Performance prediction of radio frequency based negative ion source using fusion neural network model

To accelerate the development of a radio frequency negative ion source (RF-NIS), a fusion neural network model has been developed to simulate and predict the performance of RF-NIS under set working parameters. The model leverages the setting parameters and diagnostic data from RF-NIS to train multiple specific neural network models, thereby establishing the fusion neural network architecture. To enhance prediction accuracy, a specialized error correction neural network model has been integrated to automatically adjust discrepancies at the decision stage. Through deep learning, the model successfully extracts the actual characteristics of RF-NIS and demonstrates superior performance in experimental tests. In engineering applications, the RF-NIS performance prediction model is utilized to predict the values of negative ion current and co-extracted electron current by extraction grid under set conditions, enabling performance simulation and qualitative analysis. This analysis investigates the effects of various influence factors on the performance of ion source to determine optimal parameter ranges. Its integration with intelligent control systems is expected to enable automatic optimization and operation of the ion source. Notably, the theoretical foundations and associated algorithms of the model are not limited to this ion source. The relevant methodology can provide a reference for prediction problems under non-linear matching conditions in fusion facilities and other application scenarios.