Abstract
This paper explores battery cycle life prediction by considering constrained-resource environments. Edge devices are characterized by constrained resources such as low power supply, limited communication bandwidth, real-time application needs, and small-sized models. These constraints differ significantly from the conditions typically associated with regular devices or cloud computing environments, which often do not rely on battery for power supply. Despite the advancements in Long Short-Term Memory (LSTM) models for battery cycle life prediction, there is a gap in literature regarding its application on edge devices. To address this gap and enhance understanding of battery cycle life prediction within such resource-limited environments, this paper studies the effects of quantization and pruning techniques on LSTM models. Quantization is employed to change model from high-precision floating-point representation to low-precision, while pruning aims to reduce model size. Experimental results indicate that the proposed method can speed up inference times, though it also reveals that prediction accuracy is slightly compromised and dependent on the tiny LSTM models’ parameter settings. The proposed method is promising to accelerate battery cycle life prediction on edge devices. Given the multiple characteristics associated with constrained resources on edge devices, future work will consider additional factors such as memory usage and power consumption.