A deep learning approach to analysis of voltage fluctuations of Li-ion batteries

We were studying ways to improve the method of monitoring lithium power sources by their voltage noise, using numerical methods and problem-oriented programming. To do this the electrical parameters of lithium current sources were measured in various operating modes. The problem of signal-to-noise separation has been solved by methods of mathematical data processing. The analysis of the stochastic composition of the noise of current sources was carried out using clustering. A neural network was trained to determine the charge state of the element from its noises based on the obtained data on voltage fluctuations. It is shown that long short-term memory layers are efficient for solving this problem.