Researchers utilize theoretical studies, supercomputing simulations, and experimental plasma turbulence measurements to understand plasma transport mechanisms. While numerical simulations based on physics can align with experimental data to some degree, discrepancies persist, limiting predictive reliability. Empirical models derived from experimental data offer another approach but may not accurately extend to future experimental devices due to data constraints. Each method has strengths and gaps, necessitating a blended approach for comprehensive plasma predictions.
To address this, researchers have adopted multi-fidelity modeling to enhance the accuracy of limited high-fidelity data by integrating abundant low-fidelity data. A novel method, nonlinear auto-regressive Gaussian process regression (NARGP), has been introduced for plasma turbulence modeling. Unlike traditional regression models relying on single input-output data pairs, NARGP incorporates multiple outputs of varying fidelity for the same input. This enables predictions of high-fidelity data by leveraging corresponding low-fidelity data.
The multi-fidelity approach has demonstrated improved prediction accuracy in various scenarios, including:
1. Integrating low- and high-resolution simulation data.
2. Predicting turbulent diffusion coefficients using experimental fusion plasma datasets.
3. Combining simplified theoretical models with turbulence simulation data.
By incorporating theory- and simulation-based predictions as low-fidelity inputs, the method compensates for gaps in experimental high-fidelity data, achieving better accuracy. These findings were published in *Scientific Reports* by Nature Publishing Group.
Traditionally, plasma turbulent transport modeling relied on either theoretical and simulation-based predictions or empirical models from experimental data. This new approach bridges these methods, combining theoretical predictability with the precision of experimental insights to forecast future nuclear fusion burning plasmas.
The multi-fidelity modeling method extends beyond fusion research. It holds promise for applications in other fields, providing a general framework for creating accurate, efficient prediction models using limited high-precision data. This could enhance performance predictions, optimize reactor designs, and spur innovation across diverse technological domains.
Research Report:Multi-Fidelity Information Fusion for Turbulent Transport Modeling in Magnetic Fusion Plasma
