# US 2023/0267304 A1 · Correlation recurrent unit for improving prediction performance of time-series data and correlation recurrent neural network

> Agents: this data is available as JSON from the patentgrep API and MCP server — see https://patentgrep.com/llms.txt

- Number: US 2023/0267304 A1
- Document ID: US20230267304A1
- Status: Published application
- Kind: A1
- Application number: 18/086,646
- Filed: 2022-12-22
- Published: 2023-08-24
- Earliest claimed: 2022-02-24
- Assignees: Pusan National University Industry-University Cooperation Foundation
- Applicants: Pusan National University Industry-University Cooperation Foundation
- Inventors: Hye Rim Bae; Sung-hyun Sim; Do Hee Kim; Han Seok Kang
- Web page: https://patentgrep.com/patent/US20230267304A1

## CPC

- G06N 3/044: Recurrent networks, e.g. Hopfield networks
- G06N 3/0442: characterised by memory or gating, e.g. long short-term memory [LSTM] or gated recurrent units [GRU]
- G06N 3/045: Combinations of networks
- G06N 3/09: Supervised learning
- G06N: COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS

## Abstract

A correlation recurrent unit and a correlation recurrent neural network are provided. An aspect provides a new neural network structure for training each decomposed time-series component by decomposing input time-series data xt into a plurality of time-series components (e.g., trend, seasonal, and remainder components) and improving the prediction performance of time-series data by training a relation between different time-series components.

## Claims

### Claim 1

A correlation recurrent unit comprised in a correlation recurrent neural network, the correlation recurrent unit comprising:
a seasonal-trend decomposition using LOESS (STL) cell configured to decompose time-series data (xt) corresponding to a time (t) when the time-series data (xt) is input, based on a selected time-series composition algorithm, into a plurality of time-series components;
an autocorrelation gate configured to obtain an autocorrelation degree with a hidden state value corresponding to a previous time (t−1) of each of the plurality of time-series components with respect to each of the plurality of time-series components;
a correlation gate configured to obtain a correlation degree with a hidden state value corresponding to a previous time (t−1) of another time-series component different from each of the plurality of time-series components with respect to each of the plurality of time-series components; and
a generation gate configured to determine whether to update the hidden state value of the previous time (t−1) generated for each of the plurality of time-series components, based on the autocorrelation degree and the correlation degree, when previous time-series data (xt-1) prior to the time-series data (xt) is input.

### Claim 2 (depends on 1)

The correlation recurrent unit of claim 1, wherein
the generation gate is configured to
identify a first time-series component, of which the autocorrelation degree or the correlation degree is closer to 1 in a predetermined range between 0 to 1 than other time-series components of the plurality of time-series components, and
with respect to the first time-series component, use a hidden state value (ht) at the time (t) when the hidden state value (ht) is generated without updating a previous hidden state value (ht-1) corresponding to the previous time (t−1).

### Claim 3 (depends on 2)

The correlation recurrent unit of claim 2, wherein
the generation gate is configured to identify a second time-series component, of which the autocorrelation degree or the correlation degree is closer to 0 in a predetermined range between 0 to 1 than other time-series components of the plurality of time-series components, and with respect to the second time-series component, generate the hidden state value (ht) at the time (t) by updating the previous hidden state value (ht-1) corresponding to the previous time (t−1).

### Claim 4 (depends on 1)

The correlation recurrent unit of claim 1, wherein
the STL cell is configured to decompose, based on an STL algorithm, the time-series data (xt) into the plurality of time-series components comprising at least a seasonal component (xt_seasonal) that is a time-series component relevant to a periodic pattern repeated from the time (t), and a trend component (xt_trend) that is a time-series component relevant to an upward or downward trend from the time (t).

### Claim 5 (depends on 4)

The correlation recurrent unit of claim 4, wherein
the autocorrelation gate is configured to
estimate and obtain a first autocorrelation degree (âtt) with a first hidden state value (ht-1_trend) corresponding to the previous time (t−1) of the same trend component as the trend component (xt_trend) of the plurality of time-series components, and
estimate and obtain a second autocorrelation degree (âts) with a second hidden state value (ht-1_seasonal) corresponding to the previous time (t−1) of the same seasonal component as the seasonal component (xt_seasonal) of the plurality of time-series components.

### Claim 6 (depends on 5)

The correlation recurrent unit of claim 5, wherein
the correlation gate is configured to
estimate and obtain a first correlation degree (ĉtt) with the second hidden state value (ht-1_seasonal) corresponding to the previous time (t−1) of a seasonal component different from the trend component (xt_trend) of the plurality of time-series components, and
estimate and obtain a second correlation degree (ĉts) with the first hidden state value (ht-1_trend) corresponding to the previous time (t−1) of a trend component different from the seasonal component (xt_seasonal) of the plurality of time-series components.

### Claim 7 (depends on 6)

The correlation recurrent unit of claim 6, wherein
the generation gate is configured to
respectively perform a Hadamard product operation with the previous hidden state value (ht-1) at the previous time (t−1) on the first and second autocorrelation degrees (âtt) and (âts) and the first and second correlation degrees (ĉtt) and (ĉts),
evaluate a correlation of the previous hidden state value (ht-1) at the previous time (t−1) with the time-series data (xt), based on a value obtained from the Hadamard product operation, and
determine, based on the correlation, whether to update the previous hidden state value (ht-1) at the previous time (t−1) that is generated for each of the plurality of time-series components.

### Claim 8

A correlation recurrent neural network comprising a plurality of correlation recurrent units for predicting time-series data, wherein each of the plurality of correlation recurrent units comprises:
a seasonal-trend decomposition using LOESS (STL) cell configured to decompose, into a trend component (xt_trend), a seasonal component (xt_seasonal), and a remainder component (xt_remainder), time-series data (xt) at a time (t) when the time-series data (xt) is input;
an autocorrelation gate configured to estimate and obtain, with respect to the trend component (xt_trend) at the time (t), a first autocorrelation degree (âtt) with a first hidden state value relevant to a trend component (xt-1_trend) corresponding to a previous time (t−1) prior to the time (t) and estimate and obtain, for the seasonal component (xt_seasonal) at the time (t), a second autocorrelation degree (âts) with a second hidden state value relevant to the seasonal component (xt-1_seasonal) corresponding to the time (t−1);
a correlation gate configured to estimate and obtain a first correlation degree (ĉtt) with the second hidden state value with respect to the trend component (xt_trend) at the time (t) and estimate and obtain a second correlation degree (ĉts) with the first hidden state value with respect to the seasonal component (xt_seasonal) at the time (t); and
a generation gate configured to, when generating a hidden state value of each of a plurality of time-series components at the time (t), increase a proportion of using the first hidden state value and the second hidden state value that are generated when previous time-series data (xt-1) at the previous time (t−1) is input as the first and second autocorrelation degrees (âtt) and (âts) and the first and second correlation degrees (ĉtt) and (ĉts) increase.
