Optimal PID controller parameters for first order and second order systems with time delay using a connectionist approach

• Published in: Engineering optimization Vol. 42; no. 3; pp. 295 - 303
• Main Authors: Madhuranthakam, C. R., Singh, J., Elkamel, A., Budman, H.
• Format: Journal Article
• Language: English
• Published: Abingdon Taylor & Francis 01.03.2010; Taylor & Francis Ltd
• Subjects: back propagation; Control systems; Controllers; Feedback control systems; FOPTD; Gain; Mathematical models; Neural networks; Optimization; Optimization algorithms; Parameter estimation; PID controller; process control; Proportional integral derivative; SOPTD; Time delay; Tuning
• ISSN: 0305-215X; 1029-0273
• DOI: 10.1080/03052150903196917

This article focuses on developing a neural controller based on a proportional integral derivative (PID) law. The main objective is to show that using neural networks represents a better alternative compared with other conventional models that were used in the past to express the tuning parameters as a function of process parameters such as process gain (K P ), process time constants (τ 1 , τ 2 , etc.,) and process time delay (θ). A Levenberg-Marquardt backward propagation algorithm is used to get the required PID parameters corresponding to different types of processes. In the present study, the PID parameters for the first order plus time delay (FOPTD) and the second order plus time delay (SOPTD) are obtained. It was observed that very high R 2 values between the actual PID parameters and the parameters obtained from the NN were achieved. Furthermore, the performance of PID control systems with FOPTD and SOPTD processes are applied on a number of case studies and compared with the conventional Zeigler-Nichols (Z-N) method of tuning PID controllers. Using the proposed NN controller was found to be efficient and the current method could be of potential use in control systems with gain scheduling also where the controller parameters are tuned according to a continuous gain schedule variable that changes based on different levels of process parameters.