Residual capacity estimation for ultracapacitors in electric vehicles using artificial neural network
- Publication Type:
- Conference Proceeding
- IFAC Proceedings Volumes (IFAC-PapersOnline), 2014, 19 pp. 3899 - 3904
- Issue Date:
Files in This Item:
|Residual Capacity Estimation for Ultracapacitors in Electric Vehicles Using Artificial Neural Network.pdf||Published version||691.75 kB|
Copyright Clearance Process
- Recently Added
- In Progress
- Closed Access
This item is closed access and not available.
© IFAC. The energy storage system (ESS) plays a significant role in fulfilling the driving performance requirements and ensuring operational safety in an electric vehicle. Ultracapacitors (UCs) are being actively studied and used in parallel with batteries or fuel cells forming hybrid energy storage systems in electric vehicles. They show excellent potential in terms of the sourcing and sinking of power, particularly for the peak-power demand encountered in aggressive regenerative braking. Since there are an increasing number of ultracapacitor applications, which now includes commercial automotive applications, establishing a good model to represent their dynamics, especially the residual capacity estimation (RCE), is vital; but this is challenging. This paper presents a residual capacity estimation model which is based on an artificial neural network (ANN). This takes both charging and discharging current and temperature into consideration. The proposed ANN model comprises of three inputs and one output: the inputs are temperature, current and voltage, and the output is the residual charge. The model is trained and validated by feeding a test database which is extracted from experimental testing of ultracapacitors at different currents and temperatures on a well-established test rig. The training data should span the whole prediction scope, therefore the test currents and temperatures both vary over a wide range and cover all the possible operational conditions of the on-board ultracapacitors. The Back-Propagation (BP) algorithm, together with an early stopping strategy, is adopted to train the proposed ANN model to assure adequately accurate prediction while avoiding overfitting risks. The model performance is validated with experimental results over a set of test data randomly selected.
Please use this identifier to cite or link to this item: