Prognostication of lithium-ion battery capacity fade based on data space compression visualization and SMA-ISVR

Accurate battery capacity fade estimation is essential for the reliable and safe operation of lithium-ion batteries. Current research on capacity estimation has achieved remarkable advancements. However, the presence of multiple characteristics data during degradation affects capacity fade, which are numerous and high dimensional. A capacity degradation prediction method based on data compression visualization and SMA-ISVR is proposed in this paper. Firstly, principal component analysis (PCA) method is used to determine the compressed dimensionality of battery aging data based on the proportion of main features. Secondly, similarity between feature samples is calculated by t-distributed stochastic neighbor embedding (t-SNE) method, which compresses the high-dimensional data into low dimensions in a visualization space and preserves the local structure among samples at the same time. In addition, a Density-Based Spatial Clustering with Noise Application (DBSCAN) method is used to deal with the noise present in compressed spatial data. Finally, an improved support vector regression (ISVR) is proposed to predict the battery capacity, linking the capacity decline trend through a dual kernel function. Meanwhile, global search capability of slime mould algorithm (SMA) is employed to optimize the kernel function and penalty factors of ISVR model. The SMA-ISVR algorithm is used in simulation experiments containing inputs from two different battery datasets. The results show that simulation of battery aging with compressed data and SMA-ISVR algorithm significantly improves the prediction accuracy.