In order to monitor the heat transfer efficiency of the direct air-cooled radiator，the method of modeling is used to study and predict the heat transfer performance of the air-cooled radiator. The outlet temperature of the air-cooled radiator can indirectly reflect its heat transfer capacity，which can be used as an evaluation index of the heat transfer performance of the air-cooled radiator. According to the analysis of the radiator heat transfer model and the influencing factors of heat transfer performance，the Radial Basis Function (RBF) neural network model was established with environmental wind speed，environmental temperature，fan speed，exhaust steam pressure，exhaust steam temperature and unit load as the input and outlet temperature as the output. In order to avoid the model falling into local optimum，Particle Swarm Optimization (PSO) algorithm was used to optimize the parameters of RBF neural network，and a large number of air cooling tower operation data were used to train the RBF neural network，and then simulation verification was carried out. Experimental results show that the MAE and RMSE of the optimized model are the lowest, and the comparison with RBF and PSO-BP models verifies the superiority of the proposed algorithm in temperature prediction.