| Abstract [eng] |
The prognostic model of lithium-ion batteries capacity is designed during the final project work. This model lets to estimate one of the battery management system state – state of health (SOH). Three internal battery variables and their combinations, two data processing methods, and a feedforward neural network with different architecture were used for modelling. To evaluate the accuracy of the implemented models, the mean average error (MAE) and root mean square error (RMSE) were calculated. After analysing the obtained results, it was found out that the data preparation methods and neural network architecture make a strong impact on the accuracy of the results. Irrespective of the influence of internal parameters on the prognostic model, it was found that a feedforward neural network with ten neurons in the hidden layer is more suitable for capacity prediction, and min-max normalization would be most appropriate for data processing. After the analysis of the influence of the internal variables for battery capacity prognosis, it was found that estimating the errors according to the MAE, it is most appropriate to include the internal variables of the voltage and temperature to the prognostic model. Applying such a model to the available data, the mean of MAE was 0,022 Ah, this is about 0,9%. Comparing the prognostic model developed in this work with the studies performed by others, it was found that the errors of the capacity predictions of this model are smaller than in previous studies, therefore the obtained state of health performance states better reflect the possible values in reality. A more detailed study and application of this model in the future could contribute to a more accurate assessment of the battery lifetime and enhancement of the battery management system. |
