Soft measurement modeling method based on mLASSO-MLP model

Abstract

The invention provides a soft measurement modeling method based on an mLASSO-MLP model, and the method can be used for soft measurement modeling of a complex nonlinear industrial process with multiplevariables and other characteristics, and achieves the prediction of a quality variable which is difficult to directly measure. Firstly, input and output data are trained and learned, and an MLP modelis established; and then introducing the mLASSO algorithm into the established MLP model for training, and evaluating the performance of the established mLASSO-MLP model. The method has the advantages that an effective soft measurement model is established, and the problems of multivariable redundancy and the like in the nonlinear process of industrial production can be solved; the model has thecharacteristics of high prediction precision, strong universality, high flexibility, low cost and the like.

Description

technical field [0001] The invention relates to the field of artificial intelligence, in particular to a soft sensor modeling method based on the mLASSO-MLP model. Background technique [0002] With the development of science and technology, the industrial production process is becoming more and more complex, usually a multi-variable, nonlinear and strongly coupled system. In the process of industrial production, there are usually the following problems in the measurement of some key quality indicators: (1) the existing means can not be measured - can not be measured; (2) the measurement cost is high or the measurement time lag is long - the measurement is difficult. This will lead to some key quality variables are often difficult to measure automatically online. However, these important variables play an extremely important role in ensuring product quality and improving production efficiency, so they must be strictly monitored during industrial production. In order to mea...

AI Technical Summary

Problems solved by technology

In the process of industrial production, the measurement of some key quality indicators usually has the following problems: (1) the existing means can not be measured - can not be measured; (2) the measurement cost is high or the measurement time lag is long - the measurement is difficult

Its performance depends largely on the quality of the data and the structure of the model, however, in the implementation of ANN soft sensor technology, it always has two redundant problems, namely redundant input variables and redundant model structure

Redundant input variables will greatly affect the accuracy of the model, increase unnecessary model complexity, and affect the reliability of the model; while the redundant model structure may lead to overfitting, especially when the complexity of the model increases, which in Neural network model is a problem that cannot be ignored

[0004] In the existing literature and patent materials, many researchers have developed effective ANN model variable selection methods, and there are also many methods for neural network model structure optimization. These methods respectively solve the problem of input variable redundancy and model structure redundancy. However, for nonlinear systems, few studies consider both input variable redundancy and model structure redundancy

Images