BATTERY STATE PREDICTION
With rising concerns about global warming, the development of electrical alternatives in different sectors (electric vehicles, solar farms, etc.) is an important vision for humanity.
For example, the successful development of electric vehicles depends highly on the performance, cost and safety of the batteries. Rechargeable lithium-ion batteries are currently the best choice in various situations. For safety, optimizing battery behavior and monitoring of the entire electrification system an advanced battery management system is needed.
Today, one of the major barriers to widespread adoption of electrical solutions is battery range anxiety (especially for electric cars). To be able to predict the remaining charge and health of batteries, it is crucial to alleviate this problem. With this information, we can predict the remaining usable time of the battery. This way, the batteries will be used to their fullest potential. In addition to that, spent batteries can be redeployed in less demanding, second-life applications (stationary grid storage at photovoltaic farms).
The goal of the project was the creation and implementation of a machine learning model, which predicts the state of a lithium-ion battery based on various sensory data. The model predicted the remaining charge of the battery, the overall health of the battery and the current available power output. The model runs on a Microcontroller, which can be integrated into different Battery Management Systems.
Used Tools: Python, Tensorflow, C
Many nonlinear degenerative processes within the battery lead to performance decrease. The prediction of these processes, which are coupled with thermal and mechanical heterogeneities within a cell, is very challenging. So far, many physical and semi-empirical models have tried to solve this task. Recently, advances in computational power and data generation have enabled machine learning techniques to accelerate progress on these challenges. Machine learning approaches offer promising opportunities for improving upon prediction accuracy, greater interpretability and broader application to a wide range of conditions.