UKF Algorithm Enhances Accuracy of EV Battery Monitoring
As the global automotive industry accelerates its shift toward electrification, one of the most pressing challenges remains the accurate estimation of a battery’s state of charge (SOC). For electric vehicle (EV) owners, the ability to trust the displayed remaining range is not just a matter of convenience—it directly influences driving confidence, charging behavior, and long-term battery health. Behind this seemingly simple metric lies a complex web of electrochemical dynamics, nonlinear behavior, and real-time data processing. A recent study led by Wenqiang Shu from the College of Electrical and Electronic Engineering at Anhui Institute of Information Technology has made significant progress in refining the precision of SOC estimation, particularly for lithium iron phosphate (LiFePO₄) batteries, which are increasingly favored for their safety, longevity, and thermal stability.
Published in the Journal of Foshan University (Natural Sciences Edition), the research introduces an advanced computational framework based on the Unscented Kalman Filter (UKF) to improve the accuracy and reliability of SOC monitoring in pure electric vehicles. Unlike traditional methods that often suffer from cumulative errors or impractical operational constraints, the UKF-based approach demonstrates a consistent estimation error of less than 4%, a performance threshold that meets the stringent demands of modern EV energy management systems.
The significance of this development cannot be overstated. Battery management systems (BMS) are the central nervous system of any electric vehicle, responsible for monitoring cell voltage, current, temperature, and—most critically—state of charge. An inaccurate SOC reading can lead to premature shutdowns, unexpected range anxiety, or worse, overcharging and deep discharging, both of which degrade battery life and pose safety risks. While several methods exist for SOC estimation, each comes with inherent limitations that hinder their effectiveness in real-world driving conditions.
One of the earliest and most straightforward techniques is the discharge test method, which involves fully discharging the battery under controlled conditions to measure total capacity. While highly accurate, this approach is impractical for in-vehicle use, as it requires the battery to be taken offline and rendered unusable during testing. Similarly, the open-circuit voltage (OCV) method relies on measuring the battery’s voltage after a prolonged rest period, during which internal chemical reactions stabilize. However, this method is incompatible with dynamic driving scenarios where continuous monitoring is required.
The ampere-hour (Ah) integration method, also known as Coulomb counting, calculates SOC by integrating the current over time. While conceptually simple, this method suffers from two major drawbacks: it requires an accurate initial SOC value, which is often unknown, and it is prone to error accumulation due to sensor inaccuracies and unaccounted self-discharge. Over time, these small errors compound, leading to significant deviations from the true SOC.
More sophisticated approaches, such as neural networks and fuzzy logic systems, have been explored to model the nonlinear behavior of lithium-ion batteries. These machine learning techniques can adapt to complex operating conditions and learn from historical data. However, they demand extensive training datasets and are sensitive to variations in driving patterns, temperature fluctuations, and aging effects. Moreover, their “black box” nature makes them difficult to validate and integrate into safety-critical automotive systems.
In contrast, model-based filtering techniques offer a balanced solution by combining physical battery models with statistical estimation algorithms. Among these, the Kalman Filter (KF) and its variants have gained widespread adoption due to their ability to fuse noisy sensor measurements with dynamic system models to produce optimal state estimates. The Extended Kalman Filter (EKF), a popular choice in early BMS designs, linearizes nonlinear system dynamics around the current operating point. While effective in mildly nonlinear systems, the EKF can introduce significant errors when applied to highly nonlinear processes such as battery electrochemistry, especially under varying load conditions.
Recognizing these limitations, Shu’s research focuses on the Unscented Kalman Filter (UKF), a more robust alternative that avoids linearization by using a deterministic sampling technique known as the Unscented Transform (UT). Instead of approximating the system’s nonlinearities through derivatives, the UKF selects a set of sample points—called sigma points—that capture the mean and covariance of the state distribution. These points are then propagated through the actual nonlinear system model, and the resulting transformed points are used to compute the updated state estimate and its uncertainty.
This approach preserves the statistical properties of the system more accurately than linearization, making it particularly well-suited for applications where the relationship between SOC and open-circuit voltage is highly nonlinear—a characteristic prominently observed in LiFePO₄ batteries. In fact, the study highlights that the OCV-SOC curve for the tested 10Ah lithium iron phosphate cell exhibits a relatively flat profile across much of its operating range, making it especially challenging to estimate SOC from voltage alone. Small measurement errors can lead to large uncertainties in SOC, underscoring the need for a filtering method capable of handling such ambiguity.
To implement the UKF algorithm, Shu first developed a second-order Thevenin equivalent circuit model of the battery. This model represents the battery’s internal behavior using a combination of resistors and capacitors to simulate ohmic resistance, polarization effects, and transient voltage responses. The open-circuit voltage is treated as a nonlinear function of SOC, derived from experimental pulse discharge tests conducted under controlled conditions. By subjecting the battery to a series of 1C discharge pulses with 60-minute rest periods between steps, the researchers were able to map the OCV-SOC relationship across the full state-of-charge range from 100% to 0%.
The resulting data was then imported into MATLAB for curve fitting, yielding a high-fidelity functional representation of the OCV-SOC dependency. This mathematical model serves as the foundation for the UKF’s output equation, allowing the filter to predict the expected terminal voltage based on the estimated SOC and compare it with actual measurements. The difference between predicted and measured voltage—known as the innovation—is used to correct the SOC estimate in real time, minimizing estimation error.
What sets the UKF apart is its ability to handle the inherent noise and uncertainty present in real-world measurements. All sensors, whether measuring current, voltage, or temperature, are subject to random fluctuations and systematic biases. Additionally, the battery itself behaves as a stochastic system, with internal parameters such as internal resistance and capacity changing over time due to aging, temperature variations, and usage patterns. The UKF accounts for these uncertainties by maintaining a covariance matrix that quantifies the confidence in the state estimate. When new measurements arrive, the filter computes an optimal gain that balances the relative trust between the model prediction and the sensor data.
In the simulation experiments conducted by Shu, the UKF was benchmarked against the EKF under identical conditions: a constant 5A discharge starting from an initial SOC of 95%. The reference SOC trajectory was generated using the ampere-hour method, assuming perfect knowledge of the initial state and no measurement errors—a best-case scenario that provides a reliable baseline for comparison. The results revealed a clear performance gap between the two filters.
During the early stages of discharge, both algorithms tracked the reference SOC closely, demonstrating their capability to initialize and respond to dynamic loads. However, as the discharge progressed, the EKF began to diverge, with its estimation error increasing steadily. By the end of the 80-minute test, the EKF’s error had reached 6.08%, exceeding the acceptable threshold for most automotive applications. In contrast, the UKF maintained a much tighter error bound, staying within 4% throughout the entire discharge cycle. This superior performance can be attributed to the UKF’s more accurate handling of nonlinearities and its ability to preserve higher-order statistical moments during state propagation.
The implications of this research extend beyond academic interest. For automakers and battery manufacturers, the adoption of UKF-based SOC estimation could lead to more reliable range predictions, reduced battery degradation, and improved user experience. Accurate SOC estimation enables smarter energy management strategies, such as predictive regenerative braking, optimized charging profiles, and adaptive thermal control. It also supports second-life applications for retired EV batteries, where precise health assessment is critical for repurposing in stationary energy storage systems.
Moreover, the methodology presented by Shu is not limited to laboratory settings. The use of MATLAB for simulation and algorithm development aligns with industry-standard practices in automotive control system design. The modular structure of the UKF framework allows for integration with existing BMS architectures, and the reliance on standard electrical measurements—voltage, current, and temperature—ensures compatibility with commercial sensor hardware.
However, the study also acknowledges certain limitations that warrant further investigation. One key challenge is the assumption of fixed process and measurement noise covariances. In practice, these noise characteristics can vary significantly depending on operating conditions, battery age, and environmental factors. A static noise model may lead to suboptimal filter performance, especially during transient events or under extreme temperatures. To address this, future work could explore adaptive filtering techniques that dynamically adjust the noise parameters based on real-time data, thereby enhancing robustness and long-term accuracy.
Another area for improvement lies in the integration of aging models. As batteries degrade over time, their capacity fades and internal resistance increases, altering the OCV-SOC relationship and affecting the accuracy of the SOC estimate. Incorporating online capacity estimation and impedance tracking into the UKF framework could enable the system to self-calibrate and maintain high accuracy over the battery’s entire lifespan.
From a broader perspective, this research contributes to the growing body of knowledge aimed at making electric vehicles more intelligent, efficient, and user-friendly. As governments worldwide implement stricter emissions regulations and consumers become more environmentally conscious, the demand for reliable and high-performance EVs will continue to rise. Technologies that enhance battery monitoring and management are therefore not just technical advancements—they are enablers of sustainable transportation.
The automotive industry has already begun to recognize the value of advanced battery algorithms. Major OEMs and Tier 1 suppliers are investing heavily in software-defined vehicles, where over-the-air updates and AI-driven diagnostics play a central role in vehicle performance and safety. In this context, the UKF represents a mature and proven technology that can be deployed today to improve battery intelligence without requiring costly hardware upgrades.
Educational institutions and research centers also stand to benefit from this work. By publishing detailed methodologies and simulation results, Shu’s study provides a valuable resource for students, engineers, and researchers working in the field of battery systems. The transparency of the approach—using publicly available tools like MATLAB and standard battery testing protocols—encourages reproducibility and collaboration, fostering innovation across academic and industrial boundaries.
In conclusion, the application of the Unscented Kalman Filter to lithium-ion battery SOC estimation marks a significant step forward in the quest for more accurate and reliable electric vehicle energy management. By leveraging the strengths of nonlinear filtering and physics-based modeling, this approach overcomes many of the shortcomings of traditional methods, delivering estimation accuracy within acceptable industry standards. As the world moves closer to a zero-emission future, advancements like these will play a crucial role in building the trust and confidence needed to accelerate the adoption of electric mobility.
Wenqiang Shu, College of Electrical and Electronic Engineering, Anhui Institute of Information Technology, Journal of Foshan University (Natural Sciences Edition)