This repository contains a MATLAB/Simulink graduation project for lithium-ion battery State of Charge (SOC) estimation using three methods:
- Coulomb Counting (CC)
- Extended Kalman Filter (EKF)
- Unscented Kalman Filter (UKF)
The project compares the accuracy and robustness of the three estimators using a first-order Thevenin equivalent circuit battery model. The model is parameterized for a Panasonic UR18650ZTA lithium-ion cell and tested under different current profiles, initial SOC errors, voltage sensor noise, and current sensor bias.
State of Charge (SOC) is one of the most important internal states in a Battery Management System (BMS). It cannot be measured directly, so it must be estimated from measurable signals such as battery current and terminal voltage.
This project implements a simulation-based SOC estimation framework in MATLAB/Simulink. A battery plant model is connected to three SOC estimators running in parallel:
- Coulomb Counting as the baseline method.
- Extended Kalman Filter as a model-based nonlinear estimator.
- Unscented Kalman Filter as a sigma-point Kalman filter estimator.
The estimators are compared using:
- Root Mean Square Error (RMSE)
- Maximum absolute SOC error
- Convergence time from wrong initial SOC
The main objectives of this project are:
- Implement a first-order Thevenin Equivalent Circuit Model (1-RC ECM) for a lithium-ion battery in MATLAB/Simulink.
- Develop EKF and UKF algorithms for battery SOC estimation.
- Compare EKF and UKF performance against the traditional Coulomb Counting method.
- Evaluate estimator robustness under initial SOC error, voltage measurement noise, and current sensor bias.
The battery plant is based on a first-order Thevenin equivalent circuit model. The model includes:
- Open-circuit voltage lookup table,
OCV(SOC) - Ohmic resistance,
R0 - One RC polarization branch,
R1-C1 - Terminal voltage measurement
- Current measurement
- True SOC output from the Simscape battery model
The battery is implemented using the Battery (Table-Based) block in Simscape Electrical.
Main battery parameters are stored in:
scripts/BatterySOCEstimationData.m
The parameter file includes:
- SOC lookup breakpoints
- Temperature lookup points
- OCV table
- R0 table
- R1 table
- Time constant table
- Cell capacity
- Kalman filter tuning values
The project was implemented in MATLAB/Simulink using a physical battery plant and estimator blocks.
The EKF and UKF estimators are implemented using MATLAB Function blocks. Readable copies of the estimator code are provided in:
src/estimators/
Two current profiles were used to test the estimators.
A constant discharge current of 2 A was used to test estimator behavior under a simple load condition.
A pulse current profile of 3 A for 10 seconds followed by 20 seconds rest was used to test estimator behavior under dynamic loading.
Pulse generator configuration:
Amplitude: 3 A
Period: 30 s
Pulse width: 33% of period, approximately 10 s
Phase delay: 0 s
The estimators were evaluated under six test scenarios.
| Test | Current Profile | Plant SOC₀ | EKF/UKF SOC₀ | CC SOC₀ | Disturbance |
|---|---|---|---|---|---|
| T1 | Constant 2 A | 0.5 | 0.5 | 0.5 | None |
| T2 | Constant 2 A | 0.5 | 0.8 | 0.8 | Wrong initial SOC (+30%) |
| T3 | Pulse 3 A / 0 A | 0.5 | 0.5 | 0.5 | None |
| T4 | Pulse 3 A / 0 A | 0.5 | 0.8 | 0.8 | Wrong initial SOC (+30%) |
| T5 | Pulse 3 A / 0 A | 0.5 | 0.8 | 0.8 | Voltage sensor noise |
| T6 | Pulse 3 A / 0 A | 0.5 | 0.8 | 0.8 | Current sensor bias (+0.1 A) |
The CSV version of this table is available at:
results/tables/test_scenarios.csv
In this test, the plant and all estimators start from the same initial SOC of 0.5.
In this test, the true battery SOC starts at 0.5, while EKF, UKF, and Coulomb Counting start at 0.8.
The Kalman filter estimators correct the initial SOC error using voltage feedback. Coulomb Counting remains offset because it has no correction mechanism.
This test uses a pulse current profile with matched initial SOC.
This test uses a pulse current profile with 30% initial SOC error.
This test evaluates estimator robustness when voltage measurement noise is added.
This test evaluates estimator robustness when a +0.1 A current sensor bias is added.
Coulomb Counting accumulates current measurement error over time, while EKF and UKF maintain better performance because they use terminal voltage feedback.
| Test | Profile | EKF RMSE (%) | UKF RMSE (%) | CC RMSE (%) | EKF Max Error (%) | UKF Max Error (%) | CC Max Error (%) | EKF Conv. (s) | UKF Conv. (s) |
|---|---|---|---|---|---|---|---|---|---|
| T1 | Constant | 0.80 | 0.74 | 0.00 | 1.30 | 1.37 | 0.01 | - | - |
| T2 | Constant, wrong initial SOC | 0.83 | 0.78 | 30.00 | 5.70 | 5.67 | 30.00 | 4.0 | 4.0 |
| T3 | Pulse | 0.46 | 0.27 | 0.03 | 0.79 | 0.55 | 0.04 | - | - |
| T4 | Pulse, wrong initial SOC | 0.52 | 0.37 | 29.98 | 5.70 | 5.65 | 30.00 | 4.0 | 4.0 |
| T5 | Pulse + voltage noise | 1.50 | 1.51 | 29.98 | 4.78 | 5.01 | 30.00 | - | - |
| T6 | Pulse + current bias | 0.62 | 0.45 | 31.03 | 5.62 | 5.57 | 32.08 | 3.0 | 3.0 |
The CSV version of this table is available at:
results/tables/metrics_summary.csv
The simulation results show that:
- Coulomb Counting performs well only when the initial SOC is correct and there is no current sensor bias.
- Coulomb Counting cannot correct wrong initial SOC because it is an open-loop method.
- EKF and UKF can correct wrong initial SOC using terminal voltage feedback.
- EKF and UKF strongly outperform Coulomb Counting under realistic disturbances.
- UKF generally gives lower RMSE than EKF in most scenarios.
- Current sensor bias has a major effect on Coulomb Counting because the error accumulates over time.
battery-soc-estimation-ekf-ukf/
│
├── README.md
├── requirements.md
├── .gitignore
│
├── models/
│ └── battery_soc_estimation_ekf_ukf.slx
│
├── scripts/
│ ├── BatterySOCEstimationData.m
│ └── compute_metrics.m
│
├── src/
│ └── estimators/
│ ├── ekf_soc.m
│ └── ukf_soc.m
│
├── docs/
│ └── images/
│ ├── model_coulomb_counting.png
│ ├── model_ekf.png
│ ├── model_ukf.png
│ ├── current_constant_2A.png
│ ├── current_pulse_3A_10s_20s.png
│ └── pulse_generator_configuration.png
│
└── results/
├── figures/
│ ├── T1_constant_matched_initial_soc.png
│ ├── T2_constant_wrong_initial_soc.png
│ ├── T3_pulse_matched_initial_soc.png
│ ├── T4_pulse_wrong_initial_soc.png
│ ├── T5_pulse_voltage_noise.png
│ └── T6_pulse_current_bias.png
│
└── tables/
├── test_scenarios.csv
└── metrics_summary.csv
-
Open MATLAB.
-
Clone or download this repository.
-
Add the repository folder to the MATLAB path.
-
Run the battery parameter setup script:
run("scripts/BatterySOCEstimationData.m")- Open the Simulink model:
open_system("models/battery_soc_estimation_ekf_ukf.slx")-
Run the simulation from Simulink.
-
After the simulation finishes, run the metrics script:
run("scripts/compute_metrics.m")The script calculates:
- EKF RMSE
- UKF RMSE
- Coulomb Counting RMSE
- Maximum absolute error
- EKF convergence time
- UKF convergence time
This project was developed using MATLAB/Simulink.
Recommended software:
- MATLAB
- Simulink
- Simscape
- Simscape Electrical
The project was tested on a MATLAB/Simulink environment with a Simscape Electrical battery model.
The battery parameter script is based on a MATLAB/Simscape battery SOC estimation example and keeps the original MathWorks copyright notice where applicable.
The EKF and UKF implementations were added as custom MATLAB Function blocks for the graduation project comparison.
- The work is simulation-based and was not validated using real experimental battery data.
- The battery plant uses a first-order Thevenin equivalent circuit model.
- Thermal effects are not deeply analyzed.
- Aging effects are only represented indirectly through disturbance and robustness testing.
- The estimator parameters are fixed and not adapted online.
Possible future improvements include:
- Validate the estimators using experimental battery data.
- Add temperature-dependent SOC estimation.
- Test adaptive EKF and adaptive UKF methods.
- Add online parameter estimation for R0, R1, and C1.
- Implement the estimator on a microcontroller.
- Extend the model from a single cell to a battery pack.
- Muhammed AbdulZahra Zaboun
- Ali Majeed Kadhuim
Department of Electronics and Communications Engineering
College of Engineering
Gilgamesh University
Academic year: 2025–2026
Battery Management System, State of Charge, Lithium-ion Battery, Coulomb Counting, Extended Kalman Filter, Unscented Kalman Filter, MATLAB, Simulink, Simscape Electrical, Thevenin Model