High-performance modeling of bidirectional DC-DC converter using zebra optimization algorithm with deep learning for electric vehicles

Abstract Electric vehicles (EVs) make a significant contribution to the former and present transportation as electricity and global environmental concerns become worse quickly. With the advancing pervasiveness of EVs, more attention should be paid to battery energy storage systems. EVs depend on a battery to supply power. Fundamental needs (braking and acceleration) will worsen the battery, or it might be impossible to utilize it in certain cases, owing to the compactness of the battery. Power electronic systems, batteries, and electric motors are used for establishing the essential current balance. The DC-DC converters including a slew of interlocking features were introduced by the group. The DC-DC converters could enhance the voltage conversion effectiveness for DC and EV microgrid methods. EV batteries need a bi-directional DC-DC converter (BDCC) to take high-voltage power from the microgrid. The BDCC matches the voltage levels of a high-voltage traction machine and a low-voltage battery in an EV. Therefore, this study provides modeling of BDCC using a zebra optimization algorithm with deep learning (BDCC-ZOADL) technique for EVs. The BDCC-ZOADL technique involves two phases of operations for effective predictive modeling of BDCC on EVs. At the initial phase, the BDCC-ZOADL technique involves recalling enhanced recurrent neural networks (RERNNs) which allows to reproduction of the non-linear switching device characteristics. In addition, the RERNN model specifies the input and its output respective to the measurement of the input-output terminals (voltage and current) of the BDCC. In the second phase, the BDCC-ZOADL technique performs hyperparameter tuning of the RERNN model by the use of ZOA which in turn boosts the overall performance of the DL model. Simulation results show that the BDCC-ZOADL method achieves an RMSE of 0.0476 for current and 0.0017 for voltage, outperforming other models like LSTM-NN and NARX-NN with higher R 2 values (0.9995 and 0.9997). To illustrate the performance of the BDCC-ZOADL technique, a series of experimentations were involved. The simulation outcomes highlight the promising solution of the BDCC-ZOADL model over existing approaches.