4th China International Youth Conference on Electrical Engineering (CIYCEE 2023)

203 / 2023-10-20 23:09:12

Inference of the fast ion 1D velocity distribution from collective Thomson scattering spectrum with Knowledge-Based Neural Network

Collective Thomson scattering,Fast ion,1D velocity distribution,Knowledge-Based Neural Network,Inversion problem

6.AI-driven technology > (1)、Intelligent Anomaly Detection, Fault Diagnosis and Prognostics

Fusion energy holds great potential as a sustainable and environmentally friendly energy source. In order to characterize the fast ion distribution in future fusion devices, the diagnostic techniques based on collective Thomson scattering (CTS) have emerged as crucial tools. In this study, an inference approach has been developed to address the inverse problem for inferring the 1D velocity distribution of fast ions. This approach combines a neural network model and a theoretical simulation model, and incorporates prior knowledge of experimental data characterization during data preprocessing. The datasets are generated by an electrostatic forward model and augmented with noise at different levels. With proper implementation, the knowledge-based neural network achieves an accuracy with over 95.45% at the noise scale of 0.5. Overall, the results demonstrate favorable outcomes in various performance metrics, highlighting the approach's robustness against noise and independence from nuisance parameters.