Unscented Kalman Filter Observer-Based Model Predictive Control for Doubly-Fed Induction Generator

Abstract

A doubly-fed induction generator (DFIG) is among the primary actuators for wind energy generation due to merits, such as low converter cost, controllable power factor, reduced power losses, variable speed operation, constant grid frequency, maximal wind energy production, and active/reactive power control. The control and estimation problems of the DFIGs are of prime importance to be accurately and properly investigated. The recent advancement in microprocessor technology has led to the implementation of more sophisticated and reliable control schemes such as direct power control, direct torque control, sliding mode control and predictive control. For the DFIGs, the conventional control system often lacks to attain satisfactory results during the transient period due to nonlinearity and a highly coupled control system. Finite control set model predictive control (FCS-MPC) seems to be a very promising solution to control the active and reactive power and to regulate the switching states of rotor side converter due to its flexibility in defining the control objectives, improved dynamic performance and constraint handling. Apart from the inherently non-linear nature of the DFIG, what makes the problem particularly challenging is the unavailability of the measurements such as rotor speed and position. The use of sensors for determining the rotor speed and position not only increases the size, hardware complexity and maintenance cost of the DFIG systems but also decreases the system robustness. The researchers have developed various sensorless algorithms such as unscented Kalman filter, extended Kalman filter, model reference adaptive system, sliding mode observer, luenberger observer for the estimation of parameters. The previous algorithms are less efficient in a way that some of them cause inadequate results for highly non-linear systems and others are unable to operate in the low-speed range. The unscented Kalman filter (UKF) comes to the rescue to deal with the aforementioned issues and generate better results by the estimation of the parameters without linearization. This thesis presents a novel hybrid technique of Finite Control Set Model Predictive Control (FCS-MPC) with Unscented Kalman Filter (UKF) to a challenging control and estimation problem of DFIG. The proposed technique deals with the inherently non linear nature of DFIG and the unavailability of the measurements such as rotor speed and position. FCS-MPC is used for regulating the switching states of the rotor side converter. UKF is selected as an observer to estimate the dynamic states of DFIG. x The aforementioned proposed hybrid control scheme is thoroughly compared with the combination of FCS-MPC with Extended Kalman Filter (EKF) for performance analysis. The hybrid schemes, FSC-MPC-EKF and FCS-MPC-UKF have never been implemented in the literature for active and reactive power control and parametric estimation of speed and position of the DFIGs.For the implementation of the proposed scheme, the discrete-time dynamic model of the DFIG in the 𝑑𝑞-reference framework is used and the simulations are performed in MATLAB/Simulink environment. The proposed hybrid technique FCS-MPC-UKF shows better results compared to FCS MPC-EKF by accurately estimating the machine parameters, efficiently dealing with the issues of high non-linearities, and accurately calculating the active and reactive power of DFIG. The benefit of using FCS-MPC can be justified since the controller has demonstrated efficient results compared to controllers with the modulation phase. A detailed quantitative analysis is performed that shows the superiority of the proposed FCS-MPC-UKF algorithm compared to the FCS-MPC-EKF algorithm in the form of fewer current ripples, short peak time, improved settling time, controlled overshoot