector including: Perform feature embedding processing on the historical target variable data and the target covariate information respectively through a preset embedding module to obtain a feature vector corresponding to the historical target variable data and a feature vector corresponding to the target covariate information; The feature vector corresponding to the historical target variable data and the feature vector corresponding to the target covariate information are fused to obtain the fused feature vector.

5. The method according to claim 4, characterized in that The preset embedding module includes a first embedding module, a second embedding module, a third embedding module and a fourth embedding module. The first embedding module, the second embedding module, the third embedding module and the fourth embedding module process different types of information. If the modality of the target covariate information includes text, numerical value and enumeration, the historical target variable data and the target covariate information are respectively subjected to feature embedding processing by the preset embedding module, and the feature vectors corresponding to the historical target variable data and the feature vectors corresponding to the target covariate information are obtained, including: Performing information extraction and embedding processing on the historical target variable data through the first embedding module to obtain a first feature vector; Performing information extraction and embedding processing on the timestamp information whose modality is a numerical value in the target covariate information by the second embedding module to obtain a second feature vector; Performing information extraction and embedding processing on the time series covariate data whose modality is numerical value in the target covariate information through the third embedding module to obtain a third eigenvector, and performing information extraction and embedding processing on the event information whose modality is enumeration in the target covariate information through the third embedding module to obtain a fourth eigenvector; The fourth embedding module extracts and embeds the event information whose modality is text in the target covariate information to obtain a fifth feature vector.

6. The method according to claim 5, characterized in that The eigenvector corresponding to the historical target variable data and the eigenvector corresponding to the target covariate information are fused to obtain the fused eigenvector, including: The first feature vector, the second feature vector, the third feature vector, the fourth feature vector and the fifth feature vector are fused to obtain the fused feature vector.

7. The method according to claim 1, characterized in that The pre-trained model is trained according to the fused feature vector to obtain the target time series prediction model corresponding to the preset prediction task, including: The fused feature vector is input into the pre-trained model, and the model parameters of the pre-trained model are adjusted according to a preset loss function to obtain the target time series prediction model.

8. A data processing method, characterized in that: include: Acquire historical time series data corresponding to a preset prediction task and predetermined target covariate information, wherein the preset prediction task at least includes a time series prediction task in the electric power field, and the target covariate information is information that affects the prediction of the target prediction time series data; The historical time series data and the target covariate information are fused through a target time series prediction model to obtain a target feature vector, wherein the target time series prediction model is obtained by using the method for generating a time series prediction model according to any one of claims 1 to 7; The target prediction time series data corresponding to the preset prediction task is obtained according to the target feature vector prediction.

9. A data processing method, characterized in that: include: Acquire historical time series data and predetermined target covariate information corresponding to a preset prediction task uploaded by a client, wherein the preset prediction task at least includes a time series prediction task in the electric power field, and the target covariate information is information that affects the prediction of the target prediction time series data; In the cloud server, the historical time series data and the target covariate information are fused and processed by a target time series prediction model to obtain a target feature vector, wherein the target time series prediction model is obtained by using the time series prediction model generation method described in any one of claims 1 to 7; and the target prediction time series data corresponding to the preset prediction task is predicted based on the target feature vector; The target prediction time series data is fed back to the client.

10. A device for generating a time series prediction model, characterized in that: include: A first acquisition unit is used to acquire historical target variable data and a covariate data set corresponding to a preset prediction task, wherein the preset prediction task at least includes a time series prediction task in the power field, the historical target variable data is historical prediction time series data, and the covariate data set is a collection of data related to the prediction of the historical target variable data; A first determining unit, configured to determine target covariate information from the covariate data set, wherein the target covariate information is information that has an impact on the prediction of the historical target variable data; A first processing unit is used to perform feature embedding and fusion processing on the historical target variable data and the target covariate information to obtain a fused feature vector; The second processing unit is used to train the pre-trained model according to the fused feature vector to obtain a target time series prediction model corresponding to the preset prediction task.

11. A data processing device, characterized in that: include: A second acquisition unit is used to acquire historical time series data corresponding to a preset prediction task and predetermined target covariate information, wherein the preset prediction task at least includes a time series prediction task in the electric power field, and the target covariate information is information that affects the prediction of the target prediction time series data; a third processing unit, configured to fuse the historical time series data and the target covariate information through a target time series prediction model to obtain a target feature vector, wherein the target time series prediction model is obtained by using the method for generating a time series prediction model according to any one of claims 1 to 7; The fourth processing unit is used to predict the target prediction time series data corresponding to the preset prediction task according to the target feature vector.

12. An electronic device, characterized in that: include: A memory storing an executable program; A processor is used to run the program, wherein the program, when running, executes the method for generating a time series prediction model according to any one of claims 1 to 7.

13. A computer program product, characterized in that It comprises a computer program or instructions, which, when executed by a processor, implements the method for generating a time series prediction model according to any one of claims 1 to 7.

Images