Process regulation control using Echo State Networks: an ESN-based deep neural network approach for PID control

In automation systems, Proportional-Integral-Derivative (PID) controllers are extensively used for their simplicity, reliability, and ability to maintain desired process levels by continuously adjusting control inputs to minimize the error between the set-point and the actual process variable. Despite the advantages, they have to face challenges such as handling non-linearities, sensitivity to parameter tuning, limited adaptability to changing process dynamics, and susceptibility to disturbances. For these reasons, various advanced control strategies and adaptive algorithms based on deep neural network are being developed and implemented. The aim of the proposed study is to introduce an innovative method for the regulation of processes, using an ESN-based deep neural network model. As preliminary results a case study on a heating system is considered. The model was trained using data generated under the control of a PID controller and tested on a using a TCLab module. The results indicate that the trained model effectively predicts the outputs of the heating system and the corresponding temperature values in alignment with the target temperatures. This study would be a starting point for a future deep methodological exploration of the mathematical aspects of ESN models to ensure the long-term stability of the system and its limitations.