Residual cycle forecasting method and system for enhancing time series forecasting, and device and medium

Abstract

A residual cycle forecasting method and system for enhancing time series forecasting, and a device and a medium, which aim to solve the problem of an existing technique having low efficiency and limited forecasting accuracy in a long-term time series forecasting task. The method explicitly models periodic patterns in time series data and uses a forecasting model to forecast a periodic residual, so as to improve the forecasting accuracy and efficiency of a time series forecasting model. The method comprises: determining the cycle length of a data set; generating a learnable recurrent cycle; separating a periodic component and a residual component of a time series; performing forecasting on the basis of a residual; and finally, in view of a forecast value of the residual component and the corresponding periodic component, forming a final forecasting result. The method can effectively capture the inherent periodicity in a time series, thereby reducing forecasting errors, and significantly improving the accuracy of long-term series forecasting; moreover, the method can maintain the simplicity and efficiency of a model, and is particularly applicable to application scenarios with clear periodic characteristics, such as power consumption, traffic flow and meteorological data.

Description

Residual cycle forecasting method, system, device and medium for enhancing time series forecasting TECHNICAL FIELD

[0001] The present application belongs to the technical field of machine learning, and particularly relates to a residual cycle forecasting method, system, device and medium for enhancing time series forecasting. BACKGROUND

[0002] Time series forecasting plays a crucial role in many application fields, such as power demand forecasting, weather forecasting, financial market analysis, and intelligent transportation systems. Accurate time series forecasting not only improves the efficiency of the system, but also provides strong support for decision-making. In the task of time series forecasting, especially long-term time series forecasting, the key to accurate prediction lies in whether the model can accurately capture the long-term dependence and intrinsic periodicity in the data.

[0003] In the prior art, in order to solve the problem of long-term time series forecasting, many models focus on improving the ability to capture long-range dependencies. For example, Transformer-based models (such as Informer, Autoformer) take advantage of the Transformer's ability to model long-range dependencies and have been widely applied to LTSF tasks. Convolutional neural networks (such as ModernTCN) enhance the model's ability to capture long-range dependencies through large convolutional kernel design, while recurrent neural networks (such as SegRNN) improve the processing of long sequences through segmented iteration. However, most of these models extract periodic features in time series by building deep and complex structures.

[0004] This means that existing methods usually require higher computational resource requirements and longer training times, and still have performance bottlenecks when faced with long-term time series forecasting tasks. Therefore, how to accurately model the periodicity in time series while maintaining efficiency has become a major challenge in the field of time series forecasting.

[0005] Based on this, the present application proposes a new time series forecasting method, namely the residual cycle forecasting method (Residual Cycle Forecasting, RCF). This method explicitly models the periodic patterns in time series data and predicts the periodic residual based on this, thereby significantly improving the accuracy and efficiency of the model in long-term time series forecasting tasks. SUMMARY

[0006] The main objective of this invention is to overcome the shortcomings and deficiencies of the prior art and provide a method, system, device and medium for enhancing the prediction of residual periods in time series forecasting. By explicitly modeling the periodic patterns in time series data and predicting periodic residuals based on them, the prediction accuracy is improved and the computational cost of the model is significantly reduced.

[0007] To achieve the above objectives, the present invention adopts the following technical solution:

[0008] In a first aspect, the present invention provides a residual periodicity prediction method for enhancing time series prediction, comprising the following steps:

[0009] Determine the period length W of the dataset; the dataset includes time series data;

[0010] For time series data with D variables, generate learnable cyclic periods. By cyclically copying the cycle period Q, a periodic component sequence C of the same length as the time series X is obtained;

[0011] Extracting sample x from time series X t-L+1:t Where t is the time index of the sample, and L is the length of the historical lookback window of the sample; the periodic component c corresponding to the sample is extracted from the periodic component sequence C. t-L+1:t ;

[0012] By using the original input sequence x t-L+1:t Subtract the periodic component c t-L+1:t Obtain the residual component x , t-L+1:t ;

[0013] The residual component x , t-L+1:t The data is input into the backbone prediction model for residual prediction to obtain the predicted values ​​of the residual components. Where H is the prediction window length of the sample; the periodic component c corresponding to the prediction window of the sample is extracted from the periodic component sequence C. t+1:t+H ;

[0014] The predicted values ​​of the residual components With the corresponding periodic component c t+1:t+H Add them together to get the final time series prediction result.

[0015] As a preferred technical solution, the period length W of the dataset is determined by manually inferring the characteristics of the time series dataset or by calculating the autocorrelation within the dataset.

[0016] As a preferred technical solution, the cyclic period Q is trained together with the backbone prediction model through gradient backpropagation during the training process, thereby learning a feature representation that reflects the inherent periodic pattern of the time series.

[0017] As a preferred technical solution, the backbone prediction model is a core model used for residual prediction.

[0018] As a preferred technical solution, the backbone prediction model is a single-layer linear model or a two-layer MLP model.

[0019] As a preferred technical solution, the periodic component c t-L+1:t The generation includes the following steps:

[0020] Shifting the cycle period Q to the left by t mod W positions yields Q. (t) ;

[0021] Q (t) copy Next, and splice together A periodic component c of length L is obtained. t-L+1:t .

[0022] As a preferred technical solution, the periodic component c t+1:t+H The generation includes the following steps:

[0023] Shifting the cycle period Q to the left by (t+L) mod W positions yields Q. (t+L) ;

[0024] Q (t+L) copy Next, and splice together A periodic component c of length H is obtained. t+1:t+H .

[0025] Secondly, the present invention provides a residual period prediction system for enhancing time series prediction, which is applied to the residual period prediction method for enhancing time series prediction, including a period determination module, a cycle period generation module, a period component acquisition module, a residual component calculation module, a residual component prediction module, and a time series prediction module.

[0026] The period determination module is used to determine the period length W of the dataset; the dataset includes time series data.

[0027] The cycle generation module is used to generate learnable cycles for time series data with D variables. By cyclically copying the cycle period Q, a periodic component sequence C of the same length as the time series X is obtained;

[0028] The periodic component acquisition module is used to extract sample x from the time series X. t-L+1:t Where t is the time index of the sample, and L is the length of the historical lookback window of the sample; the periodic component c corresponding to the sample is extracted from the periodic component sequence C. t-L+1:t ;

[0029] The residual component calculation module is used to calculate the original input sequence x. t-L+1:t Subtract the periodic component c t-L+1:t Obtain the residual component x , t-L+1:t ;

[0030] The residual component prediction module is used to predict the residual component x. , t-L+1:t The data is input into the backbone prediction model for residual prediction to obtain the predicted values ​​of the residual components. Where H is the prediction window length of the sample; the periodic component c corresponding to the prediction window of the sample is extracted from the periodic component sequence C. t+1:t+H ;

[0031] The time series prediction module is used to generate the predicted values ​​of the residual components. With the corresponding periodic component c t+1:t+H Add them together to get the final time series prediction result.

[0032] Thirdly, the present invention provides an electronic device, the electronic device comprising:

[0033] At least one processor; and,

[0034] A memory communicatively connected to the at least one processor; wherein,

[0035] The memory stores computer program instructions that can be executed by the at least one processor to enable the at least one processor to perform the residual periodic forecasting method for enhanced time series forecasting.

[0036] Fourthly, the present invention provides a computer-readable storage medium storing a program that, when executed by a processor, implements the aforementioned method for predicting residual periods in enhanced time series forecasting.

[0037] Compared with the prior art, the present invention has the following advantages and beneficial effects:

[0038] (1) Effectiveness: The present invention can capture the periodic characteristics of data more accurately, thereby improving the accuracy of long-term series prediction tasks.

[0039] (2) Efficiency: Compared with traditional complex deep models, the present application can achieve high-precision prediction by combining a simple backbone model for residual prediction, significantly reducing the computational cost of the model and improving the prediction efficiency.

[0040] (3) Universality: The method proposed in the present application can be combined with various existing time series prediction models to further improve the prediction accuracy of existing models.

[0041] (4) Explainability: The present application explicitly models the periodic patterns in time series data by constructing a learnable cycle period, and the fully trained cycle period can accurately reflect the periodic characteristics existing in the data, providing a clear explanation basis for the prediction results.

[0042] In summary, the present application provides an efficient and accurate solution for long time series prediction tasks of time series through the residual cycle prediction method, which has a wide application prospect. BRIEF DESCRIPTION OF DRAWINGS

[0043] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings needed in the embodiment description will be briefly introduced. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor.

[0044] FIG. 1 is a flowchart of the residual cycle prediction method for enhancing time series prediction in an embodiment of the present application;

[0045] FIG. 2 is a schematic diagram of using Linear and MLP as backbone prediction models in an embodiment of the present application;

[0046] FIG. 3 is a schematic diagram of the cycle component generation process in an embodiment of the present application;

[0047] FIG. 4 is a comparison chart of prediction accuracy and parameter efficiency in a user power consumption prediction task in an embodiment of the present application;

[0048] FIG. 5 is a comparison chart of prediction accuracy improvement of a user power consumption prediction task on existing prediction models in an embodiment of the present application;

[0049] FIG. 6 is a cycle period pattern obtained from the modeling results in a user power consumption prediction task in an embodiment of the present application;

[0050] FIG. 7 is a structural schematic diagram of a residual cycle prediction system for enhancing time series prediction in an embodiment of the present application;

[0051] FIG. 8 is a structural schematic diagram of an electronic device in an embodiment of the present application. DETAILED DESCRIPTION

[0052] In the following, the technical solutions in the embodiments of the present application will be described clearly and completely in conjunction with the drawings in the embodiments of the present application, so that those skilled in the art can better understand the solutions in the present application. Obviously, the described embodiments are only some of the embodiments of the present application, but not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative work fall within the scope of protection of the present application.

[0053] In the present application, the phrase "embodiments" means that the specific features, structures or characteristics described in conjunction with the embodiments can be included in at least one embodiment of the present application. The appearance of this phrase at various places in the specification does not necessarily mean that it refers to the same embodiment, nor is it independent or alternative to other embodiments. Those skilled in the art explicitly and implicitly understand that the embodiments described in the present application can be combined with other embodiments.

[0054] As shown in FIG. 1, the present embodiment provides a residual cycle prediction method for enhancing time series prediction, which explicitly models the periodic pattern in time series data and predicts the periodic residual on this basis to improve prediction accuracy and significantly reduce the computational overhead of the model. The method specifically includes the following steps:

[0055] (1) Determine the cycle length W of the data set;

[0056] Specifically, the cycle length W can be obtained by manually reasoning the characteristics of the time series data set or calculating the autocorrelation within the data set.

[0057] (2) For time series data with D variables, generate a learnable cyclic period Q By cyclically copying the cyclic period Q, a cycle component sequence C equal in length to the time series X can be obtained.

[0058] It can be understood that the cyclic period Q will be trained together with the backbone prediction model through gradient backpropagation during the training process, so as to learn the feature representation reflecting the inherent periodic pattern of the time series.

[0059] (3) Extract sample x from time series X t-L+1:t , where t is the time index of the sample, and L is the length of the historical lookback window of the sample; extract the cycle component c corresponding to the sample from the cycle component sequence C t-L+1:t .

[0060] Specifically, the generation of the cycle component c t-L+1:t includes the following steps:

[0061] (31) Q is obtained by shifting the cycle period Q to the left by t mod W positions. (t) ;

[0062] (32) Then Q (t) copy Next, and splice together A periodic component c of length L is obtained. t-L+1:t ;

[0063] (4) By using the original input sequence x t-L+1:t Subtract the periodic component c t-L+1:t Obtain the residual component x , t-L+1:t ;

[0064] (5) The residual component x , t-L+1:t The data is input into the backbone prediction model for residual prediction to obtain the predicted values ​​of the residual components. Where H is the prediction window length of the sample, and the backbone prediction model is the core model used for residual prediction, including but not limited to basic neural network models and existing time series prediction models. Simultaneously, the periodic component c corresponding to the prediction window of the sample is extracted from the periodic component sequence C. t+1:t+H .

[0065] Specifically, the periodic component c t+1:t+H The generation includes the following steps:

[0066] (51) Q is obtained by shifting the cycle period Q to the left by (t+L) mod W positions. (t+L) ;

[0067] (52) Q (t+L) copy Next, and splice together A periodic component c of length H is obtained. t+1:t+H .

[0068] (6) The predicted values ​​of the residual components With the corresponding periodic component c t+1:t+H Add them together to get the final time series prediction result.

[0069] In another specific embodiment of this application, the technical solution of this patent is used to further illustrate the electricity consumption situation of power industry users.

[0070] Accurate prediction of user's power consumption curve is essential for power system planning. Therefore, there is an urgent need for a more accurate user power consumption prediction technology to assist the long-term scheduling and planning of the power system. The power consumption data of 321 users in a certain area is recorded every hour, and the historical data needs to be modeled to predict the user's power consumption curve in the next month (720 data points) based on the user's historical power consumption curve in the next 4 days (96 data points). The present application is used to model the user power consumption prediction task:

[0071] Firstly, the historical training data is standardized. Since the power consumption data patterns and value ranges of different users are different, the variables (user power consumption) need to be standardized to ensure data consistency. The specific method is to calculate the average value and standard deviation of each variable, and then divide each variable by the average value and standard deviation of each variable to achieve standardization.

[0072] Further, the period length W of the data is determined. Power data usually has a daily cycle (different patterns during the day and night) and a weekly cycle (different patterns on weekdays and weekends). Combined with the recording interval of the data, it can be inferred that the period length of the power consumption curve is one week, i.e. W = 168.

[0073] Further, for user power consumption data with D = 321 variables, a learnable cycle Q is generated By copying the cycle Q cyclically, a period component sequence C equal in length to the time series X can be obtained. The cycle Q will be trained together with the main prediction model through gradient backpropagation during the training process, so as to learn the feature representation reflecting the inherent periodicity of the time series.

[0074] Further, in each specific prediction process, first, a sample x t-L+1:t is cut from the time series X, where t is the time index of the sample, and L = 96 is the length of the historical lookback window of the sample; At the same time, the period component c t-L+1:t corresponding to the sample is cut from the period component sequence C. (t) Specifically, the cycle Q can be obtained by shifting Q to the left by t mod W positions. (t) Then Q t-L+1:t is copied times and spliced to obtain a period component c t-L+1:t with a length of L. t-L+1:t .

[0075] Further, by subtracting the period component c t-L+1:t from the original input sequence x t-L+1:t , a residual component x , is obtained.t-L+1:t .

[0076] Further, the residual component x , t-L+1:t is input into the trunk prediction model for residual prediction, obtaining the predicted value of the residual component where H=720 is the length of the prediction window of the sample. The trunk prediction model is a backbone model for residual prediction, including but not limited to a basic neural network model and an existing time series prediction model. In the present embodiment, a single-layer linear model and a double-layer MLP model are adopted as the trunk prediction model, and two specific prediction models, CycleNet / Linear and CycleNet / MLP, are constructed.

[0077] Further, the periodic component c t+1:t+H corresponding to the prediction window of the sample is cut from the periodic component sequence C (t+L) . (t+L) , is copied (t+L) times, and spliced to obtain the periodic component c with a length of H t+1:t+H .

[0078] Further, the predicted value of the residual component is added to the corresponding periodic component c t+1:t+H , to obtain the final time series prediction result

[0079] To clearly show the overall flow of the present embodiment, FIG. 1 shows a flowchart of a residual cycle prediction method for enhancing time series prediction, FIG. 2 shows a schematic diagram of using Linear and MLP as the trunk prediction model, and FIG. 3 further shows a schematic diagram of the periodic component generation process after using the method of the present application.

[0080] To verify the effective performance of the present application, the present embodiment compares CycleNet / Linear and CycleNet / MLP of the present application with existing prediction models. FIG. 4 shows the comparison results of the prediction accuracy and the parameter amount of the original direct modeling method and the method of the present application in user power consumption prediction modeling. It can be found that the present application achieves higher prediction accuracy with smaller parameter size.

[0081] Further, FIG. 5 shows the prediction accuracy improvement brought by the present application, i.e., the residual cycle prediction method (RCF), combined with existing time series prediction models. It can be found that the present application can significantly improve the prediction accuracy of existing models.

[0082] Further, FIG. 6 shows the learned periodic features in the data of the invention in the cycle period after the end of the training. It can be found that the invention fully captures the periodic features existing in the user power consumption curve, including the daily and weekly cycle patterns, providing further interpretability for the accurate prediction of time series.

[0083] In summary, the residual periodic prediction method for enhancing time series prediction proposed by the invention can effectively improve the prediction accuracy of user power consumption prediction tasks, providing an important basis for power system power dispatch and an important guarantee for the safe and stable operation of the power system.

[0084] It should be noted that for the foregoing method embodiments, in order to facilitate description, they are all described as a combination of a series of actions, but those skilled in the art should know that the invention is not limited by the order of the described actions, because according to the invention, certain steps can be performed in other orders or simultaneously.

[0085] Based on the same idea as the residual periodic prediction method for enhancing time series prediction in the above embodiments, the invention also provides a residual periodic prediction system for enhancing time series prediction, which can be used to execute the above residual periodic prediction method for enhancing time series prediction. For the sake of convenience, in the structural schematic diagram of the residual periodic prediction system for enhancing time series prediction embodiment, only the part related to the embodiment of the invention is shown, and those skilled in the art can understand that the illustrated structure does not constitute a limitation on the device, which can include more or fewer components than the illustrated, or combine certain components, or different component arrangements.

[0086] Please refer to FIG. 7, in another embodiment of the present application, a residual periodic prediction system 100 for enhancing time series prediction is provided, which includes a cycle determination module 101, a cycle period generation module 102, a periodic component acquisition module 103, a residual component calculation module 104, a residual component prediction module 105 and a time series prediction module 106;

[0087] The cycle determination module 101 is configured to determine the cycle length W of the data set; the data set includes time series data;

[0088] The cycle period generation module 102 is configured to generate a learnable cycle period Q for time series data with D variables The cycle period Q is obtained by cyclically copying the cycle period Q to obtain a periodic component sequence C equal in length to the time series X;

[0089] The periodic component acquisition module 103 is configured to intercept a sample x from the time series X t-L+1:twherein t is a time index of the sample, L is a history lookback window length of the sample; the periodic component c corresponding to the sample is intercepted from the periodic component sequence C t-L+1:t ;

[0090] The residual component calculation module 104 is configured to obtain a residual component x by subtracting the periodic component c from the original input sequence x t-L+1:t ; t-L+1:t , t-L+1:t ;

[0091] The residual component prediction module 105 is configured to input the residual component x into a backbone prediction model for residual prediction to obtain a predicted value of the residual component , t-L+1:t ; wherein H is a prediction window length of the sample; the periodic component c corresponding to the prediction window of the sample is intercepted from the periodic component sequence C t+1:t+H ;

[0092] The time series prediction module 106 is configured to add the predicted value of the residual component x to the corresponding periodic component c to obtain a final time series prediction result ; t+1:t+H

[0093] It should be noted that the residual periodic prediction system for enhancing time series prediction of the present application corresponds to the residual periodic prediction method for enhancing time series prediction of the present application one by one, and the technical features and advantages described in the above embodiment of the residual periodic prediction method for enhancing time series prediction are applicable to the embodiment of the residual periodic prediction system for enhancing time series prediction, and the specific content can be referred to the description in the method embodiment of the present application, which will not be repeated here, and hereby declared.

[0094] In addition, in the embodiment of the residual periodic prediction system for enhancing time series prediction of the above embodiment, the logical division of each program module is only an example, and in actual application, the above functions can be completed by different program modules according to the needs, for example, the configuration requirements of the corresponding hardware or the convenience of software implementation, that is, the internal structure of the residual periodic prediction system for enhancing time series prediction is divided into different program modules to complete all or part of the functions described above.

[0095] Please refer to FIG. 8, in one embodiment, an electronic device for implementing the residual periodic prediction method for enhancing time series prediction is provided, the electronic device 200 can include a first processor 201, a first memory 202 and a bus, and can further include a computer program stored in the first memory 202 and executable on the first processor 201, such as a residual periodic prediction program for enhancing time series prediction 203.​​

[0096] The first memory 202 includes at least one type of readable storage medium, such as flash memory, a mobile hard disk, a multimedia card, a card-type memory (e.g., an SD or DX memory, etc.), a magnetic memory, a disk, an optical disk, etc. In some embodiments, the first memory 202 can be an internal storage unit of the electronic device 200, such as a mobile hard disk of the electronic device 200. In other embodiments, the first memory 202 can also be an external storage device of the electronic device 200, such as a plug-in mobile hard disk, a smart media card (SMC), a secure digital (SD) card, a flash card, etc. Further, the first memory 202 can include both an internal storage unit and an external storage device of the electronic device 200. The first memory 202 can be used to store application software and various data installed in the electronic device 200, such as the code of the residual period prediction program 203 for enhancing time series prediction, and can also be used to temporarily store data that has been output or will be output.

[0097] The first processor 201 can be composed of an integrated circuit in some embodiments, such as a single packaged integrated circuit, or a plurality of packaged integrated circuits with the same or different functions, including one or more combinations of a central processing unit (CPU), a microprocessor, a digital processing chip, a graphics processor, and various control chips, etc. The first processor 201 is the control unit of the electronic device, which connects various components of the entire electronic device through various interfaces and lines, and executes various functions and processes data of the electronic device 200 by running or executing programs or modules stored in the first memory 202 and calling data stored in the first memory 202.

[0098] FIG. 8 only shows an electronic device with components, and those skilled in the art can understand that the structure shown in FIG. 8 does not constitute a limitation on the electronic device 200, and can include fewer or more components than shown, or combine certain components, or different component arrangements.

[0099] The residual period prediction program 203 for enhancing time series prediction stored in the first memory 202 in the electronic device 200 is a combination of a plurality of instructions, which, when running in the first processor 201, can achieve:

[0100] determining a period length W of a data set; the data set includes time series data;

[0101] For time series data with D variables, generate learnable cyclic periods. By cyclically copying the cycle period Q, a periodic component sequence C of the same length as the time series X is obtained;

[0102] Extracting sample x from time series X t-L+1:t Where t is the time index of the sample, and L is the length of the historical lookback window of the sample; the periodic component c corresponding to the sample is extracted from the periodic component sequence C. t-L+1:t ;

[0103] By using the original input sequence x t-L+1:t Subtract the periodic component c t-L+1:t Obtain the residual component x , t-L+1:t ;

[0104] The residual component x , t-L+1:t The data is input into the backbone prediction model for residual prediction to obtain the predicted values ​​of the residual components. Where H is the prediction window length of the sample; the periodic component c corresponding to the prediction window of the sample is extracted from the periodic component sequence C. t+1:t+H ;

[0105] The predicted values ​​of the residual components With the corresponding periodic component c t+1:t+H Add them together to get the final time series prediction result.

[0106] Furthermore, if the modules / units integrated in the electronic device 200 are implemented as software functional units and sold or used as independent products, they can be stored in a non-volatile computer-readable storage medium. The computer-readable medium may include: any entity or device capable of carrying the computer program code, a recording medium, a USB flash drive, a portable hard drive, a magnetic disk, an optical disk, a computer memory, or a read-only memory (ROM).

[0107] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by a computer program instructing relevant hardware. The program can be stored in a non-volatile computer readable storage medium, and when executed, can include the processes of the above-mentioned embodiment methods. Any reference to memory, storage, database or other medium used in the embodiments provided by the present application can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. As an illustration but not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDR SDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0108] Any combination of the technical features of the above embodiments can be made. In order to make the description simple, all possible combinations of the technical features in the above embodiments are not described, however, as long as the combination of the technical features does not exist, it should be considered as the scope of the present application.

[0109] The above embodiments are the preferred embodiments of the present application, but the embodiments of the present application are not limited to the above embodiments, and any changes, modifications, substitutions, combinations and simplifications made without departing from the spirit and principles of the present application shall be equivalent replacement methods and shall be within the scope of protection of the present application.

Claims

1. A residual period prediction method for enhancing time series prediction, characterized in that, The method comprises the following steps: determining a cycle length W of a data set; the data set comprises time series data; For time series data with D variables, generating a learnable cycle period obtaining a cycle component sequence C of the same length as the time series X by cyclically copying the cycle Q; Taking a sample x from a time series X t-L+1:t where t is the time index of the sample, L is the length of the historical look-back window of the sample; taking a periodic component c corresponding to the sample from a periodic component sequence C t-L+1:t ; By subtracting the periodic component c from the original input sequence x t-L+1:t Subtracting the periodic component c t-L+1:t Obtaining the residual component x' t-L+1:t ; The residual component x' is input into the main prediction model for residual prediction, obtaining a predicted value of the residual component t-L+1:t The residual component x' is input into the main prediction model for residual prediction, obtaining a predicted value of the residual component where H is a prediction window length of the sample; a periodic component c corresponding to the prediction window of the sample is taken from the periodic component sequence C t+1:t+H ; predicting a value for the residual component with the corresponding periodic component c t+1:t+H adding, to obtain a final time series prediction result 2. The residual periodicity prediction method for enhancing time series prediction according to claim 1, characterized in that, The cycle length W of the data set is determined by manually reasoning the characteristics of the time series data set or calculating the autocorrelation within the data set.

3. The residual periodicity prediction method for enhancing time series prediction according to claim 1, characterized in that, The cycle Q is trained together with the backbone prediction model through gradient back propagation in the training process, so as to learn the feature representation reflecting the inherent periodic pattern of the time series.

4. The residual periodicity prediction method for enhancing time series prediction according to claim 1, characterized in that, The backbone prediction model is a backbone model for residual prediction.

5. The residual periodicity prediction method for enhancing time series prediction according to claim 4, characterized in that, The backbone prediction model is a single-layer linear model or a two-layer MLP model.

6. The residual periodicity prediction method for enhancing time series prediction according to claim 1, wherein, The periodic component c t-L+1:t The generation of the periodic component c comprises the following steps: Q = (Q + t) mod W (t) ; Q (t) copy subsequently, and spliced obtaining a periodic component c of length L t-L+1:t .

7. The residual periodicity prediction method for enhancing time series prediction according to claim 1, wherein, The periodic component c t+1:t+H The generation of the periodic component c comprises the following steps: Q = (t + L) mod W (t+L) ; Q (t+L) copy subsequently, and spliced obtaining a periodic component c of length H t+1:t+H .

8. A residual period prediction system that enhances time series prediction, characterized by, The residual cycle prediction method for enhancing time series prediction according to any one of claims 1-7 comprises a cycle determination module, a cycle period generation module, a cycle component acquisition module, a residual component calculation module, a residual component prediction module, and a time series prediction module. The cycle determination module is configured to determine a cycle length W of a data set; the data set comprises time series data; The cycle period generation module is configured to generate a learnable cycle period for time series data with D variables. obtaining a cycle component sequence C of the same length as the time series X by cyclically copying the cycle Q; The periodic component obtaining module is configured to intercept a sample x from a time sequence X t-L+1:t wherein t is a time index of the sample, and L is a length of a historical lookback window of the sample; and intercept a periodic component c corresponding to the sample from a periodic component sequence C t-L+1:t ; The residual component calculation module is configured to calculate a residual component x′ by subtracting the periodic component c from the original input sequence x t-L+1:t t-L+1:t t-L+1:t ;​​ The residual component prediction module is configured to input the residual component x′ t-L+1:t into the backbone prediction model for residual prediction to obtain a predicted value of the residual component where H is a prediction window length of the sample; a periodic component c corresponding to the prediction window of the sample is taken from the periodic component sequence C t+1:t+H ; The time series prediction module is used to generate the predicted values ​​of the residual components. with the corresponding periodic component c t+1:t+H adding, to obtain a final time series prediction result 9. An electronic device, comprising: The electronic device comprises: at least one processor; and a memory connected to the at least one processor in communication; wherein The memory stores computer program instructions executable by the at least one processor, and the computer program instructions are executed by the at least one processor to enable the at least one processor to execute the residual cycle prediction method for enhancing time series prediction according to any one of claims 1-7.

10. A computer-readable storage medium storing a program, characterized in that, The program is executed by the processor to implement the residual cycle prediction method for enhancing time series prediction according to any one of claims 1-7.

Images