Industrial equipment monitoring data prediction method

Abstract

The invention discloses an industrial equipment monitoring data prediction method, which comprises the steps of S1, acquiring an industrial equipment monitoring data sequence which comprises industrial equipment monitoring data from a first moment to a t-1-th moment; s2, inputting the industrial equipment monitoring data sequence into a prediction model based on a multi-head attention mechanism; and S3, outputting industrial equipment prediction data at the t moment. When a traditional RNN and similar structure LSTM and GRU are used for training long-sequence samples, the memory limitation of the traditional RNN and similar structure LSTM and GRU can hinder batch processing of the training samples, and the training efficiency is low. According to the method, a multi-head attention mechanism is adopted, so that the training efficiency can be effectively improved, and the industrial equipment monitoring data prediction efficiency is further improved.

Description

technical field [0001] The invention relates to the technical field of industrial equipment monitoring, in particular to a method for predicting industrial equipment monitoring data. Background technique [0002] In recent years, with the advent of the Industry 4.0 era, industrial big data has become an important research topic. Due to the complex production process, large number of sensors and fast sampling frequency of industrial equipment, it is easy to accumulate a large amount of data in a short period of time. Due to the characteristics of label data and other characteristics, if there are special working conditions, it will often cause large economic losses. Therefore, if the abnormalities in the production process can be predicted in time, the efficiency of the entire production process will be improved, resulting in greater application. value. [0003] Among the existing prediction methods for industrial equipment monitoring data, models based on LSTM and GRU deep...

AI Technical Summary

Problems solved by technology

However, RNN and LSTM and GRU with similar structures hinder the parallelization between training samples due to their inherent sequential properties. For long-sequence training samples, memory constraints hinder the batch processing of training samples, and the training efficiency is low.

Images