New IFFRLS-IAEKF Algorithm Enhances Battery SOC Estimation for EVs
As the global automotive industry accelerates its shift toward electrification, one of the most pressing challenges remains accurate and reliable estimation of a battery’s state of charge (SOC). For drivers, this metric is more than just a number on a dashboard—it directly influences range anxiety, charging behavior, and overall confidence in electric vehicles (EVs). Behind the scenes, sophisticated algorithms within the battery management system (BMS) are responsible for calculating this vital parameter. However, traditional methods often fall short under dynamic driving conditions, where rapid changes in load, temperature, and usage patterns can degrade estimation accuracy.
Now, a breakthrough approach developed by researchers at Anhui Polytechnic University promises to significantly improve the precision of real-time SOC estimation. The team—comprising Yan Huihui, Zhang Yan, Zhang Peixian, Ma Wenjing, and Zhou Yuan—has introduced an innovative hybrid algorithm known as IFFRLS-IAEKF, which combines enhanced parameter identification with adaptive filtering techniques to deliver superior performance under complex operating scenarios.
Published in the Journal of Harbin University of Commerce (Natural Sciences Edition), this research presents a comprehensive solution that addresses two critical weaknesses in existing SOC estimation frameworks: static model parameters and non-adaptive noise handling in Kalman filtering. By rethinking how data is weighted over time and how errors are dynamically corrected, the proposed method sets a new benchmark for online battery state monitoring.
The significance of precise SOC estimation cannot be overstated. Inaccurate readings can lead to premature shutdowns, inefficient energy use, or even overcharging risks—all of which compromise both safety and user experience. Current approaches such as ampere-hour integration suffer from cumulative drift due to sensor inaccuracies, while open-circuit voltage (OCV) methods require long rest periods impractical in real-world driving. Extended Kalman Filter (EKF)-based strategies have become standard in many BMS platforms because they balance computational efficiency with reasonable accuracy. Yet, their dependence on fixed process and measurement noise covariances limits adaptability when battery characteristics evolve during operation.
Recognizing these limitations, the Anhui Polytechnic team focused on enhancing both the modeling fidelity and the filtering intelligence of the SOC estimation pipeline. Their strategy begins with refining the recursive least squares (RLS) technique—a widely used method for identifying parameters in equivalent circuit models like the first-order Thevenin model. While conventional forgetting factor RLS (FFRLS) improves tracking capability by emphasizing recent data, it relies on a predetermined, constant forgetting factor. This rigidity becomes problematic under fluctuating workloads, leading either to sluggish response or excessive sensitivity to outliers.
To overcome this, the researchers introduced the Improved Forgetting Factor Recursive Least Square (IFFRLS) algorithm, featuring a segmented, data-dependent forgetting factor function. Instead of using a single fixed value, the algorithm dynamically adjusts the weight given to historical versus current measurements based on the volume of incoming data. Specifically, when the sample count is low (≤32), a conservative factor of 0.95 ensures stability and prevents overfitting. Beyond this threshold, the forgetting factor transitions into a continuously adaptive regime governed by an inverse relationship with data length, effectively minimizing the risk of “data saturation”—a phenomenon where older but still relevant information is prematurely discarded.
This innovation allows the IFFRLS method to maintain high responsiveness during transient events—such as sudden acceleration or regenerative braking—while preserving steady-state accuracy during prolonged cruising. As a result, key parameters including ohmic resistance (Ro), polarization resistance (Rp), and polarization capacitance (Cp) are tracked with greater fidelity across diverse driving cycles.
But accurate parameter identification alone is insufficient without a robust state estimator capable of fusing this information into a reliable SOC prediction. Here, the team turned to the Adaptive Extended Kalman Filter (AEKF), known for its ability to self-adjust noise statistics. However, even AEKF has its shortcomings, particularly when initial assumptions about process and measurement noise diverge from actual conditions.
To address this, the researchers developed the Improve Adapted Extended Kalman Filter (IAEKF), a novel variant that incorporates an error-weighting mechanism derived from the discrepancy between EKF-based estimates and those obtained via coulomb counting (ampere-hour integration). This difference, representing the accumulated bias in the filter’s output, serves as a correction signal that modulates the process noise covariance matrix in real time.
By introducing this feedback loop, the IAEKF gains a form of self-awareness—its internal uncertainty model evolves in response to observed deviations, making it far more resilient to modeling inaccuracies and sensor imperfections. Furthermore, the adjustment of the error covariance matrix indirectly influences the Kalman gain, allowing the filter to selectively trust certain inputs more heavily depending on current reliability.
The synergy between IFFRLS and IAEKF creates a closed-loop system where improved model parameters feed into a smarter filter, which in turn produces cleaner state estimates that further refine the identification process. This co-adaptive architecture represents a significant advancement over sequential or decoupled approaches commonly found in prior literature.
To validate their methodology, the team conducted extensive simulations under the Dynamic Stress Test (DST) protocol—a rigorous driving cycle designed to mimic aggressive urban driving with frequent stops, starts, and speed variations. Using experimental data collected from a Leoch AGM lead-acid battery tested under controlled thermal conditions (25°C), the researchers compared the performance of several algorithms: standard RLS, FFRLS, EKF, AEKF, and their proposed IFFRLS-IAEKF combination.
Voltage prediction accuracy served as the primary metric for evaluating parameter identification. Results showed that the IFFRLS algorithm achieved lower voltage tracking errors throughout the test cycle, with the cumulative error reduced by 24% compared to basic RLS and 14.8% relative to FFRLS. These improvements stem directly from the algorithm’s ability to avoid over-reliance on outdated data while maintaining numerical stability—a delicate balance that fixed-factor methods struggle to achieve.
More importantly, the ultimate goal was SOC estimation accuracy. Under DST conditions, the IAEKF demonstrated tighter convergence to the true SOC trajectory, with minimal overshoot and faster recovery after disturbances. The average estimation error remained below previously reported thresholds, consistently outperforming both EKF and AEKF counterparts. Notably, the maximum deviation never exceeded acceptable safety margins, underscoring the method’s suitability for deployment in commercial BMS environments.
One of the standout features of this research is its practical orientation. Unlike purely theoretical studies that rely on idealized datasets or simplified assumptions, this work integrates real battery characterization data obtained through Hybrid Pulse Power Characterization (HPPC) testing. This empirical foundation ensures that the identified OCV-SOC relationship reflects genuine electrochemical behavior rather than synthetic approximations. The resulting sixth-order polynomial fit provides a highly accurate representation of the nonlinear dependency between terminal voltage and charge level, forming a solid basis for subsequent filtering operations.
Moreover, the computational complexity of the proposed algorithm remains compatible with embedded systems typically found in modern EVs. While some advanced filters like Unscented Kalman Filters (UKF) offer higher-order accuracy, they come at the cost of increased processing demands. The IFFRLS-IAEKF framework retains the lightweight structure of classical EKF implementations while delivering performance gains through intelligent adaptation—making it an attractive candidate for integration into next-generation BMS chips.
From an engineering perspective, the implications of this research extend beyond mere accuracy improvements. More precise SOC estimation enables automakers to extract additional usable capacity from batteries without compromising safety margins. It also enhances predictive capabilities for range estimation, reducing driver anxiety and improving route planning functionality in navigation systems. Additionally, better state awareness supports advanced functions such as fast-charging optimization, cell balancing, and health diagnostics—all crucial components of holistic battery lifecycle management.
For fleet operators and mobility service providers, such advancements translate into tangible operational benefits. Accurate SOC tracking reduces unplanned downtime, optimizes charging schedules, and extends battery lifespan through more refined usage patterns. In shared mobility applications, where vehicle availability directly impacts revenue, even small gains in predictability can yield substantial returns.
Regulatory bodies may also find value in this technology. As governments worldwide push for stricter emissions standards and incentivize zero-emission transportation, ensuring the reliability and transparency of EV performance metrics becomes increasingly important. Standardized, high-fidelity SOC estimation methods could support certification processes and consumer protection initiatives, fostering greater trust in electric mobility solutions.
Looking ahead, the research opens several avenues for future exploration. One promising direction involves extending the IFFRLS-IAEKF framework to account for aging effects and temperature dependencies, which were held constant in the current study. Incorporating online capacity fade detection and thermal compensation would make the algorithm even more robust across seasons and years of service.
Another area ripe for development is hardware-in-the-loop (HIL) validation. While simulation results are compelling, real-world testing on prototype BMS units would provide definitive proof of viability. Collaborations with automotive suppliers or OEMs could accelerate the transition from academic concept to industrial application.
Additionally, the modular nature of the algorithm lends itself well to integration with machine learning components. For instance, neural networks trained on large-scale driving datasets could assist in predicting upcoming load profiles, enabling proactive adjustments to filtering parameters. Such hybrid AI-classical control architectures represent the frontier of smart battery management.
Cybersecurity considerations must also be addressed as BMS software grows more complex. Ensuring the integrity and authenticity of SOC estimates will be essential, especially as vehicles become more connected and autonomous. Robust encryption, secure boot mechanisms, and anomaly detection protocols should accompany any deployment of advanced estimation algorithms.
In conclusion, the IFFRLS-IAEKF algorithm developed by Yan Huihui, Zhang Yan, Zhang Peixian, Ma Wenjing, and Zhou Yuan marks a meaningful step forward in the quest for dependable battery state estimation. By intelligently combining adaptive parameter identification with error-aware filtering, the method delivers measurable improvements in accuracy, stability, and responsiveness under realistic operating conditions. Its publication in the Journal of Harbin University of Commerce (Natural Sciences Edition) underscores the growing contribution of Chinese academic institutions to cutting-edge automotive technologies.
As the EV market continues to expand, innovations like this will play a pivotal role in shaping the driving experience of tomorrow. With consumers demanding ever-greater reliability and transparency, the behind-the-scenes mathematics of battery management will increasingly come into the spotlight. The work of this Anhui Polytechnic team exemplifies how rigorous scientific inquiry, grounded in practical experimentation, can yield solutions that are not only technically sound but also commercially viable.
For engineers, policymakers, and drivers alike, the message is clear: the future of electric mobility depends not just on bigger batteries or faster chargers, but on smarter ways of understanding and managing the energy we already have. And with algorithms like IFFRLS-IAEKF entering the conversation, that future looks brighter—and more predictable—than ever.
Yan Huihui, Zhang Yan, Zhang Peixian, Ma Wenjing, Zhou Yuan, School of Electrical Engineering, Anhui Polytechnic University; Journal of Harbin University of Commerce (Natural Sciences Edition), DOI: 10.1672-0946(2024)06-0658-06