Real-time system identification and self-tuning control of DC-DC power converter using Kalman Filter approach

Abstract

Switch-mode power converters (SMPCs) are employed in many industrial and consumer devices. Due to the continuous reduction in cost of microprocessors, and improvements in the processing power, digital control solutions for SMPCs have become a viable alternative to traditional analogue controllers. However, in order to achieve high-performance control of modern DC-DC converters, using direct digital design techniques, an accurate discrete model of the converter is necessary. This model can be acquired by means of prior knowledge about the system parameters or using system identification methods. For the best performance of the designed controller, the system identification methods are preferred to handle the model uncertainties such as component variations and load changes. This process is called indirect adaptive control, where the model is estimated from input and output data using a recursive algorithm and the controller parameters are tuned and adjusted accordingly. In the parameter estimation step, Recursive Least Squares (RLS) method and its modifications exhibit very good identification metrics (fast convergence rate, accurate estimate, and small prediction error) during steady-state operation. However, in real-time implementation, the accuracy of the estimated model using the RLS algorithm is affected by measurement noise. Moreover, there is a need to continuously inject an excitation signal to avoid estimator wind-up. In addition, the computational complexity of RLS algorithm is high which demands significant hardware resources and hence increase the overall cost of the digital system. For these reasons, this thesis presents a robust parametric identification method, which has the ability to provide accurate estimation and computationally efficient self-tuning controller suitable for real-time implementation in SMPCs systems. This thesis presents two complete real-time solutions for parametric system identification and explicit self-tuning control for SMPCs. The first is a new parametric estimation method, based on a state of the art Kalman Filter (KF) algorithm to estimate the discrete model of a synchronous DC-DC buck converter. The proposed method can accurately identify the discrete coefficients of the DC-DC converter. This estimator possesses the advantage of providing an independent strategy for adaptation of each individual parameter; thus offering a robust and reliable solution for real-time parameter estimation. To improve the tracking performance of the proposed KF, an adaptive tuning technique is proposed. Unlike many other published schemes, this approach offers the unique advantage of updating the parameter vector coefficients at different rates. This thesis also validates the performance of the identification algorithm with time-varying parameters; such as an abrupt load change. Furthermore, the proposed method demonstrates robust estimation with and without an excitation signal, which makes it very well suited for real-time power electronic control applications. Additionally, the estimator convergence time is significantly shorter compared to many other schemes, such as the classical Exponentially weighted Recursive Least Square (ERLS) method. To design a computationally efficient self-tuning controller for DC-DC SMPCs, the second part of the thesis develops a complete package for real-time explicit self-tuning control. The novel partial update KF (PUKF) is introduced for real-time parameter estimation. In this approach, a significant complexity reduction is attained as the number of arithmetic operations are reduced, more specifically the computation of adaptation gains and covariance updates. The explicit self-tuning control scheme is constructed via integrating the developed PUKF with low complexity control algorithm such as Bányász/Keviczky PID controller. Experimental and simulation results clearly show an enhancement in the overall dynamic performance of the closed loop control system compared to the conventional PID controller designed based on a pre-calculated average model. Importantly, in this thesis, unlike a significant proportion of existing literature, the entire system identification, and closed loop control process is seamlessly implemented in real-time hardware, without any remote intermediate post processing analysis.