|Title||Battery diagnosis for electrical vehicles through semi-physical fuzzy models|
|Publication Type||Conference Paper|
|Year of Publication||2016|
|Authors||Sánchez, L., Otero J., Couso I., and Blanco C.|
|Conference Name||2016 IEEE International Conference on Fuzzy Systems (FUZZ-IEEE)|
|Keywords||Batteries, battery diagnosis, battery health evaluation, battery management systems, Battery models, battery powered vehicles, Computational modeling, differential equations, Discharges (electric), electrical vehicles, fuzzy models, Fuzzy rule-based systems, Fuzzy systems, gradient descent algorithms, graphics processing units, learning (artificial intelligence), learning algorithm, Li-FePO4batteries, Li-ion battery, Liquids, massively parallel GPU, Mathematical model, nonlinear blocks, Numerical models, parallel processing, power engineering computing, secondary cells, semiphysical fuzzy models, TensorFlow, TensorFlow™, Vehicles|
An intelligent model of a Li-Ion battery for electrical vehicles is proposed that allows for a fast battery health evaluation without the need of removing the battery from the vehicle. The only data needed for performing the condition monitoring are logged performance records (currents and voltages) that are commonly available at Battery Management Systems. The model comprises a combination of differential equations and fuzzy rule-based systems, these last being embedded as non-linear blocks in the differential equations. Fuzzy rules are learnt from data with the help of metaheuristics and fine tuned with gradient descent algorithms. A TensorFlow™ implementation of the learning algorithm has been developed that takes advantage of the numerical processing capabilities of a massively parallel GPU and improves the learning speed by a high margin.