Cascade control of superheated steam temperature with neuro-PID controller

Area of Science:

• Process Control
• Artificial Intelligence
• Thermodynamics

Background:

• Superheated processes require precise temperature control to ensure efficiency and safety.
• Traditional PID controllers may struggle with complex dynamics and disturbances in superheated systems.
• Existing neural network approaches often focus on minimizing squared error, potentially leading to temperature fluctuations.

Purpose of the Study:

• To develop an improved cascade control methodology for superheated processes.
• To implement a Proportional-Integral-Derivative (PID) controller using neural networks trained via an error entropy criterion.
• To combine feedback and feedforward control within the proposed neuro-PID controller.

Main Methods:

• Developed an improved cascade control methodology for superheated processes.
• Implemented a primary PID controller using neural networks trained by minimizing the error entropy criterion.
• Estimated tracking error entropy recursively using a receding horizon window technique.
• Integrated measurable disturbances as inputs to the neuro-PID controller for combined feedback and feedforward control.

Main Results:

• The proposed neuro-PID controller, trained with the error entropy criterion, demonstrated a reduction in superheated steam temperature fluctuations.
• Analysis of the neural network's convergent conditions was performed.
• Implementation procedures for the cascade control approach were summarized.
• Simulation examples validated the advantages of the proposed method over controllers minimizing squared error.

Conclusions:

• The developed neuro-PID controller offers enhanced performance for superheated processes by minimizing error entropy.
• The integration of feedback and feedforward control improves robustness against measurable disturbances.
• The proposed method provides a promising alternative for precise temperature regulation in industrial superheated applications.