Smart Neural Network Steers Dual-Input Converter for Electric Vehicles

Researchers at Vellore Institute of Technology in Chennai, India, have proposed a non-isolated bidirectional dual-input single-output (DISO) DC–DC converter paired with an intelligent energy management method for electric vehicle applications. The design integrates photovoltaic and battery energy sources within a single power conversion structure, which the authors say results in fewer components, higher efficiency and bidirectional power flow capability.

The converter is built on a two-phase interleaved boost topology with a relay-assisted reconfiguration stage. This arrangement facilitates flexible source selection, battery charging and discharging, and regenerative braking operation, allowing energy recovered during braking to be routed productively back into the battery rather than wasted.

At the heart of the energy management strategy is a Modified Deep Neural Network, or MODNN. Unlike a conventional single-task feedforward DNN, the proposed model uses a shared-trunk, multi-head architecture that jointly predicts two things at once: the discrete operating mode of the converter and the continuous duty ratios that control its power electronics. Training relies on a composite loss function combining classification and regression objectives.

The MODNN was benchmarked against recent machine-learning baselines, including k-nearest neighbours, a bagged neural-network ensemble and a long short-term memory (LSTM) network, all under identical training and test conditions. Inputs to the model include PV voltage, PV current, battery state of charge, battery voltage and load demand, which together determine the optimal operating mode and power flow under different driving scenarios.

According to the authors, the system provides flexible power distribution between available energy sources, productive use of regenerative braking energy, uninterrupted load supply and effective battery management. Five operating modes were defined to cover the range of conditions an EV powertrain might encounter, from solar-assisted driving to battery charging and braking recovery.

Performance was first verified through MATLAB/Simulink simulations, and the converter topology was then experimentally validated with a laboratory prototype rated at 500 watts. The experimental results showed that all five operating modes functioned successfully, with a peak measured efficiency of 96.7 percent — a figure the team presents as validation of the converter’s suitability for EV energy management.

The work, published in Frontiers in Artificial Intelligence in the journal’s Machine Learning section, sits at the intersection of power electronics and applied machine learning. Rather than using hard-coded control rules, the neural network adapts power sharing dynamically as conditions change, an approach the authors argue enhances energy utilization in vehicles that combine onboard solar generation with battery storage.

The study is an original research article describing a laboratory-scale prototype, so the reported efficiency and operating results reflect controlled bench conditions rather than road-tested vehicle performance. The authors also note, in the journal’s standard disclaimer, that all claims are solely those of the authors and do not necessarily represent those of their affiliated organizations or the publisher.

Nevertheless, the combination of a multiport converter architecture and a multi-task learning controller addresses a practical challenge: managing multiple energy sources without adding separate conversion stages for each. By merging PV and battery inputs into one structure and letting a learned controller decide mode and duty cycle, the design points toward more compact and efficient energy systems for electrified transport.

Subject of Research: A bidirectional dual-input DC–DC converter with a modified deep neural network energy management strategy for electric vehicles

Article Title: Intelligent energy management of a novel non-isolated bidirectional dual-input single-output converter for electric vehicle applications

Article References: M. R., F. B., & R., J. (2026). Intelligent energy management of a novel non-isolated bidirectional dual-input single-output converter for electric vehicle applications. Frontiers in Artificial Intelligence, 9, Article 1950048. https://doi.org/10.3389/frai.2026.1950048

Image Credits: AI Generated

DOI: 10.3389/frai.2026.1950048

Keywords: electric vehicles, DC-DC converter, energy management, deep neural network, photovoltaics, battery storage, regenerative braking, interleaved boost topology, power electronics, machine learning, Intelligent, energy