KR-20260067894-A - DEVICE AND METHOD FOR GATE CIRCULATION UNIT-BASED TIME SERIES

Abstract

A gate-circulation unit-based time series prediction device according to one embodiment of the present invention It may include a normalization unit that normalizes training data having a length, a patching unit that performs patching to group the normalized data into a predetermined length, a gate circulation unit that takes the data patched through the patching unit as input and converts it into a latent vector of a preset dimension, and a prediction unit that flattens the latent vector and derives a time series prediction result through inverse normalization of the predicted value having a length through a single linear layer.

Inventors

• 정성원
• 김윤영

Assignees

• 서강대학교산학협력단

Dates

Publication Date

20260513

Application Date

20241106

Claims (20)

1. In a time series prediction device based on a gate circulation unit, A normalization unit that normalizes training data having a length; A patching unit that performs patching to group normalized data into predetermined lengths; A gate circulation unit that takes data patched through the patching unit as input and converts it into a latent vector of a preset dimension; and A gate-circulating unit-based time series prediction device comprising a prediction unit that flattens the above-mentioned latent vector and derives a time series prediction result through inverse normalization of a predicted value having length through a single linear layer .
2. In paragraph 1, The above normalization unit is, A gate cycle unit-based time series forecasting device that normalizes by applying a reversible instance normalization technique to mitigate the distribution difference between the above training data and label data .
3. In paragraph 1, The above patching part is, A gate cycle unit-based time series forecasting device that, based on a sliding window algorithm, repeats the last value of the training data by a stride to connect it to the last value of the training data and truncates by a preset patch length.
4. In the third, The above patching part is, A gate cycle unit-based time series prediction device that generates patched training data by accumulating cut data through repeated cutting of the patch length.
5. In paragraph 4, A gate cycle unit-based time series forecasting device in which the patched training data has a length shortened compared to the length of the training data.
6. In paragraph 4, The above gate circulation unit is, The patched data is input into each cell of the gate circulation unit as many times as the number of times the above-mentioned cut data has accumulated, and A gate cycle unit-based time series forecasting device that performs learning to output weights and learning weights through the latent vector of the previous cell and the data input to the cell.
7. In paragraph 6, The above weight indicates the degree to which unnecessary information of the potential vector of the previous cell is reset, and A gate cycle unit-based time series forecasting device in which the above learning weight is a weight value that reflects the information of the previous cell and the information of the current cell to generate a potential vector of the current cell.
8. In paragraph 6, The above gate circulation unit is, A gate-circulating unit-based time series forecasting device that performs learning to generate a temporary vector using the value obtained by multiplying the learned weights and the latent vector of the previous cell and the data entered into the cell as inputs.
9. In paragraph 8, The above gate circulation unit is, A gate cycle unit-based time series forecasting device that generates a potential vector of the current cell based on the above temporary vector, the above potential vector, and the above learning weight.
10. In paragraph 1, The above prediction unit is, A gated circulation unit-based time series forecasting device that learns the process of backpropagating the result of calculating the mean square error loss between the above time series forecast result and the actual value.
11. In a gated cyclic unit-based time series forecasting method, (a) A step of normalizing training data having a normalization part of length; (b) A step in which the patching unit performs patching to group normalized data into a predetermined length; (c) a step in which a gate circulation unit takes data patched through the patching unit as input and converts it into a latent vector of a preset dimension; and (d) A gate cycle unit-based time series prediction method comprising the step of flattening the above potential vector and deriving a time series prediction result through inverse normalization of a prediction value having length through a single linear layer.
12. In Paragraph 11, The above normalization unit is, A gated cycle unit-based time series forecasting method that normalizes by applying a reversible instance normalization technique to mitigate the difference in distribution between the above training data and label data.
13. In Paragraph 11, The above patching part is, A gate cycle unit-based time series forecasting method based on a sliding window algorithm, which repeats the last value of the training data by a stride to connect it to the last value of the training data and truncates by a preset patch length.
14. In Paragraph 13, The above patching part is, A gated cycle unit-based time series forecasting method that generates patched training data by accumulating cut data through repeated cutting of the patch length.
15. In Paragraph 14, A gated cycle unit-based time series forecasting method in which the patched training data has a length shortened compared to the length of the training data.
16. In Paragraph 14, The above gate circulation unit is, The patched data is input into each cell of the gate circulation unit as many times as the number of times the above-mentioned cut data has accumulated, and A gate cycle unit-based time series forecasting method that performs learning to output weights and learning weights through the latent vector of the previous cell and the data input to the cell.
17. In Paragraph 16, The above weight indicates the degree to which unnecessary information of the potential vector of the previous cell is reset, and A gated cycle unit-based time series forecasting method in which the above learning weight is a weight value that reflects the information of the previous cell and the information of the current cell to generate a latent vector of the current cell.
18. In Paragraph 16, The above gate circulation unit is, A gate cycle unit-based time series forecasting method that performs learning to generate a temporary vector using the value obtained by multiplying the learned weights and the latent vector of the previous cell and the data entered into the cell as input.
19. In Paragraph 18, The above gate circulation unit is, A gate cycle unit-based time series forecasting method that generates a potential vector of the current cell based on the above temporary vector, the above potential vector, and the above learning weight.
20. In Paragraph 11, The above prediction unit is, A gated recurrence unit-based time series forecasting method that learns the process of backpropagating the result of calculating the Mean Square Error loss between the above time series forecast result and the actual value.

Description

Device and Method for Gate Circulation Unit-Based Time Series Forecasting The present invention relates to a time series prediction device and method based on a gate cycle unit. Time series forecasting involves predicting future data based on time series data recorded in chronological order, and it is a critical task for businesses and the field. Time series forecasting is used in many sectors closely related to daily life, such as energy, finance, transportation, and weather. Furthermore, the importance of time series data is growing due to the mass production of data driven by the Internet of Things, digital transformation, and the rise of smart cities. As data and computing power increase, machine learning models are becoming the dominant models in the field of time series forecasting, and consequently, deep learning models for handling time series data are continuously being researched. Figure 1 is a diagram illustrating the structure of a machine learning model used for conventional time series forecasting. FIG. 2 is a diagram illustrating the configuration of a gate circulation unit-based time series prediction device according to one embodiment of the present invention. FIG. 3 is a diagram illustrating the structure of a gate-circulation unit-based time series prediction device in operation according to one embodiment of the present invention. FIG. 4 is a diagram illustrating the patching process of a gate circulation unit-based time series prediction device according to one embodiment of the present invention. FIG. 5 is a diagram illustrating the structure of a GRU of a gate circulation unit-based time series prediction device according to one embodiment of the present invention. FIG. 6 is a diagram illustrating a dataset of an experiment for evaluating the performance of a gate-circulating unit-based time series prediction device according to one embodiment of the present invention. FIGS. 7 to 9 are drawings illustrating a performance comparison table between a gate circulation unit-based time series prediction device according to one embodiment of the present invention and other models. FIG. 10 is a diagram illustrating the flow of a gate-circulating unit-based time series prediction method according to one embodiment of the present invention. The objectives and effects of the present invention, and the technical configurations for achieving them, will become clear by referring to the embodiments described in detail below in conjunction with the accompanying drawings. In describing the present invention, if it is determined that a detailed description of known functions or configurations may unnecessarily obscure the essence of the invention, such detailed description will be omitted. However, this is not intended to limit the invention to specific embodiments, and it should be understood that it includes all modifications, equivalents, and substitutions that fall within the spirit and scope of the invention. Furthermore, the terms described below are defined considering their functions in the present invention, and these may vary depending on the intentions or practices of the user or operator. However, the present invention is not limited to the embodiments disclosed below but may be implemented in various different forms. These embodiments are provided merely to ensure that the disclosure of the present invention is complete and to fully inform those skilled in the art of the scope of the invention, and the present invention is defined only by the scope of the claims. Therefore, such definition should be based on the content throughout this specification. The terms used in this application are used merely to describe specific embodiments and are not intended to limit the invention. The singular expression includes the plural expression unless the context clearly indicates otherwise. In this application, terms such as "comprising" or "having" are intended to specify the presence of the features, numbers, steps, actions, components, parts, or combinations thereof described in the specification, and should be understood as not precluding the existence or addition of one or more other features, numbers, steps, actions, components, parts, or combinations thereof. Unless otherwise defined, all terms used herein, including technical or scientific terms, have the same meaning as generally understood by those skilled in the art to which the present invention pertains. Terms such as those defined in commonly used dictionaries should be interpreted as having a meaning consistent with their meaning in the context of the relevant technology, and should not be interpreted in an ideal or overly formal sense unless explicitly defined in this application. Hereinafter, preferred embodiments of the present invention will be described in more detail with reference to the attached drawings. In order to facilitate an overall understanding of the present invention, the same reference numerals are used for identical c