Artificial Neural Network-Based Adaptive Voltage Control of a Hybrid Zeta DC/DC Converter Using Multilayer Perceptron and Radial Basis Function Architectures With Back-Propagation Learning

This paper proposes the design and control of the output voltage of a Zeta hybrid DC/DC boost converter, an essential component for advancing clean energy technology, using an artificial neural network algorithm based on back-propagation (BP) learning method. The proposed artificial neural network control includes multilayer perceptron (MLP) and radial basis function (RBF) methods in both online and offline supervised modes. The first novelty of this control method involves four input cells comprising input voltage, reference output voltage, discrepancy between the output voltage and the reference value, and the accumulation of the output voltage error over time . The incorporation of the integral of output voltage error enables proportional-integral (PI) control capability in the proposed artificial neural network, utilizing input voltage for adaptive control capability. The neural network output consists of a single cell corresponding to the converter's duty cycle. In the proposed method, a recursive neural network is established through feedback signals of voltage. The second novelty is the incorporation of input voltage feedback within the ANN structure, enabling real-time adjustments and improved disturbance rejection. Additionally, an offline ANN control strategy using the Levenberg-Marquardt algorithm is proposed for accelerated convergence in buck mode operation. MATLAB simulations demonstrate that with the proposed control method, the output voltage of the Zeta hybrid DC/DC boost converter can accurately track its reference values. Even under disturbances and fluctuations in the voltage supplied to the converter and output load, the output voltage remains stable around its operating points. Furthermore, the converter's generalization capability is verified through arbitrary changes in input voltage values. Calculated LMS error for reference DC voltage disturbance using the BP offline, BP online, RBF online, and PI algorithms, are achieved 0.05, 0.1, 0.16, and 2.4, respectively. Overall, the proposed neural network control method shows promising results in maintaining stable output voltage performance for the Zeta hybrid DC/DC boost converter, validating its effectiveness under varying operational conditions.