Recurrent neural network training method

A technology of cyclic neural network and training method, which is applied in the direction of neural learning method, biological neural network model, neural architecture, etc., and can solve the problem that memory items cannot converge in training results, data content that cannot accurately predict time changes, gradient explosion, etc. problem, to achieve the effect of reducing the possibility of overfitting, high accuracy, strong early warning and control ability

Pending Publication Date: 2022-04-29
CHINA EVERBRIGHT BANK
View PDF14 Cites 3 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

[0005] However, the chain calculation method in actual operation may have a gradient explosion problem, that is, the memory items are superimposed in the cyclic neural network, resulting in the inability to converge the training results in the end.
[0006] Existing recurrent neural network training methods cannot accurately predict data content that changes over time

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
  • Recurrent neural network training method
  • Recurrent neural network training method
  • Recurrent neural network training method

Examples

Experimental program
Comparison scheme
Effect test

Embodiment 2

[0135] The following is based on the LSTM model to train and predict the daily frequency pressure of the database index in a certain system, in order to obtain the data trend of the index in the future, and then obtain the pressure of the database under the system by calculating the increase, which is used to judge the risk of the database and Make early warning management.

[0136] First of all, the daily frequency pressure index of the database is a processing index, and the original basic index is derived from the concurrent time of the database.

[0137] Calculated in the day dimension, the proportion of data concurrent time in a day is taken as the daily frequency pressure. That is to say, there is a value between 0-100 every day, and the time series data (T1, T2, ..., Tn) of this indicator is obtained in the time dimension. The target indicator is T, and the business date is t, then the target indicator value is Tt, and the forecast target starts from t+1 day.

[0138]...

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 recurrent neural network training method, and the method comprises the steps: carrying out the training and prediction of the daily frequency pressure of a database index under a certain system based on an LSTM model, so as to obtain the data trend of the index in a period of time in the future, and obtaining the pressure condition of the database under the system through the calculation of the amplification, and enabling the pressure condition to be used for judging the risk of the database and carrying out the early warning management. A daily frequency pressure index of the database is a processing index, and data preprocessing of missing value complementation is carried out on the data; constructing a complete time series data set (Xt, Tt) n; and after feature combination is completed, starting to train the model. According to the recurrent neural network training method, after data processing and arrangement are carried out on database indexes supported in a computer software system, modeling is carried out in a neural network type machine learning mode, so that the prediction condition of the database indexes in future time is obtained; based on a prediction result, index management can be carried out on a database under the system, and meanwhile, high early warning and control capabilities are achieved under various risks.

Description

technical field [0001] The invention relates to a dynamic process monitoring and generating method for automatic data modeling, and specifically relates to a training method for a cyclic neural network. Background technique [0002] The input and output of traditional neural networks are independent of each other, that is, there is no correlation between the input set and the output set. In real cases, there are indeed many cases where the input set and the output set are related in time or space. [0003] Through a certain method, the input and output sets of each training can be connected to a certain extent during the training of the neural network. Based on this, the recurrent neural network (RNN) creatively introduces the concept of "memory", and by adding memory items, the training process of the neural network can provide a relationship bridge for the input and output sets to a certain extent. [0004] In practice, first try to apply the recurrent neural network mode...

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): G06N3/04G06N3/08G06K9/62G06F16/21
CPCG06N3/084G06F16/211G06N3/048G06N3/044G06F18/217G06F18/214
Inventor 王丽史晨阳彭晓王岗潘竹邢世伟
Owner CHINA EVERBRIGHT BANK
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