Adapting physics-informed neural networks to improve ODE optimization in mosquito population dynamics.

Area of Science:

• Computational Science
• Ecological Modeling
• Machine Learning

Background:

• Physics-informed neural networks (PINNs) integrate physical laws into data-driven models.
• PINNs offer data efficiency and robustness for complex dynamical systems.
• Existing PINN frameworks struggle with real-world ordinary differential equation (ODE) systems, particularly those with multi-scale behavior.

Purpose of the Study:

• To propose an improved PINN framework for ODE systems.
• To address challenges in modeling mosquito population dynamics.
• To enhance solutions for both forward and inverse problems in complex ODE systems.

Main Methods:

• Developed a novel PINN framework with specific improvements for ODE systems.
• Addressed gradient imbalance and stiffness issues inherent in mosquito population ODEs.
• Implemented a time domain expansion strategy to resolve time causality problems in PINNs.

Main Results:

• The proposed framework effectively models mosquito population dynamics.
• Preliminary experiments with simulated data demonstrate the approach's effectiveness.
• The method shows promise in handling multi-scale behaviors and stiff ODEs.

Conclusions:

• The enhanced PINN framework shows significant potential for ecological system modeling.
• Physics-informed machine learning can advance the study of complex ecological dynamics.
• The approach offers a robust solution for ODE systems with challenging characteristics.