This paper comprehensively reviews bidirectional DC-DC converters and their control techniques for Electric Vehicle applications. A classification is proposed based on the three power ratings levels of the SAEJ1772 standard. Circuit topologies are compared based on power rating, switching frequency, static voltage gain, operating modes, number of components and power switch material. High switching-frequency topologies that include emerging Wide Bandgap (WBG) devices are also discussed. Cost comparison of traditional Si power switches and WBG devices employed in the reviewed topologies is also included. Traditional control methods for power converters in EVs are presented while also considering the emergence of Artificial Intelligence algorithms applied in systems controls that offer alternative methods to improve the efficiency of bidirectional DC-DC converters.

Recently, it has been produced an accelerated presence of non-linear elements in the residential, commercial and industrial sectors, which has affected the voltage waveform, altering its amplitude and frequency. These sectors obtain the electric energy from distribution networks, so it is essential to check them systematically in order to ensure the power quality according to operational guidelines and national standards. Power quality involves a wide variety of electromagnetic phenomena on the power system, highlighting the harmonics. This paper presents an application based on the k-means clustering to identify similarities in the behavior of the harmonic distortion of voltage and current in distribution networks and its validation and refinement using the rough sets theory. The main result lies in the characterization of those groups of distribution networks by means of their centroids and according to the IEEE Standard 519-2014. To verify the methodology, field measurements of a group of distribution networks in Cuba are analyzed.

This paper presents the design and embedded implementation of a robust controller for the transformerless multilevel active monophase rectifier. In order to reduce the effects caused by the uncertainty originated by the output load, an algebraic estimator is devised. Then, a linear controller based on the Exact Static Error Dynamics Passive Output Feedback (ESEDPOF) is proposed, where the uncertainty estimation is taken into account. Since the controller estimator is based on the continuous time plant model, its real-time implementation in a digital platform requires a discretization of the controller under sufficiently fast sampling, such that the properties of the closed loop nonlinear sampled data system are preserved. For this reason, the medium-scale FPGA Spartan-6 XC6SLX16 is used for implementing the ESEDPOF controller, the online algebraic estimator, the enhanced phase locked loop (EPLL), and the multilevel PWM. The parallel processing provided by these devices and the capability in the design of custom modules, allow optimizing the hardware description and obtaining an update time for the control law of 9.683 micro seconds. Experimental validation shows an excellent dynamical performance and a nearunit power factor.

This study undertakes a comprehensive review of control techniques applicable to DC-DC power converters, categorized into Traditional Control (TC) methods and those based on Artificial Intelligence (AI). Succinct descriptions of prevalent strategies in both classifications are furnished, shedding light on their fundamental principles. Further, the current progress in the field is evaluated, anchoring the discussion in the provided categorization. In assessing the merits and potential drawbacks of each method, specific emphasis is laid on the target converter topology. Predominant topologies such as the elementary buck, boost, bidirectional buck-boost, and dual-activebridge (DAB) are scrutinized. To furnish a thorough analysis and facilitate comparison of principal control methods, simulations of four fundamental off-the-shelf algorithms are undertaken, employing a 1 MHz switching frequency.

Los vehículos eléctricos presentan una alternativa viable para reducir las emisiones de gases tóxicos en las concentraciones urbanas y para disminuir los efectos de los gases de invernadero. La batería de los vehículos eléctricos debe ser monitoreada con precisión para asegurar su funcionamiento adecuado y seguro. Para esto, es necesario desarrollar algoritmos eficientes que permitan estimar de forma precisa el estado de carga mediante dispositivos embarcados en el vehículo. En este trabajo, se utiliza un conjunto de datos de ciclado de una batería de Litio para entrenar una red neuronal para la estimación del estado de carga. Se realiza una optimización bayesiana para establecer la mejor arquitectura de red neuronal y se valida el comportamiento frente a las mediciones reales que ofrece el conjunto de datos. Para su utilización en un dispositivo embarcado, la red neuronal se valida con un modelo de hardware-en-lazo (HIL) en un FPGA con aritmética de punto fijo. Después del entrenamiento se observa un error promedio cuadrático menor al 2% y una precisión promedio del 97.5%.