New Adaptive Algorithm Boosts EV Battery State Estimation Accuracy
In the rapidly evolving landscape of electric mobility, where range anxiety and battery longevity remain top concerns for consumers and manufacturers alike, a breakthrough in battery management system (BMS) technology has emerged from Chongqing University of Technology. A research team led by Dr. Yingzhe Kan has introduced a novel adaptive algorithm that significantly enhances the accuracy of lithium-ion battery state-of-charge (SOC) estimation—a critical parameter for vehicle performance, energy efficiency, and user experience.
Published in the Journal of Chongqing University of Technology (Natural Science), the study presents an innovative fusion of real-time parameter identification and state estimation techniques designed to overcome the limitations of conventional methods. As electric vehicles (EVs) operate under diverse driving conditions—from city commutes to highway cruising—the internal characteristics of their batteries dynamically shift. Traditional battery models, which rely on fixed parameters derived from offline testing, often fail to capture these transient behaviors, leading to inaccuracies in SOC prediction. This discrepancy not only undermines driver confidence but also limits the effectiveness of energy management strategies.
Dr. Kan, an expert in new energy vehicles and advanced powertrain systems, emphasized the importance of adapting to real-world variability. “Battery behavior is inherently nonlinear and time-varying,” he explained. “Using static models in dynamic environments leads to accumulating errors. Our goal was to develop a method that continuously learns and adjusts, ensuring the BMS always operates with up-to-date information about the battery’s condition.”
The core innovation lies in the integration of a self-regulating forgetting factor within a recursive least squares (RLS) framework—termed Adaptive Forgetting Factor Recursive Least Squares (AFFRLS)—combined with the Extended Kalman Filter (EKF) for SOC estimation. Unlike traditional RLS methods that apply a fixed forgetting factor, this new approach dynamically tunes the factor based on real-time prediction errors. When the model’s output deviates significantly from actual measurements, the algorithm reduces the influence of older data, placing greater emphasis on recent observations. Conversely, when predictions are stable and accurate, it maintains a higher memory of historical data, balancing responsiveness with stability.
This adaptive mechanism allows the system to respond swiftly to sudden changes in load, temperature, or aging effects, all of which impact battery impedance, capacitance, and open-circuit voltage (OCV). By continuously recalibrating the equivalent circuit model parameters—specifically the ohmic resistance, polarization resistances, and capacitances—the AFFRLS-EKF method ensures that the underlying model remains representative of the battery’s current physical state.
The research team, including graduate researcher Min Yang and colleagues Huaze Sun and Yunfei Xie, conducted extensive experimental validation using a 19Ah ternary lithium-ion cell under multiple driving cycles. These included the Hybrid Pulse Power Characterization (HPPC) test, the Dynamic Stress Test (DST), and intermittent constant-current discharge scenarios—each designed to simulate different aspects of real-world driving, from aggressive acceleration to steady cruising.
Results demonstrated exceptional performance across all test conditions. Under the HPPC cycle, the maximum SOC estimation error remained below 0.65%, while the DST and intermittent discharge tests yielded errors of 0.55% and 0.57%, respectively. In every case, the error stayed within 1%, a significant improvement over conventional RLS-EKF methods, which exhibited errors as high as 2% under similar conditions. This level of precision surpasses industry benchmarks and brings EVs closer to the ideal of “truthful” range prediction—a key factor in consumer adoption.
One of the most compelling aspects of the new algorithm is its ability to maintain high accuracy without sacrificing computational efficiency. Many advanced estimation techniques, such as particle filters or deep learning models, offer improved precision at the cost of increased processing demands—making them impractical for embedded BMS hardware. The AFFRLS-EKF method, however, operates efficiently on standard microcontrollers, enabling real-time implementation without requiring high-end computing resources.
Yang, whose research focuses on battery dynamics and control systems, highlighted the practical implications. “What sets this method apart is its balance between accuracy and feasibility. It doesn’t require massive datasets or cloud connectivity. Everything happens onboard, in real time, using only voltage and current measurements that are already available in every EV.”
The dual-polarization equivalent circuit model used in the study strikes a careful balance between complexity and fidelity. While more elaborate models with multiple RC networks can offer higher precision, they come with increased computational overhead and parameter interdependence, complicating identification. The dual-RC structure captures the essential transient and diffusion-related voltage drops while remaining tractable for online estimation.
Crucial to the success of any SOC estimator is the accuracy of the SOC-OCV relationship. The team performed detailed Hybrid Pulse Power Characterization testing at 25°C, collecting voltage relaxation data after each 10% SOC decrement. Using seventh-order polynomial fitting, they established a precise mapping between SOC and open-circuit voltage—a foundational element that anchors the entire estimation process.
During operation, the algorithm functions in a closed loop: the AFFRLS module continuously updates the model parameters using incoming voltage and current data; these updated parameters are then fed into the EKF, which computes the current SOC estimate. The EKF accounts for both process noise (uncertainties in the system dynamics) and measurement noise (sensor inaccuracies), refining its estimate with each iteration.
This feedback architecture creates a self-correcting system. If sensor drift or model degradation begins to affect accuracy, the parameter identification stage detects the anomaly and adjusts accordingly. Over time, this enables the BMS to compensate not only for operational transients but also for long-term aging effects such as capacity fade and internal resistance growth—key indicators of battery health.
The implications extend beyond mere SOC accuracy. With more reliable state estimation, vehicle manufacturers can optimize thermal management, extend battery life through gentler charging protocols, and improve regenerative braking efficiency. Accurate SOC data also enables smarter preconditioning of the battery before fast charging, reducing stress and improving charge acceptance rates.
Moreover, precise SOC estimation enhances safety. Overestimating remaining charge can lead to unexpected shutdowns, while underestimation may cause unnecessary limitations on performance. In extreme cases, inaccurate state monitoring can contribute to overcharging or deep discharging—conditions that accelerate degradation and increase the risk of thermal runaway.
From a systems engineering perspective, the AFFRLS-EKF method exemplifies the trend toward intelligent, adaptive control in automotive electronics. As vehicles become more connected and autonomous, the demand for self-aware, self-tuning components grows. This algorithm represents a step toward truly cognitive battery systems—ones that learn from experience and adapt to individual usage patterns.
The research also underscores the importance of interdisciplinary collaboration. The team combined expertise in mechanical engineering, control theory, electrochemistry, and signal processing to tackle a problem that sits at the intersection of physics and computation. Their work reflects a broader shift in automotive R&D, where breakthroughs increasingly emerge from the convergence of traditionally separate domains.
Industry response to the findings has been cautiously optimistic. While the method has been validated in laboratory settings, real-world deployment will require extensive durability testing under diverse environmental conditions—including extreme temperatures, humidity, and vibration. Additionally, integration with existing BMS architectures and compliance with automotive functional safety standards (such as ISO 26262) will be necessary before widespread adoption.
Nevertheless, the potential benefits are substantial. For consumers, it means more trustworthy range displays, longer battery life, and improved driving confidence. For automakers, it offers a pathway to reduce battery oversizing—a common practice used to compensate for estimation uncertainty—thereby lowering vehicle weight and cost. For fleet operators, accurate SOC tracking enables better route planning and charging scheduling, improving operational efficiency.
The method’s adaptability also makes it suitable for second-life applications, such as stationary energy storage, where batteries from retired EVs are repurposed. In these scenarios, the ability to accurately assess the state of aged and heterogeneous battery packs is crucial for safety and performance.
Looking ahead, the research team is exploring extensions of the algorithm to estimate additional battery states, including state-of-health (SOH) and state-of-energy (SOE). Preliminary work suggests that the same adaptive identification framework can be applied to track capacity fade and internal resistance growth over time, providing a comprehensive picture of battery condition.
Another area of investigation involves temperature compensation. Battery behavior varies significantly with temperature, and future versions of the algorithm may incorporate real-time thermal modeling to maintain accuracy across a wide operating range—from sub-zero winters to scorching summers.
The publication of this research in a peer-reviewed journal marks a significant milestone, but the journey from academic discovery to commercial product is often long and complex. The team has filed preliminary intellectual property disclosures and is in discussions with several automotive suppliers about potential collaboration.
As global EV sales continue to rise—surpassing 14 million units in 2023 according to industry analysts—the need for smarter, more efficient battery management will only intensify. Innovations like the one developed at Chongqing University of Technology are not just technical achievements; they are enablers of a sustainable transportation future.
In an era where software increasingly defines vehicle capability, this adaptive algorithm exemplifies how intelligent algorithms can extract maximum value from existing hardware. It is a reminder that sometimes, the most impactful advances are not in building bigger batteries, but in understanding them better.
The success of the AFFRLS-EKF method also highlights the growing role of Chinese institutions in advancing EV technology. With strong government support for new energy vehicles and a rapidly expanding domestic market, research centers across China are making significant contributions to battery science, motor control, and charging infrastructure.
Dr. Kan expressed optimism about the broader impact of the work. “We are not just improving a number on a dashboard. We are building trust between the driver and the machine. When people see that their car’s range prediction is consistently accurate, they begin to rely on it. That trust is essential for the mass adoption of electric vehicles.”
As the automotive industry transitions from internal combustion engines to electrified powertrains, the role of the battery will only grow in importance. And with it, the need for sophisticated, reliable, and adaptive state estimation methods will become paramount. The work of Kan, Yang, Sun, and Xie represents a meaningful step forward in that direction—bringing the promise of seamless, predictable, and efficient electric mobility one step closer to reality.
Yingzhe Kan, Min Yang, Huaze Sun, Yunfei Xie, School of Mechanical Engineering, Chongqing University of Technology. Journal of Chongqing University of Technology (Natural Science), doi:10.3969/j.issn.1674-8425(z).2024.11.003