In order to ensure the ecological security of inland river basins in northwest China, it is urgently needed to conduct assessments on the safety of sudden water pollution in inland river basins in northwest China. This study focuses on the Dunhuang area of the Shule River Basin. By applying the pressure-state-response (PSR) model framework and based on historical data from 2017 to 2022 in this basin, a radial basis function (RBF) neural network model integrated with whale optimization algorithm-simulated annealing (WOA-SA) was used to assess the risk level of sudden water pollution in this area. The assessment results are compared with those obtained from particle swarm optimization-RBF (PSO-RBF), genetic algorithm-RBF (GA-RBF) neural network models, and the traditional Technique for order preference by similarity to ideal solution (TOPSIS) method. The analysis results show that the risk level of sudden water pollution in the Dunhuang section of the Shule River is rated as Grade Ⅱ between 2017 and 2018, and then decrease to Grade Ⅲ from 2019 to 2022, indicating a trend of gradually decreasing and stabilizing risks. The results are consistent with those of the TOPSIS method and align with the basin management situation, effectively validating the accuracy of the assessment model. The research findings are helpful in improving the prevention and control capabilities and emergency response efficiency of the Shule River Basin in response to sudden water pollution incidents. They are of great significance to water resources management in northwest inland river basins and ecological protection in the Qilian Mountain area.