Online Prediction of the Li-Ion Battery Remaining Useful Life for Smart Grid by Using Event-Driven Approach

Abstract

The latest innovative advancements have included renewable energy sources based on smart grids and electric vehicles (EVs). Technologies and equipment related to remote and integrated power systems offer improved developments for smart and microgrids. The battery is a crucial element of modern power systems; it is indispensable in many vital applications such as EVs, drones, avionics, satellites, mobile phones, and energy storage for renewable smart grids. Due to their superior characteristics, including but not limited to having a compact size, high power supply capability, and a higher number of charging/discharging cycles, the Lithiumion (Li-ion) batteries are one of the most dominant energy storage technologies. On the other hand, due to one of their most significant disadvantages being expensive, and to optimize their performance and ensure they last longer in the smart grid, their use is monitored using battery management systems (BMSs). The extensive processing resources that modern BMSs need can result in higher overhead power consumption. Regarding this, several embedded and integrated systems-based solutions have been proposed. This thesis focuses on upgrading the present Liion BMSs, reconstructing their associative data acquisition by employing the event-driven sensing mechanism. It aims at efficiently predicting the Li-ion battery's remaining useful life (RUL) online using machine learning (ML). Firstly, a suitable current high-power Li-ion battery real dataset has been distinguished and utilized during offline processing. Then, this dataset is reconstructed to make its form compatible with MATLAB. Using MATLAB, the proposed system is modeled. The eventdriven peak sensing model efficiently extracted the features from the studied battery consumption parameters based on the phenomenon of shape context feature extraction. The feature extractor is sufficiently achieved via embedding the event-driven peak sensing phenomenon in the system. These features are used for training the proposed event-driven remaining useful life (RUL) predictor using robust ML classifiers, namely k-Nearest Neighbor (kNN), Artificial Neural Network (ANN), Linear Regression (LR), Random Tree (RT), and Random Forest (RF). During online processing, the event-driven sensing-based acquired parameters are used to predict the battery capacity. Besides, by applying an event-driven peak sensing approach, the shape context features are extracted. Beyond, those fused features are used for online RUL prediction of the studied Li-ion battery cell by employing Weka software. The devised solution excelled its counterparts in terms of power consumption with a compression gain of 437.5-fold on average. The results showed superior performance for the LR among its studied counterparts, followed by the ANN, kNN, RF, then RT. The achieved MAE, RMSE, RAE, and RRSE by the LR based on the training and testing split of 70:30% are 0.0019, 0.0023, 1.0411%, and 1.1606%, respectively, and by the 5-fold cross-validation are 0.00289, 0.0051, 1.6405%, and 2.6656%, respectively. The proposed solution is parametrizable and can be utilized in different expected applications such as smart and microgrids, hybrid electric vehicles (HEVs), drones, distributed sensors, and satellites.