# US 2023/0274180 A1 · Machine Learning Super Large-Scale Time-series Forecasting

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

- Number: US 2023/0274180 A1
- Document ID: US20230274180A1
- Status: Published application
- Kind: A1
- Application number: 17/652,863
- Filed: 2022-02-28
- Published: 2023-08-31
- Earliest claimed: 2022-02-28
- Assignees: Google Inc.
- Applicants: Google LLC
- Inventors: Xi Cheng; Jiashang Liu; Lisa Yin; Amir Hossein Hormati; Mingge Deng; Weijie Shen; Kashif Yousuf
- Web page: https://patentgrep.com/patent/US20230274180A1

## CPC

- G06N 20/00: Machine learning
- G06F 16/248: Presentation of query results
- G06N: COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06F: ELECTRIC DIGITAL DATA PROCESSING (computer systems based on specific computational models G06N)

## Abstract

A method for forecasting time-series data, when executed by data processing hardware, causes the data processing hardware to perform operations including receiving a time series forecasting query from a user requesting a time series forecast forecasting future data based on a set of current time-series data. The operations include obtaining, from the set of current time-series data, a set of training data. The operations include training, using a first portion of the set of training data, a first sub-model of a forecasting model and training, using a second portion of the set of training data, a second sub-model of the forecasting model. The second portion is different than the first portion. The operations include forecasting, using the forecasting model, the future data based on the set of current time-series data and returning, to the user, the forecasted future data for the time series forecast.

## Claims

### Claim 1

A computer-implemented method when executed by data processing hardware causes the data processing hardware to perform operations comprising:
receiving a time series forecasting query from a user, the time series forecasting query requesting the data processing hardware to perform a time series forecast forecasting future data based on a set of current time-series data, the set of current time-series data comprising a series of data points listed in time order;
obtaining, from the set of current time-series data, a set of training data;
training, using a first portion of the set of training data, a first sub-model of a forecasting model;
training, using a second portion of the set of training data, a second sub-model of the forecasting model, the second portion different than the first portion;
forecasting, using the forecasting model, the future data based on the set of current time-series data; and
returning, to the user, the forecasted future data for the time series forecast requested by the time series forecasting query.

### Claim 2 (depends on 1)

The method of claim 1, wherein obtaining the set of training data comprises sequentially splitting the set of current time-series data into the set of training data and a set of testing data.

### Claim 3 (depends on 1)

The method of claim 1, wherein:
the first portion of the set of training data comprises an entirety of the set of training data; and
the second portion of the set of training data comprises a configurable ratio of the entirety of the set of training data.

### Claim 4 (depends on 3)

The method of claim 3, wherein the second portion of the set of training data comprises a most recent portion of the set of training data.

### Claim 5 (depends on 1)

The method of claim 1, wherein training, using the second portion of the set of training data, the second sub-model of the forecasting model comprises selecting the second portion of the set of training data based on a minimum training data threshold, a maximum training data threshold, and a training data ratio.

### Claim 6 (depends on 1)

The method of claim 1, wherein:
the first sub-model of the forecasting model comprises a seasonal model, and
the second sub-model of the forecasting model comprises a trend model.

### Claim 7 (depends on 1)

The method of claim 1, wherein training, using the first portion of the set of training data, the first sub-model of the forecasting model comprises performing hyper-parameter tuning.

### Claim 8 (depends on 7)

The method of claim 7, wherein performing-hyper parameter tuning comprises reducing a search space of each respective hyper-parameter of a plurality of hyper-parameters.

### Claim 9 (depends on 8)

The method of claim 8, wherein reducing the search space of each respective hyper-parameter of the plurality of hyper-parameters comprises, for each respective hyper-parameter of the plurality of hyper-parameters, obtaining a training hyper-parameter minimum and a training hyper-parameter maximum, the training hyper-parameter minimum greater than a minimum of the respective hyper-parameter, and the training hyper-parameter maximum less than a maximum of the respective hyper-parameter.

### Claim 10 (depends on 1)

The method of claim 1, wherein forecasting, using the forecasting model, the future data based on the set of current time-series data comprises aggregating a first forecast predicted by the first sub-model using the current time-series data and a second forecast predicted by the second sub-model using the current time-series data.

### Claim 11

A system comprising:
data processing hardware; and
memory hardware in communication with the data processing hardware, the memory hardware storing instructions that when executed on the data processing hardware cause the data processing hardware to perform operations comprising:
receiving a time series forecasting query from a user, the time series forecasting query requesting the data processing hardware to perform a time series forecast forecasting future data based on a set of current time-series data, the set of current time-series data comprising a series of data points listed in time order;
obtaining, from the set of current time-series data, a set of training data,
training, using a first portion of the set of training data, a first sub-model of a forecasting model;
training, using a second portion of the set of training data, a second sub-model of the forecasting model, the second portion different than the first portion;
forecasting, using the forecasting model, the future data based on the set of current time-series data; and
returning, to the user, the forecasted future data for the time series forecast requested by the time series forecasting query.

### Claim 12 (depends on 11)

The system of claim 11, wherein obtaining the set of training data comprises sequentially splitting the set of current time-series data into the set of training data and a set of testing data.

### Claim 13 (depends on 11)

The system of claim 11, wherein:
the first portion of the set of training data comprises an entirety of the set of training data; and
the second portion of the set of training data comprises a configurable ratio of the entirety of the set of training data.

### Claim 14 (depends on 13)

The system of claim 13, wherein the second portion of the set of training data comprises a most recent portion of the set of training data.

### Claim 15 (depends on 11)

The system of claim 11, wherein training, using the second portion of the set of training data, the second sub-model of the forecasting model comprises selecting the second portion of the set of training data based on a minimum training data threshold, a maximum training data threshold, and a training data ratio.

### Claim 16 (depends on 11)

The system of claim 11, wherein: the first sub-model of the forecasting model comprises a seasonal model; and
the second sub-model of the forecasting model comprises a trend model.

### Claim 17 (depends on 11)

The system of claim 11, wherein training, using the first portion of the set of training data, the first sub-model of the forecasting model comprises performing hyper-parameter tuning.

### Claim 18 (depends on 17)

The system of claim 17, wherein performing-hyper parameter tuning comprises reducing a search space of each respective hyper-parameter of a plurality of hyper-parameters.

### Claim 19 (depends on 18)

The system of claim 18, wherein reducing the search space of each respective hyper-parameter of the plurality of hyper-parameters comprises, for each respective hyper-parameter of the plurality of hyper-parameters, obtaining a training hyper-parameter minimum and a training hyper-parameter maximum, the training hyper-parameter minimum greater than a minimum of the respective hyper-parameter, and the training hyper-parameter maximum less than a maximum of the respective hyper-parameter.

### Claim 20 (depends on 11)

The system of claim 11, wherein forecasting, using the forecasting model, the future data based on the set of current time-series data comprises aggregating a first forecast predicted by the first sub-model using the current time-series data and a second forecast predicted by the second sub-model using the current time-series data.
