KR-20260065672-A - METHOD FOR PREDICTING FUTURE PATTERNS OF TIME SERIES DATA USING LARGE LANGUAGE MODEL

KR20260065672AKR 20260065672 AKR20260065672 AKR 20260065672AKR-20260065672-A

Abstract

An apparatus for predicting time-series data of a biosignal according to the present invention comprises a sensor unit for acquiring a user's biosignal and a processor for sampling the acquired biosignal at predetermined time intervals to include physiological measurements according to changes over time, normalizing the sampled biosignal by adjusting it so that the physiological measurements are included within a predetermined range, concatenating the time and the normalized physiological measurements corresponding to the time based on the normalized biosignal to generate structured data for the biosignal, and applying the structured data for the biosignal as an input value to a tokenization module of a transformer-based large language model (LLM) to output a predicted biosignal.

Inventors

박철수
강영신
김현중
양근보

Assignees

광운대학교 산학협력단

Dates

Publication Date: 20260511
Application Date: 20241101

Claims (10)

In a device for predicting time series data of biosignals, A sensor unit for acquiring a user's biosignal; and The above-mentioned acquired biosignal is sampled at predetermined time intervals to include physiological measurements according to changes over time, and The above-mentioned sampled biosignal is normalized by adjusting it so that the physiological measurement value is included within a predetermined range, Based on the above normalized biosignal, structured data for the biosignal is generated by concatenating the above time and the above normalized physiological measurement value corresponding to the above time, and A time-series data prediction device for a biosignal, characterized by including a processor that applies structured data for the above biosignal as an input value to a tokenization module of a transformer-based large language model (LLM) and outputs a predicted biosignal.
In a device for predicting time series data of biosignals, A sensor unit for acquiring a user's biosignal; and The above-mentioned acquired biosignal is sampled at predetermined time intervals to include physiological measurements according to changes over time, and The above-mentioned sampled biosignal is normalized by adjusting it so that the physiological measurement value is included within a predetermined range, Based on the sampled biosignal above, a frequency data list including a frequency band according to the change in time and the magnitude of the frequency band is extracted, and Based on the normalized biosignal and the frequency data list, structured data for the biosignal is generated by concatenating the time, the normalized physiological measurement value corresponding to the time, and the frequency data list corresponding to the time. A time-series data prediction device for a biosignal, characterized by including a processor that applies structured data for the above biosignal as an input value to a tokenization module of a transformer-based large language model (LLM) and outputs a predicted biosignal.
In Paragraph 2, The above biosignal is, A device for predicting time-series data of a biosignal, which is one of an electrocardiogram (ECG), electroencephalography (EEG), photoplethysmography (PPG), blood pressure (BP), or saturation of partial pressure oxygen (SpO2).
In a device for predicting time series data of biosignals, A communication unit that receives the user's biosignal; and The above-mentioned received biosignal is sampled at predetermined time intervals to include physiological measurements according to changes over time, and The above-mentioned sampled biosignal is normalized by adjusting it so that the physiological measurement value is included within a predetermined range, Based on the above normalized biosignal, structured data for the biosignal is generated by concatenating the above time and the above normalized physiological measurement value corresponding to the above time, and A time-series data prediction device for a biosignal, characterized by including a processor that applies structured data for the above biosignal as an input value to a tokenization module of a transformer-based large language model (LLM) and outputs a predicted biosignal.
In a device for predicting time series data of biosignals, A communication unit that receives the user's biosignal; and The above-mentioned received biosignal is sampled at predetermined time intervals to include physiological measurements according to changes over time, and The above-mentioned sampled biosignal is normalized by adjusting it so that the physiological measurement value is included within a predetermined range, Based on the sampled biosignal above, a frequency data list including a frequency band according to the change in time and the magnitude of the frequency band is extracted, and Based on the normalized biosignal and the frequency data list, structured data for the biosignal is generated by concatenating the time, the normalized physiological measurement value corresponding to the time, and the frequency data list corresponding to the time. A time-series data prediction device for a biosignal, characterized by including a processor that applies structured data for the above biosignal as an input value to a tokenization module of a transformer-based large language model (LLM) and outputs a predicted biosignal.
In Paragraph 5, The above biosignal is, A device for predicting time-series data of a biosignal, which is one of an electrocardiogram (ECG), electroencephalography (EEG), photoplethysmography (PPG), blood pressure (BP), or saturation of partial pressure oxygen (SpO2).
In a method for predicting time series data of biosignals, A step of acquiring the user's biosignal; A step of sampling the above-mentioned biosignal at predetermined time intervals to include physiological measurements according to changes over time; A step of normalizing the sampled biosignal by adjusting it so that the physiological measurement value is included within a predetermined range; A step of generating structured data for the biosignal by concatenating the time and the normalized physiological measurement value corresponding to the time based on the normalized biosignal; and A method for predicting time series data of a biosignal, characterized by including the step of applying structured data for the above biosignal as an input value to a tokenization module of a transformer-based large language model (LLM) to output a predicted biosignal.
In a method for predicting time series data of biosignals, A step of acquiring the user's biosignal; A step of sampling the above-mentioned biosignal at predetermined time intervals to include physiological measurements according to changes over time; A step of normalizing the sampled biosignal by adjusting it so that the physiological measurement value is included within a predetermined range; A step of extracting a frequency data list including a frequency band according to the change in time and the magnitude of the frequency band based on the sampled biosignal; A step of generating structured data for the biosignal by concatenating the time, the normalized physiological measurement value corresponding to the time, and the frequency data list corresponding to the time based on the normalized biosignal and the frequency data list; and A method for predicting time series data of a biosignal, characterized by including the step of applying structured data for the above biosignal as an input value to a tokenization module of a transformer-based large language model (LLM) to output a predicted biosignal.
In Paragraph 8, The above biosignal is, A method for predicting time series data of a biosignal, which is one of an electrocardiogram (ECG), electroencephalography (EEG), photoplethysmography (PPG), blood pressure (BP), or saturation of partial pressure oxygen (SpO2).
A computer-readable recording medium storing a program for executing on a computer a method for predicting time-series data of a biosignal described in any one of claims 7 to 9.

Description

Method for Predicting Future Patterns of Time Series Data Using a Large Language Model The present invention relates to an apparatus and method for effectively predicting future data patterns in time series data using a large language model. Time series forecasting is a technique that predicts future data patterns based on past data and is performed using deep learning models. Until now, methods using recurrent neural network models such as LSTM (Long Short-Term Memory) or GRU (Gated Recurrent Unit) have been primarily studied for time series forecasting, but there was a constraint that various patterns had to be learned from a large amount of data to achieve accurate time series forecasting. Time-series data prediction of biosignals plays a crucial role in the medical field. It can be utilized for various purposes, including health monitoring, disease prevention, treatment optimization, and medical resource management. Through such predictions, potential health issues can be detected early, and personalized medical services can be provided, contributing to the improvement of overall healthcare quality and cost reduction. However, biosignal data has the unique characteristic of being limited in its collection and acquisition. In particular, technologies capable of making predictions using only small amounts of biosignals offer several advantages. They are cost-effective and enable real-time monitoring and rapid prediction. Furthermore, they enhance user convenience and support early diagnosis and rapid decision-making by medical professionals. Such technologies significantly improve access to medical services and provide efficient health management opportunities to a greater number of people. To overcome these limitations, the present invention proposes a novel time-series biosignal prediction method based on a Large Language Model (LLM). The present invention is characterized by the ability to achieve high prediction performance with only a small amount of biosignal data. The accompanying drawings, which are included as part of the detailed description to aid in understanding the present invention, provide embodiments of the present invention and explain the technical concept of the present invention together with the detailed description. Figure 1 is a diagram illustrating the layer structure of an artificial neural network. Figure 2 is a diagram illustrating an example of a deep neural network. Figure 3 is a diagram illustrating the structure of a transformer network. Figure 4 is a diagram illustrating a part of the data processing process of the LLM. FIG. 5 is an exemplary drawing for explaining a time series data prediction device for biosignals according to the present invention. FIG. 6 is a diagram illustrating a block diagram for explaining the function of a time series data prediction device (600) of a biosignal according to the present invention. FIG. 7 is a diagram illustrating the process of generating structured data through a time series data prediction device for biosignals according to the present invention. FIG. 8 is a diagram illustrating the parameter definition and value setting of an LLM for implementing a time series data prediction device for biosignals according to the present invention. FIG. 9 is a diagram illustrating the result of predicting an electrocardiogram waveform using a time series data prediction device for biosignals according to the present invention. FIG. 10 is a diagram illustrating the performance of the result of predicting an electrocardiogram waveform through a time series data prediction device for biosignals according to the present invention. FIG. 11 is a diagram illustrating the difference in performance between a time series data prediction device for biosignals according to the present invention and a device using a conventional LSTM. FIG. 12 is another exemplary drawing for explaining a time series data prediction device for biosignals according to the present invention. FIG. 13 is another example diagram illustrating the process of generating structured data through a time-series data prediction device for biosignals according to the present invention. FIG. 14 is another example drawing illustrating a block diagram to explain the function of a time series data prediction device (1400) of a biosignal according to the present invention. The present invention is capable of various modifications and may have various embodiments, and specific embodiments are illustrated in the drawings and described in detail. However, this is not intended to limit the invention to specific embodiments, and it should be understood that the invention includes all modifications, equivalents, and substitutions that fall within the spirit and scope of the invention. When it is stated that one component is "connected" or "connected" to another component, it should be understood that while it may be directly connected or connected to that other component, there may also be other com