| Abstract [eng] |
In the energy sector, sustainability encompasses a wide range of aspects. One of them could be the responsible use of resources. Lithium-ion batteries are particularly suitable for achieving responsible use of energy resources. As these batteries are rechargeable and long-lasting, they can be reused. This not only helps reduce the costs for a company or organisation but also protects the environment. Despite their positive properties, lithium-ion batteries age, are not resistant to all conditions and can fail. To identify faults and ageing, it is recommended to perform an ageing simulation and reliability assessment. This study analyses the effectiveness of machine learning models in assessing the internal states of lithium-ion batteries: state of charge and state of health. A review of the scientific literature revealed that different machine-learning models are combined for the ageing simulation and reliability assessment of lithium-ion batteries. Data is often pre-processed before being fed into the machine learning models. For this study, synthetic lithium-ion battery charging data with 10,000 data points and five variables: current, voltage, temperature, capacity, and testing time in seconds were generated. Additionally, real lithium iron phosphate battery discharge and charging data were analysed. Three artificial neural networks, one regression model, and one recursive filter were created. These models were refined, and a total of nine models were tested with synthetic and real battery data during the study. A total of 202 experiments were conducted. Of these, 181 experiments were conducted with synthetic data, and 21 experiments were conducted with real data. The results were compared with the actual values of the State of Charge and State of Health. The suitability of the models was verified with real data. It was observed that neural networks are more suitable than regression models. Based on the study results, an additional application system was created, allowing the user to independently generate synthetic lithium-ion battery charging data, apply a Kalman Filter, and calculate the state of charge of the batteries without programming. The application system also allows the user to upload their data. |
