Battery Sensing Data Recovery Model Based on Self-learning and Feature Transfer

With the development of industrial production towards large-scale, refined, and intelligent directions, the health management and operation maintenance of industrial equipment have become increasingly important. However, factors such as equipment failures and environmental disturbances in industrial application scenarios can lead to missing operational state data of the equipment, thereby hindering the application of data-driven methods in equipment health state management. This paper, based on the context of data missing due to battery sensor failures, proposes a data recovery model based on self-learning and feature transfer. First, multiple health indicators are extracted from battery operational data, and missing scenarios for health indicator sequences and capacity data are simulated. Then, the data recovery accuracy of self-learning and feature transfer methods is compared across various data missing scenarios and ratios. Finally, the final model is constructed using a selection strategy. The proposed model demonstrates excellent data recovery performance across different scenarios, and the recovered key variables enable accurate assessment of the equipment degradation process.