Unlock instant, AI-driven research and patent intelligence for your innovation.

System and method for multi-horizon time series forecasting with dynamic temporal context learning

A time series, context technology, applied in the field of end-to-end deep learning frameworks, which can solve problems such as inability to explore temporal patterns

Pending Publication Date: 2020-03-06
BEIJING JINGDONG SHANGKE INFORMATION TECH CO LTD +1
View PDF0 Cites 6 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

However, existing LSTM-based methods cannot explore temporal patterns associated with long future bounds

Method used

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
View more

Image

Smart Image Click on the blue labels to locate them in the text.
Viewing Examples
Smart Image
  • System and method for multi-horizon time series forecasting with dynamic temporal context learning
  • System and method for multi-horizon time series forecasting with dynamic temporal context learning
  • System and method for multi-horizon time series forecasting with dynamic temporal context learning

Examples

Experimental program
Comparison scheme
Effect test

example 1

[0139] Example 1: In one example, the disclosed architecture is applied to Global Optimization Challenge 2018 (GOC2018) sales forecasting. GOC2018 Online Sales Dataset (http: / / jdata.joybuy.com / html / detail.html?id=4) is a public dataset collected from actual sales data of JD.com, a global online retail company. This dataset provides 6000 time series of daily sales data of 1000 products sold in 6 regions in China from 2016 to 2017. Participants were asked to forecast the daily demand in January 2018 for all products in all demand regions. Specifically, quantile forecasts are required, and each time series i has its dedicated quantile of interest q given in the dataset i , which ranges from 0:85 to 0:99.

[0140] Sales forecasting is challenging because multiple factors have to be considered simultaneously, such as product categories, geographic regions, promotions, etc. This disclosure briefly introduces the features available as historical and future information provided in ...

example 2

[0150] Example 2: In another example, the disclosed model is evaluated on the electricity price forecasting task introduced in the 2014 Global Energy Forecasting Competition (GEFCom2014) (Hong et al. 2016). The GEFCom2014 price forecast dataset contains hourly electricity prices for three years from 2011-01-01 to 2013-12-31. The task is to provide future 24-hour forecasts over 12 evenly spaced evaluation weeks. On a rolling basis, ground truth price information from previous rounds can also be used to predict future rounds. In this data set, hourly based regional and total electric load estimates are the two temporal characteristic variables available in both past and future information. According to the competition instructions, hourly price predictions should provide quantile predictions of 0.01, 0.02, ..., 0.99, given by q 1 ,...,q 99 express.

[0151] To make our setup comparable to (Wen et al. 2017), we train our model to provide only 5 quantile predictions on 0.01, 0...

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
Login to View More

PUM

No PUM Login to View More

Abstract

A system and a method for time series forecasting. The method includes: providing input feature vectors corresponding to a plurality of future time steps; performing bi-directional long-short term memory network (BiLSTM) on the input feature vectors to obtain hidden outputs corresponding to the plurality of future time steps; for each future time step: performing temporal convolution on the hiddenoutputs using a plurality of temporal scales to obtain context features at the plurality of temporal scales, and summating the context features at the plurality of temporal scales using a plurality of weights to obtain multi-scale context features; and converting the multi-scale context features to obtain the time series forecasting corresponding to the future time steps.

Description

[0001] Cross References to Related Applications [0002] This application claims priority and benefit from U.S. Provisional Patent Application Serial No. 62 / 723,696, filed August 28, 2018, the entire contents of which are incorporated herein by reference. [0003] Several references, which may include patents, patent applications, and various publications, are cited and discussed in the description of the present disclosure. Citation and / or discussion of these references are provided solely to clarify the description of the present disclosure and are not an admission that any such reference is "prior art" to the disclosure described herein. All references cited and discussed in this specification are hereby incorporated by reference in their entirety, as if each reference were individually incorporated by reference. technical field [0004] The present disclosure relates generally to time series forecasting, and more specifically to an end-to-end deep learning framework for m...

Claims

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
Login to View More

Application Information

Patent Timeline
no application Login to View More
Patent Type & Authority Applications(China)
IPC IPC(8): G06Q10/04G06N3/04G06N3/08
CPCG06Q10/04G06N3/049G06N3/084G06N3/045G06F17/18G06N3/044G06F17/11G06N3/0418
Inventor 范晨悠张雨泽袁融张弛邬迪
Owner BEIJING JINGDONG SHANGKE INFORMATION TECH CO LTD