Dataset and Code for Hybrid GPR–CNN-Based Electron Temperature Inference from X-ray Spectra in EAST Tokamak

This repository contains the dataset and implementation code associated with the manuscript titled "A neural inference framework for electron temperature reconstruction from X-ray spectra in fusion plasmas."The dataset includes preprocessed X-ray crystal spectrometer (XCS) impurity emission spectra (from both argon and tungsten) acquired on the EAST tokamak, along with electron temperature profiles obtained from electron cyclotron emission (ECE) measurements. The data spans over 2,300 discharges under various plasma conditions.The code implements a hybrid inference model combining Gaussian Process Regression (GPR) with a non-stationary kernel and a convolutional neural network (CNN) to map impurity line intensity profiles to electron temperature distributions. Model training, validation, and synthetic benchmarking using STRAHL simulations are included.