Synthetic battery attribute generation to surmount data scarcity using auto‐correlation mechanism

• Published in: International journal of energy research Vol. 46; no. 7; pp. 9882 - 9891
• Main Authors: Channegowda, Janamejaya, Raj Urs, Vinayak, Lingaraj, Chaitanya
• Format: Journal Article
• Language: English
• Published: Chichester, UK John Wiley & Sons, Inc 10.06.2022; Hindawi Limited
• Subjects: Algorithms; Batteries; battery; Correlation; Datasets; electric vehicle; Electric vehicles; Electrification; Electronic devices; Electronic equipment; Energy; Energy charge; Energy storage; Lithium; Lithium-ion batteries; Storage batteries; Storage systems; synthetic data; time‐series analysis
• ISSN: 0363-907X; 1099-114X
• DOI: 10.1002/er.7784

Summary The energy storage sector has witnessed meteoric growth in the last decade. Lithium‐ion based batteries, at the centre of this unprecedented rise in storage systems, have been driving the electrification revolution in the automotive domain. These lithium‐ion batteries are largely used in consumer electronic devices and wearables. The energy storage research activities have been primarily focused on devising accurate algorithms to compute state‐of‐charge of these energy storage systems. State‐of‐charge helps to calculate the remaining usage time of devices and determines the range of an electric vehicle. All state‐of‐charge algorithms developed till date have used curated labelled datasets available in abundance. However, in reality, researchers do not have access to battery datasets from manufacturers due to privacy concerns. Recording battery measurements is also an expensive and herculean task. This paper introduces an auto‐correlation mechanism to compose synthetic battery datasets. This procedure is very beneficial during data scarce scenarios wherein researchers have access to finite amount of data. The leading contributions of this work include (a) synthetic battery dataset creation using auto‐correlation mechanism and (b) diverse heterogeneous data generation, which is suitable for battery capacity forecasting purposes. An auto‐correlation based synthetic battery parameter generation methodology is introduced. The battery dataset generated can be used to artificially increase training data during scarce data availability scenarios. This approach is simple and does not involve complex generative adversarial network‐based architectures to fabricate synthetic data.