A Multi-Channel Deep-Learning Prediction Algorithm for Battery State-of-Health Indicator

Lithium batteries are an essential part of many modern technologies. In most mission-critical applications it is essential to evaluate the current state-of-health (SOH) of the battery during its operating life using adequate condition monitoring tools. The acquired diagnostic data can then be used to forecast the future degradation trend of the battery, and thus to predict the remaining useful life of the device. In this work, a multi-channel deep-learning algorithm based on a Long Short-Term Memory Neural Network is presented in order to predict the future degradation trend of a set of battery's parameters. In order to test and validate the performances of the algorithm, a battery degradation dataset provided by the Toyota Research Institute has been used. The results illustrated the ability of the proposed method to predict the future degradation of several parameters, like charge and discharge capacity, internal resistance, required charging time and maximum overheating of the batteries. Using this approach, it is possible to correctly predict the degradation trend of battery SOH and the remaining useful life even after only few cycles of the battery life.