Multi-layer Perception Neural Network Soft-sensor Modeling of Grinding Process Based on Swarm Intelligent Optimization Algorithms

In modern industrial production processes, hard instruments are often used to measure quality variables in industrial processes, but they are often subject to various economic or technical limitations and cannot successfully complete real-time monitoring tasks. In order to alleviate these problems,...

Full description

Saved in:
Bibliographic Details
Published in:Engineering letters 2024-03, Vol.32 (3), p.463
Main Authors: Shang-Guan, Yi-Peng, Xing, Cheng, Wang, Jie-Sheng, Sun, Yong-Cheng, Yang, Qing-Da
Format: Article
Language:English
Subjects:
Algorithms
Error functions
Grinding
Multilayers
Neural networks
Optimization
Optimization algorithms
Performance prediction
Prediction models
Process variables
Real time
Real variables
Sensors
Online Access:Get full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:In modern industrial production processes, hard instruments are often used to measure quality variables in industrial processes, but they are often subject to various economic or technical limitations and cannot successfully complete real-time monitoring tasks. In order to alleviate these problems, the soft-sensor technology is used to estimate the quality variables in real time, and a suitable prediction model is established. The process variables that can be directly measured are selected as input variables to estimate the target variables that are difficult to measure directly. Taking the iron content of ore, a key economic and technical index in the grinding process, as the prediction object, a MLP neural network soft-sensor model of grinding process based on swarm intelligent optimization algorithms is proposed. Based on the MLP neural network, the neural network is regarded as a nonlinear function. For the error function composed of nonlinear functions, an intelligent optimization algorithm is introduced to optimize it. After the optimized optimal model parameters, namely weights and thresholds, are re-assigned to the neural network, the neural network is used to predict the target variables. The simulation results indicate that the improved model exhibits enhanced generalization performance and predictive accuracy, making it capable of meeting the real-time control demands of the grinding and classification production process.
ISSN:1816-093X
1816-0948