A dynamic sensitivity analysis scheme for air absorbed dose rate with multivariate time-series prediction model and SHAP method
摘要
A dynamic sensitivity analysis scheme combining the multivariate time-series prediction model and the SHAP method is designed. In this paper, the impact of meteorological features on the air absorbed dose rate is studied as an example. The results show that the overall performance indicators of the BiLSTM-Attention-L2 model are better than other models. In Guangzhou city, the air absorbed dose rate is most significantly influenced by air temperature, which has a positive contribution. In Shaoguan city, the air absorbed dose rate is primarily influenced by humidity and wind speed, with humidity having a negative contribution.