Dynamic Modeling of Cable-Suspended Payload Quadcopter: A Physics-Informed Deep Learning Approach

This paper introduces a novel methodology for modeling Cable-Suspended Payload Quadcopters (CSPQs). Although cable-suspended systems offer significant operational advantages, they are characterized by increased modeling complexity arising from payload swing dynamics and intricate aerodynamic effects. To address these challenges, we propose a hybrid modeling framework that integrates physics-based modeling with deep learning techniques. Specifically, our approach employs a Physics-Informed Neural Network (PINN) architecture, leveraging an attention-enhanced Temporal Convolutional Network (PIATCN), in which the quadcopter payload dynamics are incorporated as a physics-informed regularization term. This framework is capable of capturing complex, nonlinear behaviors while ensuring physically consistent predictions and reducing the volume of required training data. The proposed method is evaluated by comparing its state prediction accuracy against several alternatives, including models based on first-principles physics, recurrent neural network architectures, and comparable data-driven models that do not incorporate physics-based regularization.