Reviewed by Lexie CornerOct 22 2024
Researchers from Pandit Deendayal Energy University have published a groundbreaking study in the journal Green Energy and Intelligent Transportation, demonstrating the application of optimized Artificial Neural Networks (ANN) and principal components-based feature creation for State of Charge (SoC) estimation in LiFePO4 batteries.
Optimized ANN for LiFePO4 battery charge estimation using principal components-based feature generation. Image Credit: Green Energy And Intelligent Transportation
Electric vehicles (EVs) are driving the current energy revolution, necessitating the development of cutting-edge technology to guarantee sustainability and peak performance.
A critical component of this is accurate State of Charge (SoC) estimation, essential for maintaining battery health and optimizing efficiency—key factors for the widespread adoption of EVs.
Current SoC estimation methods face limitations in terms of accuracy, robustness, and real-time application. The researchers' approach, utilizing ANN and feature creation based on principal component analysis, aims to address these challenges.
Precise data collection is essential for developing the necessary algorithms. The team designed a custom 12V, 4Ah battery pack with specialized hardware for real-time monitoring. A DHT22 temperature sensor attached to a Raspberry Pi tracks temperature fluctuations, while a computerized battery analyzer measures the battery's voltage, current, and open-circuit voltage, ensuring accurate data for algorithm development.
The researchers analyzed the principal components of the collected battery data to create input parameters for their ANN model. Three main components were generated for feature engineering.
In optimizing the model, the team tested eleven combinations of ten different optimizers to minimize the loss function, using early stopping to improve training efficiency. The Adafactor optimizer stood out, achieving a Root Mean Square Error (RMSE) of 0.4083 and an R² Score of 0.9998.
Beyond EV applications, accurate SoC estimation can be crucial in vehicle-to-grid (V2G) systems, where EVs return electricity to the grid during peak demand. It can also enhance energy storage systems within renewable energy grids, ensuring more reliable and efficient power delivery. As the global shift towards sustainable energy continues, the demand for precise SoC prediction will increase, driving innovations in battery management and energy storage technology.
This novel approach provides valuable insights for electric vehicle operators and energy companies, offering significant progress toward smarter charging strategies and effective energy management. Future SoC estimation methods must adapt to changing driving conditions and energy usage patterns in real time, enabling dynamic energy management and ensuring EV efficiency in any situation.
Journal Reference:
Mehta, C., et al. (2024) Optimized ANN for LiFePO4 Battery Charge Estimation using Principal Components based Feature Generation. Green Energy and Intelligent Transportation. doi.org/10.1016/j.geits.2024.100175.