Modeling application of auto-encoder-based extreme learning machine in industrial production prediction

Abstract

The invention discloses a modeling application of an auto-encoder-based extreme learning machine in industrial production prediction, which comprises the steps of encoding input data by using the auto-encoder to obtain main characteristics of original data, and removing noise and redundant information in the data; comparing the auto-encoder reconstruction losses corresponding to different auto-encoder structures according to the average relative error, selecting the auto-encoder structure with the minimum loss as a final auto-encoder, and using the selected auto-encoder to encode the original data as the input of an extreme learning machine. According to output weight of the extreme learning machine calculated by Moore-Penrose generalized inverse matrix, the final output of the extreme learning machine is obtained. According to the method, noise and redundant information in the data are removed, the main characteristics extracted by the auto-encoder are used as the input of the extreme learning machine, the sensitivity of the extreme learning machine to hidden layer node selection is reduced, and the precision and robustness of model prediction are improved.

Description

technical field [0001] The invention relates to the technical field of industrial production prediction, in particular to a modeling application of an autoencoder-based extreme learning machine in industrial production prediction. Background technique [0002] In recent years, chemical production technology has made some progress, among which pure terephthalic acid (PTA) has played a key role. PTA is in increasing demand in the market and its production cost affects the overall energy efficiency level in complex chemical processes. In PTA production technology, one of the effective ways to save energy is to establish an accurate production forecast model. [0003] However, due to the need to deal with a large number of data characteristics and various influencing factors in the PTA production process, there is a strong nonlinear relationship between variables. Traditional production forecasting models based on statistics are too complex and require the support of relevant ...

AI Technical Summary

Problems solved by technology

[0003] However, due to the large amount of data characteristics and various influencing factors that need to be processed in the PTA production process, there is a strong nonlinear relationship between the variables

Traditional production forecasting models based on statistics are too complex and require the support of relevant domain knowledge, so they are no longer suitable for building adaptive production forecasting models for increasingly complex petrochemical production data

Images