A probabilistic forecasting approach towards generation of synthetic battery parameters to resolve limited data challenges. Issue 4 (2nd November 2021)

• Record Type: Journal Article
• Authors: Naaz, Falak; Channegowda, Janamejaya
• Is Part Of: Energy storage. Volume 4:Issue 4(2022)
• Page Start: n/a
• Page End: n/a
• Publication Date: 2021-11-02
• Subjects: battery -- energy storage -- state‐of‐charge -- synthetic data; Energy storage -- Periodicals; Energy storage; Periodicals; 621.04205
• DOI: 10.1002/est2.297
• Languages: English
• ISSNs: 2578-4862

Abstract

Abstract: Advent of multiple data‐driven techniques in the energy storage domain has resulted in the development of accurate battery capacity estimators. A major impediment to research in this domain is the limited availability of diverse datasets. Generation of synthetic data is vital when experimental data is unavailable. Moreover, privacy concerns and confidentiality restrictions have further fueled data scarcity in this domain. Present state‐of‐the‐art techniques for synthetic data generators are primarily built on Generative Adversarial Networks (GANs). Although the performance of GANs are exceptional, the high dimensional outputs of these models are difficult to interpret and the associated loss function is highly data specific and cannot be generalized. Moreover, GANs consist of two resource‐intensive neural networks (generator and discriminator) which renders them unsuitable for devices with limited computational resources. In this article, we introduce a simple deep learning‐based probabilistic time series model and we employ the forecasting ability of this model to generate synthetic data. The simplicity of this model makes it an effective candidate to produce synthetic data in resource‐constrained scenarios.