Data prediction model adjusting and optimizing method and apparatus based on LSTM network

A data prediction and data technology, applied in the field of data processing, can solve problems such as inability to guarantee local optimal solutions, slow calculation speed, large data sets of difficult variables, etc., and achieve the effects of fast calculation speed, good prediction and elimination of risks

Active Publication Date: 2018-11-06
HARBIN INST OF TECH
View PDF13 Cites 9 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Benefits of technology

It describes a way that uses a special type of neural net called Long Short Term Memory Network (LSTM). These networks have several layers between input/output devices and memory cells. They work by processing sequences of data like speech or images from sensors connected through communication channels such as telephone lines. By removing irrelevant parts during this process, we reduce our computational resources needed compared to traditional methods. Overall, these techniques make it possible to quickly learn new patterns without being affected by any external factors affecting their accuracy.

Problems solved by technology

This patented technology solves issues related to learning sequential relationships between values over longer periods without requiring complex parameter settings or retraining procedures.

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
  • Data prediction model adjusting and optimizing method and apparatus based on LSTM network
  • Data prediction model adjusting and optimizing method and apparatus based on LSTM network
  • Data prediction model adjusting and optimizing method and apparatus based on LSTM network

Examples

Experimental program
Comparison scheme
Effect test

Embodiment 1

[0041] Such as figure 1 As shown, an LSTM network-based data prediction model tuning method provided by Embodiment 1 of the present invention includes:

[0042] Step S1, preprocessing:

[0043] In this embodiment, assume that the variable to be predicted is Y, calculate the period value of the variable Y to be predicted according to the data of the variable Y to be predicted in the data set, arrange the period values ​​from small to large, and obtain the first N small period values ​​of the variable to be predicted , that is, let the periodic value of Y be n={n 1 ,n 2 ,...,n N}, n 1 2 N .

[0044] Calculate the correlation coefficient between each variable in the data set and the variable to be predicted, arrange each variable according to its correlation coefficient from large to small, and extract the data of the first few variables in the data set whose sum of correlation coefficients is greater than the coefficient threshold to form a training set.

[0045] The origi...

Embodiment 2

[0055] Embodiment 2 is basically the same as Embodiment 1, and the similarities will not be described in detail. The difference lies in:

[0056] In step S1, when calculating the periodic value of the variable to be predicted according to the data of the variable to be predicted in the data set, the data of the variable to be predicted is normalized according to the order of time series, and the sequence value of each zero-crossing point is recorded, and the adjacent two The difference between the sequence values ​​of zero-crossing points is recorded as the periodic value of the variable to be predicted. The sequence value here is the number of rows in the dataset.

[0057] Preferably, when acquiring the first N smaller cycle values ​​of the variable to be predicted, the value range of N is 4-7. Further preferably, the value of N is 5, that is, the first 5 smaller period values ​​are selected sequentially, and a total of 5 rounds of training are performed. It has been verifi...

Embodiment 3

[0065] Such as Figure 5 As shown, Embodiment 3 of the present invention provides an LSTM network-based data prediction model tuning device, including a preprocessing unit 100 and a model training unit 200, wherein:

[0066] The preprocessing unit 100 is used to calculate the periodic value of the variable to be predicted according to the data of the variable to be predicted in the data set, arrange the periodic values ​​from small to large, and obtain the N smallest periodic values ​​of the variable to be predicted; The correlation coefficient between the variable and the variable to be predicted, and arrange each variable according to its correlation coefficient from large to small, and extract the data of the first few variables whose sum of correlation coefficient is greater than the coefficient threshold in the data set to form a training set.

[0067] The model training unit 200 is used to construct a model using the training set obtained by the preprocessing unit 100 an...

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

The invention relates to a data prediction model adjusting and optimizing method based on an LSTM network. The method include steps of preprocessing: obtaining previous N small period values of a to-be-predicted variable, and extracting data of previous multiple variables whose correlation coefficient sums are greater than a coefficient threshold in a data set to form a training set; and model training: performing N rounds of training according to a sequence of the period values from large to small to calculate an optimal solution model, wherein each round of training comprises converting thetraining set from time sequence data to a supervised learning sequence; inputting the supervised learning sequence to the LSTM network to obtain a training model of the round; and comparing a root-mean-square error obtained by the training model of the round with a root-mean-square error obtained by training of the previous round, and reserving the training model corresponding to a smaller value as the optimal solution model. The invention also relates to a data prediction model adjusting and optimizing apparatus based on the LSTM network. According to the adjusting and optimizing method and the apparatus, optimization is performed based on the LSTM network, data prediction can be realized, the calculating speed is fast, and the prediction effect is good.

Description

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

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
Owner HARBIN INST OF TECH
Who we serve
  • R&D Engineer
  • R&D Manager
  • IP Professional
Why Eureka
  • Industry Leading Data Capabilities
  • Powerful AI technology
  • Patent DNA Extraction
Social media
Try Eureka
PatSnap group products

Browse by: Latest US Patents, China's latest patents, Technical Efficacy Thesaurus, Application Domain, Technology Topic.

© 2024 PatSnap. All rights reserved.Legal|Privacy policy|Modern Slavery Act Transparency Statement|Sitemap