Remaining Useful Lifetime Prediction of Lithium-Ion NMC 18650 Battery Cells Using Support Vector Machine

The rise of the energy needs of humans brought forth important innovations such as the lithium-ion battery. However, lithium-ion battery cells would experience a capacity drop in their usage due to imperfections in the chemical reactions. Predicting that capacity degradation behavior would help improve the efficiency of the lithium-ion battery cells’ usage, maintenance, and replacement. One of the many ways to predict said behavior is by using data-driven methods such as machine learning. In this research, the prediction of a lithium-ion battery cell’s Remaining Useful Lifetime (RUL) with the NMC cathode type and the 18650 form factor using a machine learning algorithm known as Support Vector Machine (SVM) will be conducted to assess the performance of said algorithm. The data used in this research is a secondary dataset from a study conducted at the Sandia National Laboratory, which tests numerous types of lithium-ion battery cells under several conditions to study the general degradation behavior of lithium-ion battery cells. The dataset is limited to conditions such as a charging rate of 0.5C. The dataset will then be used to train and test several models, each with varying kernel functions such as the linear kernel, polynomial kernel, radial-basis function kernel, and sigmoid kernel, before and after their hyperparameters are tuned using the random search grid methods. The acquired performances of predicting the RUL would then be compared with each other to find the best-performing models. The performance indexes used in this experiment are the Mean Absolute Error (MAE), Mean Squared Error (MSE), Root Mean Squared Error (RMSE), Coefficient of Determination (R2), Root Mean Squared Logarithmic Error (RMSLE), and Mean Absolute Percentage Error (MAPE). The best performance from the SVM model predicting the RUL of a lithium-ion NMC 18650 battery cell are as follows, MAE score of 1,347.4521, MSE score of 3,729,987.2910, RMSE score of 1,931.3175, R2 score of 0.6944, RMSLE score of 0.8844, and MAPE score of 2.0422. That performance was achieved using an SVM model with hyperparameters as follows, radial-basis function for the kernel function, a C value of 9.596, and an epsilon value of 1.9. For other hyperparameters of the radial-basis function kernel itself uses the gamma constant of 0. 10000000024877526.