Decay Regularized Stochastic Configuration Networks with Multi-Level Data Processing for Uav Battery Rul Prediction

Abstract

A precise and durable health management strategy for battery power systems is crucial to maintaining the reliable operation of Unmanned Aerial Vehicles(UAVs).This paper introduces an adaptive Decay Regularized Stochastic Configuration Networks(DRSCN) with Multi-level data processing for UAVs Battery Remaining Useful Life(RUL) Prediction. Firstly, a Multi source Signal Enhancement Analysis Framework (MSEAF) is effectively presented to criticalbattery health indicators from complex signals. Moreover, a significant innovation is enhancing the SCN model’s output layer, which introduces Decay Regularization to sparsify the layer’s weights, significantly reducing the risk of overfitting in later prediction stages. To further re fine and optimize DRSCN, a Convex Lens and Dual-Mechanism Enhanced Sand Cat Swarm Optimization (CLDM-SCSO) algorithm is employed for precise hyperparameter tuning, fur ther improving prediction accuracy. Finally, extensive control and ablation experiments on the NASA HIRF battery dataset validate the framework’s superior prediction accuracy and relia bility over existing methods, offering an efficient and reliable UAV battery health monitoring solution.

Keywords: UAV battery, RUL prediction, Stochastic Configuration Networks, signal decomposition