Prediction of Li-Ion Battery Discharge Patterns in IoT Devices Under Random Use Via Machine Learning Algorithms
CitationGokcen, A., Gokcen, A., Sahin, S. (2022). Prediction of Li-Ion Battery Discharge Patterns in IoT Devices Under Random Use Via Machine Learning Algorithms. Computer Journal, bxac089. https://doi.org/10.1093/comjnl/bxac089
This study presents foreseeing of the Lithium-ion battery discharge models for the Internet of Things (IoT) devices under randomized use patterns. IoT systems run in harmony with the human-machine interface, communication protocols and sharing data so long as uninterrupted data communication is exploited for their devices. Hence, forecasting the battery discharge duration is a very important issue for the regularization of IoT device performances. The well-known discharge duration is generally about the age-related electrochemical phenomena of an electrochemistry for Lithium-ion battery. The discharge changes of the battery were obtained from the input-output dynamics of the random battery use obtained from the randomized battery usage dataset in the NASA Ames prognostics data repository. They were investigated by machine learning methods and their results were estimated for life expectancy regularization of the IoT devices. In order to find the appropriate models of battery usage under randomized patterns, artificial neural network (ANN), Gaussian process and nonlinear regression models are evaluated in terms of battery capacity and internal resistance change as a function of discharged energy. The R-2, Adjusted R-2, root-mean-square-error (RMSE) and normalized-mean-square-error (NMSE) criteria were used to compare the performances of the obtained models for different settings. According to the results, ANN model, with settings of radial basis function activation function within single hidden-layer, and 20 hidden-layer neurons, shows the best performance in terms of R-2 = 1.0000 and NMSE = 1.7384.10(-4) metrics. RMSE = 0.9896.10(-4) is achieved by the ANN model with the settings of single hidden-layer with 10 neurons and hyperbolic-tangent activation function.