Physics-Informed Neural Networks 연구 동향 및 농업 분야 발전 방향 = Status and Development of Physics-Informed Neural Networks in Agriculture

다국어 초록 (Multilingual Abstract)

Mathematical modeling is the process of representing physical phenomena using equations, and it often describes various scientific phenomena through differential equations. Numerical analysis, which is capable of approximating solutions to partial differential equations representing physical phenomena, is widely utilized. However, in high-dimensional or nonlinear systems, computational costs can substantially increase, leading to potential numerical instability or convergence issues. Recently, Physics-Informed Neural Networks (PINNs) have emerged as an alternative approach. A PINN leverages physical laws even with limited data to provide highly reliable predictive performance and can address the convergence issues and high computational costs associated with numerical analysis. This paper analyzes the weak signals, research trends, patent trends, and case studies of PINNs. On the basis of this analysis, it proposes directions for the development of PINN techniques in the agricultural field. In particular, the application of PINNs in agriculture is expected to be more effective than in other industries because of their ability to reflect real-time changes in biological processes. While the technology readiness level of PINNs remains low, the potential for model training with minimal data and real-time prediction capabilities suggests that PINNs could replace traditional numerical analysis models. It is anticipated that the research and industrial applications of PINN will develop at an increasing pace while focusing on addressing the complexity of mathematical models in agriculture, mathematical modeling and the application of various biological processes; securing key patents related to PINNs; and standardizing PINN technology in the field of agriculture.