Design of PID Controller Based on Echo State Network With Time-Varying Reservoir Parameter

Area of Science:

• Control Systems Engineering
• Computational Intelligence
• Machine Learning

Background:

• Proportional-Integral-Derivative (PID) controllers are widely used but challenging to tune for systems with time delays.
• Traditional tuning methods can be slow and may not achieve optimal performance for complex systems.

Purpose of the Study:

• To propose a new design method for optimizing PID controller parameters in discrete-time systems with time delays.
• To enhance the speed and accuracy of PID parameter determination using a novel echo state network approach.

Main Methods:

• A time-varying reservoir parameter echo state network (TVRP-ESN) was developed for PID controller parameter optimization.
• The method involves synchronous updating of output weights and reservoir parameters using the partial derivative of system output error.
• The TVRP-ESN leverages its learning and approximation capabilities for efficient parameter convergence.

Main Results:

• The proposed TVRP-ESN method demonstrated rapid acquisition of PID controller parameters.
• Synchronous updating of network parameters significantly improved the convergence speed for determining PID parameters.
• Effectiveness was validated through three simulation examples, showing successful control performance.

Conclusions:

• The TVRP-ESN offers an effective and efficient approach for optimizing PID controller parameters in discrete-time systems with time delays.
• This method provides a faster and potentially more accurate alternative to conventional PID tuning techniques.
• The study highlights the potential of echo state networks in advanced control system design.