New Algorithm Enables Real-Time OCV-SoC Mapping for EV Batteries
In a significant leap forward for electric vehicle (EV) battery management, researchers have developed a novel method to rapidly and accurately extract the open-circuit voltage (OCV) to state-of-charge (SoC) relationship—critical for precise battery monitoring—without the need for time-consuming traditional testing procedures. This breakthrough, detailed in a recent study published in the Journal of Tongji University (Natural Science), promises to enhance the safety, efficiency, and longevity of lithium-ion batteries used in everything from passenger EVs to commercial fleets.
The conventional approach to determining the OCV-SoC curve—a fundamental input for battery management systems (BMS)—has long been a bottleneck in real-world EV operations. Historically, engineers relied on incremental OCV (IO) or low-current OCV (LO) tests that require the battery to rest for extended periods between small charge or discharge steps to reach electrochemical equilibrium. Due to hysteresis effects inherent in lithium-ion chemistries, these protocols can take several days to complete a single full cycle. Moreover, because the OCV-SoC relationship drifts as the battery ages, this laborious calibration must be repeated periodically throughout the battery’s lifespan—making it impractical for in-service vehicles.
The newly proposed method sidesteps these limitations by leveraging a key physical insight: during normal discharge, the true OCV curve evolves smoothly over time. This “smoothness hypothesis” forms the cornerstone of an innovative optimization framework that can reconstruct the OCV-SoC mapping from arbitrary, real-world current-voltage data—such as those recorded during urban driving cycles—without requiring any rest periods or controlled laboratory conditions.
Led by Jinwei Xue from the Shanghai Automotive Wind Tunnel Center at Tongji University, the research team combined a first-order equivalent circuit model (ECM) with a powerful multi-objective evolutionary algorithm known as NSGA-II (Non-dominated Sorting Genetic Algorithm II). The ECM simulates the dynamic voltage response of the battery under load, decomposing the measured terminal voltage into contributions from ohmic resistance, polarization effects, and the underlying OCV. However, without prior knowledge of the ECM parameters (such as internal resistance and RC time constants), direct calculation of OCV yields a noisy, erratic curve.
Here’s where the smoothness assumption proves transformative. The researchers observed that only when the ECM parameters are correctly identified does the reconstructed OCV-t curve exhibit the expected monotonic and smooth behavior. Incorrect parameters produce artificial oscillations and discontinuities. By defining a smoothness metric—essentially quantifying the total variation of the OCV increments over time—they turned parameter identification into an optimization problem: find the set of ECM parameters that minimizes this metric, thereby yielding the smoothest possible OCV trajectory consistent with the measured data.
NSGA-II, renowned for its robustness in navigating complex, non-convex search spaces, was employed to solve this problem efficiently. The algorithm iteratively evolves a population of candidate parameter sets, using principles inspired by natural selection—crossover, mutation, and non-dominated sorting—to converge toward optimal solutions. Crucially, the method does not require any prior calibration data or assumptions about battery chemistry beyond the general smoothness of OCV during discharge.
To validate their approach, the team used real-world battery aging data from Stanford University’s Energy Control Lab, featuring INR21700-M50T cells with graphite/silicon anodes and NMC cathodes, cycled under the Urban Dynamometer Driving Schedule (UDDS)—a standard test cycle mimicking city driving patterns. Despite the absence of ground-truth OCV or SoC labels in the dataset, the researchers used Coulomb counting (ampere-hour integration) as a reference for SoC, anchored at the discharge cutoff point.
The results were striking. The optimized ECM parameters not only produced a remarkably smooth OCV-t curve but also aligned closely with independent electrochemical impedance spectroscopy (EIS) measurements in the relevant frequency band (0–0.2 Hz), confirming the physical plausibility of the identified parameters. Further refinement using an eighth-order polynomial fit yielded a clean, continuous OCV-SoC curve ready for integration into estimation algorithms.
The ultimate test came in the form of SoC estimation using an Extended Kalman Filter (EKF)—a standard technique in BMS that fuses model predictions with real-time voltage measurements. When initialized with the extracted OCV-SoC curve, the EKF achieved a maximum estimation error of just 2% over an entire UDDS discharge cycle. More impressively, the method demonstrated exceptional robustness to initialization errors. In a stress test, researchers artificially reset the EKF’s internal SoC estimate to 80% midway through discharge—introducing a 40-percentage-point error. Within approximately 1,000 seconds (less than 17 minutes of driving), the filter converged back to the true SoC trajectory, showcasing the self-correcting power of the accurately extracted OCV relationship.
This resilience is a game-changer for real-world applications. In practice, BMS can suffer from sensor drift, communication faults, or sudden disturbances that corrupt the SoC estimate. Traditional systems relying on outdated or generic OCV-SoC curves may never recover accurately from such errors, leading to range anxiety or even safety risks from overcharge/overdischarge. The new method ensures that the core OCV-SoC mapping remains accurate and adaptive, enabling rapid error correction even under adverse conditions.
Moreover, the technique is inherently scalable and deployable. It requires only standard voltage and current telemetry—data already collected by every modern BMS—and runs on conventional embedded processors. No specialized equipment, rest periods, or offline lab testing are needed. This makes it ideal for over-the-air (OTA) updates or continuous in-vehicle calibration, allowing the BMS to autonomously refine its OCV-SoC model as the battery ages.
From a systems perspective, accurate and up-to-date SoC estimation unlocks multiple benefits. It enables more aggressive regenerative braking strategies without fear of overcharging, optimizes thermal management by anticipating high-current demands, and extends battery life by preventing operation in high-stress voltage regions. For fleet operators and OEMs, this translates to lower total cost of ownership, improved vehicle reliability, and enhanced customer satisfaction.
The implications extend beyond passenger EVs. Commercial electric trucks, buses, and even grid-scale energy storage systems—all reliant on precise SoC tracking—could benefit from this methodology. As global EV adoption accelerates and battery chemistries continue to evolve (e.g., silicon-anode, solid-state), the need for flexible, chemistry-agnostic calibration techniques becomes ever more pressing. This smoothness-based approach offers a universal framework adaptable to diverse cell formats and aging mechanisms.
Critically, the research addresses a long-standing challenge in battery modeling: the coupling between OCV identification and other ECM parameters. Previous attempts to estimate OCV online often suffered from identifiability issues, where errors in resistance or capacitance estimates propagated into OCV inaccuracies. By anchoring the solution in a physically meaningful smoothness criterion, the new method decouples these uncertainties, yielding more stable and reliable results.
Looking ahead, the team envisions integrating this OCV extraction technique with advanced state-of-health (SoH) estimators. Since OCV-SoC curves shift characteristically with capacity fade and impedance growth, continuous monitoring of these changes could provide early warnings of degradation. Combined with machine learning models trained on aging datasets, such a system could predict remaining useful life with unprecedented accuracy.
The automotive industry is already taking note. With regulatory pressures mounting for longer battery warranties and stricter safety standards, OEMs are investing heavily in next-generation BMS capabilities. Techniques like this—born from academic rigor but engineered for real-world robustness—represent the bridge between theoretical battery science and practical vehicle engineering.
In summary, this work redefines what’s possible in battery state estimation. By replacing days of lab testing with minutes of computational optimization on real driving data, it brings high-fidelity OCV-SoC mapping into the realm of real-time, in-vehicle operation. For drivers, this means more accurate range predictions and longer-lasting batteries. For manufacturers, it means smarter, safer, and more efficient electric vehicles. And for the broader energy transition, it removes a critical technical barrier to mass EV adoption.
As electric mobility continues its inexorable rise, innovations like this—quietly operating in the background of every battery pack—will be just as vital as the headline-grabbing advances in cell chemistry or charging speed. After all, the true measure of a battery isn’t just how much energy it stores, but how well we understand and manage that energy at every moment of its life.
Jinwei Xue¹, Xuzhi Du², Zhigang Yang³, Lei Zhao¹, Chao Xia⁴. “Extraction of Open Circuit Voltage-State of Charge Curve for Lithium-Ion Batteries Based on Smoothness Optimization.” Journal of Tongji University (Natural Science), Vol. 52, No. S1, Oct. 2024. DOI: 10.11908/j.issn.0253-374x.24778.
¹Shanghai Automotive Wind Tunnel Center, Tongji University, Shanghai 201804, China
²Department of Mechanical Science and Engineering, University of Illinois Urbana-Champaign, Urbana, IL 61801, USA
³COMAC Beijing Aircraft Technology Research Institute, Beijing 102211, China
⁴School of Automotive Studies, Tongji University, Shanghai 201804, China