System identification-based tuning of PID algorithm parameters by using PSO and genetic algorithms

The PID control algorithm is the most used industrial control method. This is due to its simplicity, ease of use, and because it yields stable results. PID parameters vary for each controller; therefore, finding the optimal parameters is crucial for obtaining good results. However, tuning PID parameters for complex systems is not a trivial task and is also somewhat difficult as conventional techniques rely on time and effort manual tuning or rely on simplified statistical estimation techniques of the system’s model yielding sub- optimal results. In this project, we propose to use machine learning for PID parameter tuning on proprietary historical time series operating process control data. The data is processed with the help of computational and machine-learning techniques to better identify the process model and predict the optimal PID parameters. The research methodology consists of three main steps. First, process-model identification is done by using a Radial Basis Function (RBF) neural network. Secondly, Particle Swarm Optimization (PSO) hybrid with Genetic Algorithms (GA) is used for finding the optimal PID parameters. The PID values predicted by PSO, will be fed to GA optimization process as an initial starting point. The final step consists of integrating the identified process model with the PID optimization algorithm in a computer-based simulation environment (Simulink). The experimental simulations are done for various study cases. Results showed that the predicted PID parameters error rate and standard deviation for PID control and process are decreased, which enhances the process controlling and stability.