Decay regularized stochastic configuration networks with multi-level data processing for UAV battery RUL prediction

An effective and robust health management strategy for battery power systems is essential for ensuring the reliable operation of Unmanned Aerial Vehicles (UAVs). This paper presents an adaptive Decay Regularized Stochastic Configuration Network (DRSCN) with multi-level data processing for predicting the Remaining Useful Life (RUL) of UAV batteries. We first introduce a Multisource Signal Enhancement Analysis Framework (MSEAF) designed to efficiently extract critical battery health indicators from complex signals. A key contribution is the enhancement of the SCN model's output layer using decay regularization, which sparsifies the weights and significantly reduces the risk of overfitting in later prediction stages. To further optimize DRSCN, the Convex Lens and Dual-Mechanism Enhanced Sand Cat Swarm Optimization (CLDM-SCSO) algorithm is employed for precise hyperparameter tuning, resulting in improved prediction accuracy. Extensive experiments using the NASA HIRF battery dataset demonstrate the framework's superior accuracy and reliability compared to existing methods, offering an efficient and dependable solution for UAV battery health monitoring.