Chapter 10 - Intelligent Data Analytics for Battery Health Forecasting Using Semi-Supervised and Unsupervised Extreme Learning Machines

Abstract

Online health assessment of lithium-ion (Li-ion) battery (LIB) is required to predict the remaining useful life (RUL) of the battery-based system. Generally, internal resistance (R) and capacity (Ah) of the battery are utilized as a health identification feature (i.e., health indicator-HI). However, the measurement for both R and Ah are not an easy task, which leads to use of a large number of equipment and high analyzing cost. In this chapter, an alternative way to estimate the RUL during an online operating scenario has been formulated, which can extract the RUL from easy-recorded parameter of LIBs. To form a linear relationship between formulated HI and actual battery capacity, a power transformation method is applied and then Pearson and Spearman rank correlation methods are used to validate its performance. The supervised extreme learning machine (SELM), semi-supervised ELM (SSELM), and unsupervised ELM (USELM) are applied to evaluate the RUL by using extracted HI and transformed HI. This is the first attempt to implement the SSELM and USELM for RUL estimation of LIBs to make the system online in the real-time domain. The obtained results show that the new proposed approach is effective in evaluating the battery health condition without the need to measure the R and capacity.