Research on Predicting the Impact of Different Impurity Seeding Rates on Divertor Detachment Based on Ensemble Learning

Divertor detachment is a key means to control the heat load on the target plate of tokamak devices. However, the traditional boundary simulation code SOLPS-ITER has limitations of long simulation time and high computational resource consumption, making it difficult to meet the demand for fast prediction. Taking the HL-2A tokamak as the research object, this paper constructs a weighted average fusion model of Random Forest and Multilayer Perceptron (MLP) based on the simulation data generated by SOLPS-ITER to predict the target plate electron temperature and target plate particle flux. The results show that the fusion model achieves a stable mean absolute percentage error (MAPE) of around 3% and fast operation speed. Feature importance analysis verifies the regulation law of impurity seeding rate on the weight of input features, which is consistent with the physical process simulated by SOLPS-ITER. This method provides a new path for fast and high-precision prediction of the heat load on the divertor target plate, and has important reference value for the physical design and real-time regulation of the divertor in fusion devices.