Certain investigations on applications of machine intelligence techniques to fractional order PID FOPID controller design

Abstract

Recently, Fractional Order Proportional Integral Derivative (FOPID) newlinecontrollers have been proven to outperform PID controllers due to their design newlineflexibility, greater relative stability with zero steady-state error, and less newlinesensitivity to high-frequency noises. Design of centralized FOPID controller is newlineproposed in this research and also, a FOPID tuner for determining controller newlineparameters for any system is formulated using machine intelligence techniques. newlineThe main drawback of the FOPID controller is the task of choosing the five newlinecontroller parameters. To make the design of the FOPID controller easier and newlineaccurate, a universal, Machine Learning based Optimal Tuner (MLOT) is newlineproposed for Fractional Order Proportional Integral Derivative (FOPID) newlinecontrollers. The proposed tuning approach overcomes the drawbacks of available newlineFOPID tuning rules such as interpolated curve fitting, a mathematical model in newlineterms of process parameters, need for an accurate initial model. This MLOT is newlinedevised using a non-interpolated, diversified dataset of optimal FOPID controller newlineparameters. newlineMachine intelligence techniques, Feed Forward Back Propagation newlineNeural Network (FFBPNN), Multi Least Squares Support Vector Regression newline(MLSSVR), are applied for designing the MLOTs owing to large newlineinterdependency among the controller parameters where linear/multivariate newlineregression models become inefficient. The proposed methodology is validated newlineon two laboratory experiments, a first-order single tank water level control newlinesystem and a second-order two-tank water level control system using newlineLABVIEW®. The FOPID controller performance designed using MLOT is newlinecompared with the performance of the FOPID controller designed using newlinepreviously available tuning rules. For both laboratory experiments, the proposed newlineMLOTs provide good performances in terms of improved robustness, faster newlinesettling time, and minimum Integrated Absolute Error-values when compared to newlineexisting FOPID tuning rules. newline