Penicillin fermentation process fault prediction method based on Attention-LSTM of multivariate time series

Abstract

The invention discloses a penicillin fermentation process fault prediction method based on Attention-LSTM (Long Short Term Memory) of a multivariate time series, which comprises the following steps: firstly, selecting a plurality of related process variables in a penicillin fermentation process through Pearson correlation analysis, then endowing implicit vectors at different moments of an input sequence of the LSTM with different weights through an Attention mechanism, the neural network prediction model is enabled to be more effective in processing long-time sequence input, and fault prediction modeling of the penicillin fermentation process is realized. According to the method, the fault prediction model combining the attention mechanism and the LSTM is adopted to perform fault prediction on the penicillin fermentation process, so that the problem that important time sequence information is easily ignored when the existing LSTM processes long sequence input is solved, and the fault prediction based on the LSTM is more accurate.

Description

technical field [0001] The invention belongs to the field of fault prediction modeling and application in industrial fermentation production process, and specifically relates to a multivariate time series-based Attention-LSTM fault prediction method for penicillin fermentation process. Background technique [0002] In the actual production process, the fault diagnosis of the system is usually only based on the current situation, but if the early prediction of the system fault can be made, the fault can be found and eliminated in time before the system fails, so that the fault can be reduced impact on the system. Therefore, the fault prediction of the system is of great significance to ensure the safety of the system. [0003] Time series (or called dynamic sequence) refers to the sequence in which the values ​​of the same statistical index are arranged in the order of their occurrence time. Time series analysis is a time-domain method for identifying modal parameters by us...

AI Technical Summary

Problems solved by technology

However, in the face of long-term sequences, the cyclic neural network is prone to problems such as gradient disappearance, gradient explosion, and insufficient long-term memory capacity.

Images