Data- and theory-guided learning of partial differential equations using SimultaNeous basis function Approximation and Parameter Estimation (SNAPE)

• Published in: Mechanical systems and signal processing 2023-04, Vol.189, p.110059, Article 110059
• Main Authors: Bhowmick, Sutanu, Nagarajaiah, Satish, Kyrillidis, Anastasios
• Format: Article
• Language: English
• Subjects: ADMM optimization; Basis function approximation; Parameter estimation; Partial differential equations; Theory-guided learning
• ISSN: 0888-3270; 1096-1216
• DOI: 10.1016/j.ymssp.2022.110059

Full-field discrete measurements of the continuous spatiotemporal response of physical processes often generate large datasets. Such continuous spatiotemporal dynamic models are represented by partial differential equations (PDEs). In the past, attempts have been made to identify the PDE models from the measured response by inferring its parameters via regression or deep learning-based techniques. But the previously presented regression-based methods fail to estimate the parameters of the higher-order PDE models in the presence of moderate noise. Likewise, the deep learning-based methods lack the much-needed property of repeatability and robustness in the identification of PDE models from the measured response. The proposed method of SimultaNeous Basis Function Approximation and Parameter Estimation (SNAPE) addresses such drawbacks by simultaneously fitting basis functions to the measured response and estimating the parameters of both ordinary and partial differential equations. The domain knowledge of the general multidimensional process is used as a constraint in the formulation of the optimization framework. The alternating direction method of multipliers (ADMM) algorithm is used to simultaneously optimize the loss function over the parameter space of the PDE model and coefficient space of the basis functions. The proposed method not only infers the parameters but also estimates a continuous function that approximates the solution to the PDE model. SNAPE not only demonstrates its applicability on various complex dynamic systems that encompass wide scientific domains including Schrödinger equation, chaotic duffing oscillator, and Navier–Stokes equation but also estimates an analytical approximation to the process response. The method systematically combines the knowledge of well-established scientific theories and the concepts of data science to infer the properties of the process from the observed data. •Parameter estimation of partial differential equation (PDE) models from noisy data.•Bi-convex optimization adopts the alternating direction method of multipliers (ADMM).•Simultaneously estimates an analytical approximation of the PDE solution.•Demonstrated on a wide variety of nonlinear PDE models.•Systematically combines domain knowledge and concepts of data science.