Hierarchical Deep Learning Model for Degradation Prediction Per Look-Ahead Scheduled Battery Usage Profile

Abstract

Batteries can effectively improve the security of energy systems and mitigate climate change by facilitating grid integration of wind and solar power. The installed capacity of battery energy storage system (BESS), mainly the lithium-ion batteries, has increased significantly. However, accurately quantifying battery degradation is challenging but crucial for the economics and reliability of BESS-integrated systems. This paper proposes a hierarchical deep learning-based battery degradation quantification (HDL-BDQ) model to quantify the battery degradation based on scheduled BESS operations. The HDL-BDQ model consists of two deep neural networks that work sequentially. It uses battery operational profiles as input features to accurately estimate the degree of degradation. Additionally, the model outperforms the existing fixed rate or linear rate based degradation models, as well as single-stage deep learning models. The training results demonstrate the high accuracy achieved by the proposed HDL-BDQ model. Moreover, a learning and optimization decoupled algorithm is implemented to strategically leverage the proposed HDL-BDQ model in optimization-based look-ahead scheduling (LAS) problems. Case studies demonstrate the effectiveness of the proposed HDL-BDQ model in LAS of a microgrid testbed.