Modeling of pulsed heat load in a cryogenic SHe loop using Artificial Neural Networks

Abstract

The pulsed heat load to the cryoplant is an important issue in the design and operation of tokamaks adopting superconducting (SC) magnets for the magnetic confinement, as the International Thermonuclear Experimental Reactor (ITER). The smoothing of the heat load during plasma operation is being addressed by experiments, e.g. in the HELIOS facility at CEA Grenoble, and simulations. The assessment of the operation of the cryoplant mainly requires the knowledge of the evolution of the heat load to the liquid helium (LHe) baths that are used as interfaces/buffers between the magnets cooling loops and the cryoplant itself. In this paper, an innovative approach based on Artificial Neural Networks (ANNs) is presented, leading to a simplified but fast model of the transient heat load from the magnets to the LHe baths. An ANN model is developed for the HELIOS loop and the resulting network is trained using detailed transient simulations performed with the 4C code, which was previously extensively validated against experimental data from HELIOS. The predictive capability of the (simplified) ANN model is then demonstrated by considering another, independent dataset, not used during the ANN training, and comparing the evolution of the heat load to the LHe bath computed by the ANNs with that obtained from the (detailed) 4C model.