Adaptive Kalman-Based Constrained Predictive Control With Neural Estimator for a Noninverting Buck–Boost Converter

Abstract

This article introduces an adaptive constrained model predictive control (AMPC) method for regulating the voltage of a noninverting dc buck–boost converter, capable of delivering up to 48 W output. The approach incorporates constraints on the control signal and its variations to minimize oscillations in both input current and output voltage. The AMPC controller employs a linear model, adaptively estimated via an online Kalman-based recursive least squares algorithm. To efficiently manage the computational demands of the AMPC algorithm, a dynamic neural network (DNN), trained using AMPC controller data, is utilized for control within a specific range of the output voltage's steady-state response. A constrained control variable tuning mechanism has been applied to the output of the DNN to reduce the oscillations of the steady-state response more efficiently. Experimental tests have been conducted to assess performance under varying conditions of reference voltage, load, and input voltage. Notably, the fluctuations in output voltage are lower compared to the basic AMPC, another constrained model predictive control, and a PI method. More specifically, for the proposed method, the output voltage fluctuation is about 72%, the calculation time is about 75%, and the minimum energy loss of the switch is about 8.5–10% average less than the basic AMPC.