Fig. 5: Demonstration of profile predictor performance.
From: Learning plasma dynamics and robust rampdown trajectories with predict-first experiments at TCV
a An example of predictions made by the profile predictor on a validation dataset shot against Thomson measurements, showing both the ability of the model to up-sample Thomson measurements and provide a smoothing effect. Error bars represent two standard deviations. b The distribution of prediction percent errors on the validation set for electron density, ne,20, and electron temperature, Te. The percent error is defined as the integrated error between the prediction and measurement, normalized to the average value of the profile. For the Te profile this is: \(100\frac{\int_{0}^{1}| {T}_{e,Thom}(\rho )-{T}_{e,pred}(\rho )| d\rho }{\int_{0}^{1}{T}_{e,Thom}(\rho )d\rho }\). Note that this metric is a pessimistic performance metric as it also captures error due to random measurement noise.
