A soft sensing method of billet surface temperature based on ILGSSA-LSSVM

Area of Science:

• Materials Science and Engineering
• Computational Science
• Industrial Process Control

Background:

• Accurate surface temperature measurement of continuous casting billets is crucial for quality control but remains a significant challenge.
• Lack of reliable temperature feedback parameters hinders scientific control over billet quality during continuous casting.

Purpose of the Study:

• To develop an advanced soft sensing model for predicting the surface temperature of continuous casting billets.
• To enhance the prediction accuracy and reliability of billet quality control through precise temperature estimation.

Main Methods:

• Optimization of the Least Square Support Vector Machine (LSSVM) model using a Sparrow Search Algorithm.
• Integration of Logistic Chaotic Mapping and Golden Sine Algorithm (ILGSSA) to optimize LSSVM parameters (penalty factor and kernel parameter).
• Global optimization approach to determine the optimal parameter combination for improved prediction accuracy.

Main Results:

• The proposed ILGSSA-LSSVM model demonstrated superior performance compared to traditional LSSVM, BP neural networks, and Gray Wolf optimized LSSVM.
• Experimental results showed minimal prediction errors, with a maximum error of 3.85733 °C, a minimum error of 0.0174 °C, and an average error of 0.05805 °C.
• The optimized model significantly reduced the negative impact of random parameter initialization on prediction accuracy.

Conclusions:

• The ILGSSA-LSSVM soft sensing model provides a highly accurate and reliable method for predicting continuous casting billet surface temperature.
• This advancement offers crucial feedback parameters for scientific control, leading to enhanced billet quality and optimized industrial processes.