New Battery Algorithm Aims to Boost EV Health and Range
Published · 3 min read
Scientists report a new joint estimation method for lithium-ion batteries. It predicts both the state of charge (SOC) and the state of health (SOH) at the same time. The team combines an enhanced unscented Kalman filter with a smart optimizer and a simple, proven circuit model. The goal is clear: safer batteries, longer life, and better performance.
What’s the Big Idea?
- SVD-MIUKF + UKF: a joint algorithm that estimates SOC and SOH together.
- Second-order RC model: a practical circuit model that captures battery behavior.
- Dung Beetle Optimizer (DBO): a metaheuristic that finds accurate model parameters.
- Real-time updates: SOH feeds capacity changes back into SOC for tighter accuracy.
- Robust math: singular value decomposition (SVD) stabilizes the filter covariance.
Why It Matters
EVs and grid storage depend on exact battery data. SOC shows how much energy is left. SOH shows how much the battery has aged. If either is wrong, range drops and safety risks grow. So a method that tracks both and links them can lift trust and value. It can also cut downtime for fleets and save money over the pack’s life.
How the Method Works
- Build the model. Use a second-order RC equivalent circuit to represent battery dynamics.
- Identify parameters. Apply the DBO to fit resistances and capacitances to real data.
- Estimate SOH. Run a standard UKF to track capacity fade in real time.
- Estimate SOC. Use the improved SVD-MIUKF to compute charge level with higher stability.
- Close the loop. Feed the latest SOH (capacity) into the SOC estimator on every step.
What’s Special About SVD-MIUKF?
- Handles tricky covariance. SVD helps avoid non-positive definite covariance issues that can crash a filter.
- Uses multi-innovation. It taps several past residuals but down-weights old data with smart weights.
- Stays stable. The filter resists “data oversaturation” and favors fresh, relevant signals.
Headline Results
Tests covered different operating modes and many aging cycles.
| Metric | Result | Notes |
| SOC RMSE | < 0.53% | Across various operating conditions |
| SOC MAE | < 0.42% | Across various operating conditions |
| SOH RMSE | < 0.0029% | High precision capacity tracking |
| SOH MAE | < 0.0021% | Low bias in health estimation |
| SOC under aging (RMSE) | < 0.92% | Stable even as the battery ages |
| SOC under aging (MAE) | < 0.91% | Low error across aging levels |
How It Stacks Up
There are two common paths to estimate SOC and SOH: data-driven methods and model-based methods.
| Approach | Strengths | Limits |
| Data-driven (e.g., LSTM, TCN + attention) | High accuracy with rich, clean data; learns complex patterns | Heavy compute; sensitive to noise and shifts; needs big datasets |
| Model-based (ECM + filters) | Explainable; lighter compute, and easier to validate | Needs careful parameter ID; may miss subtle effects |
| Proposed joint method (SVD-MIUKF + UKF + DBO) | Links SOC and SOH; stabilizes covariance; adapts as cells age | Still needs broad validation and embedded tuning |
What This Could Mean for EVs
- Better range planning: tighter SOC cuts surprise stops.
- Longer pack life: accurate SOH guides safe charge and discharge limits.
- Higher safety: precise states support stronger thermal and current control.
- Lower cost: fewer errors reduce warranty pain and service visits.
- Fleet wins: consistent data improves scheduling and health reports.
Limitations and Next Steps
- Real-world stress: temperature gradients, sensor drift, and cell imbalance still challenge models.
- Hardware fit: embedded chips have tight CPU and memory budgets.
- Wider tests: try more chemistries, formats, and fast-charge profiles.
- Richer data: refine parameter ID, because better parameters boost every estimate.
Bottom Line
This research shows a clear step forward. The SVD-MIUKF + UKF pair ties SOC and SOH into one loop. The DBO keeps the model honest by finding sharp parameters. The SVD and multi-innovation trick keep the filter stable and fresh. And the numbers speak for themselves: very low error across conditions and aging.
It’s not magic, but it’s close. With careful tuning and real-world trials, this method could improve EV performance, extend battery life, and raise driver confidence. That’s a win for users, fleets, and the grid.
