일과 에너지 및 각운동량을 이용한 Lagrangian PINNS
Lagrangian PINNs Using Work and Energy, Angular Momentum
1
조회
0
다운로드
0
인용
0
참고문헌
키워드
초록
This study proposes a novel model to address the convergence issues of conventional Lagrangian Neural Networks (LNNs), which are sensitive to variations in initial conditions. To enhance physical consistency and inference capability, the proposed model integrates stronger physics-based constraints with a dual-network architecture: a single-layer Lagrangian network for inferring generalized forces through nonlinear transformation layers, and a multi-layer external force network for predicting applied forces. The loss function incorporates the principles of work-energy and angular momentum to enforce physical plausibility. The model is evaluated on a single pendulum system and demonstrates stable convergence across diverse initial conditions. However, robustness to variations in system parameters such as mass and length remains a subject for future investigation.
Abstract
This study proposes a novel model to address the convergence issues of conventional Lagrangian Neural Networks (LNNs), which are sensitive to variations in initial conditions. To enhance physical consistency and inference capability, the proposed model integrates stronger physics-based constraints with a dual-network architecture: a single-layer Lagrangian network for inferring generalized forces through nonlinear transformation layers, and a multi-layer external force network for predicting applied forces. The loss function incorporates the principles of work-energy and angular momentum to enforce physical plausibility. The model is evaluated on a single pendulum system and demonstrates stable convergence across diverse initial conditions. However, robustness to variations in system parameters such as mass and length remains a subject for future investigation.