Study on a Predictive Battery Charging Control System for Series Hybrid Electric Vehicles Using Gradient Information

Yuhi Abe･Takahiro Shimizu･Hyuga Cyujo･Mai Shinkawa･Jin Kusaka (Waseda University)

This study proposes a predictive battery charging control system for series hybrid electric vehicles using only road gradient information. The system actively uses battery energy during uphill driving to maximize regenerative energy recovery during subsequent downhill driving. A simulation model identified from real-world driving data was used for evaluation on a suburban route with large elevation changes. The results confirmed a significant increase in regenerated energy and an improvement in fuel efficiency, demonstrating the effectiveness of gradient-based predictive control without requiring uncertain velocity predictions.

Efficient Kalman Filter-Based Battery Pack SOC Estimation Using Bar Delta Filtering

Seyedmehdi Hosseininasab･Changwei Lin･Lennart Bauer (FEV Europe)･Tsuyoshi Horiba･Yasutaka Ikawa (FEV Japan)

Accurate estimation of SOC in lithium-ion battery packs is essential for safe reliable and optimized BMS in electric vehicles and energy storage. This study presents an efficient framework combining a lightweight Kalman filter with the Bar-Delta method for series-connected LFP cells. A simulation model with realistic cell-to-cell variations evaluates performance. The method estimates pack average SOC and cell deviations with lower computational cost. Results show comparable accuracy to full Kalman filtering while significantly reducing computation, enabling real-time embedded applications. It is suitable for resource-constrained hardware and scalable pack monitoring deployments today.

Comparison of IR, EKF, and NARX Estimators for Battery SOC and Voltage Prediction Across EV Pack Sizes

Alessandro Massimo Bertucci･Xinwei Li･Ratnak Sok (Waseda University)･Keiki Tanabe･Goro Iijima (Mitsubishi Fuso Truck and Bus)･Jin Kusaka (Waseda University)

This study compares Internal Resistance (IR), Extended Kalman Filter (EKF), and Nonlinear Autoregressive with Exogenous Inputs (NARX) estimators for battery state-of-charge and terminal-voltage prediction across S, M, and L electric-vehicle battery pack configurations. On regular-temperature JE05 measured logger data, IR and EKF achieved similar SOC accuracy but poor voltage reproduction, especially for the L configuration. NARX achieved higher joint SOC and voltage accuracy on the same data. Cold-chamber NARX results were separated by driving cycle, showing strong performance for 50 km/h and 80 km/h cases, while JE05 at -25 degC remained the most difficult voltage condition.

Battery Load Prediction in Market Environments Using Big Data and Generative AI

Yuya Hato (Mazda)

When lithium-ion batteries are used in electric vehicles, it is necessary to ensure reliability against degradation and thermal runaway, considering real-world customer usage patterns. Therefore, by applying generative AI to market big data, we developed a method to generate driving scenarios that replicate customer usage characteristics. We then developed a method to evaluate battery load intensity under market conditions by combining the generated scenarios with a one-dimensional model.