Abstract
Closed-loop feedback controllers are commonly used for temperature regulation in the field of process control. Designed for stability and reliability, feedback controllers maintain process stability with minimal deviation from target values. However, feedback regulators face limitations in complex scenarios, as they rely solely on real-time data without learning from past historical information. Recent research on Sequence-to-Sequence machine learning models with attention mechanisms offer promising solutions for more adaptive, dynamic temperature control. The work presented in this project develops a tailored machine learning model based on scaled dot-product attention to predict temperature control signals from error values, using simulated temperature control data. The performance of an attention-based model was compared to a baseline sequence-to-sequence model, demonstrating improved alignment with target control signal and highlighting its potential for effective temperature regulation.