Abstract: In order to solve the problem of insufficient output of a nuclear power plant in summer in China, a method for optimizing the output of nuclear power steam turbine based on nonlinear autoregressive neural network and random forest algorithm was proposed. Nonlinear autoregressive neural network can achieve accurate prediction of seasonal time series. Random forest algorithm is not sensitive to outliers and has strong generalization ability, which is widely used in classification and regression problems. The nonlinear autoregressive neural network was used to establish a seawater temperature time series prediction model, and the random forest algorithm is used to establish a regression model of the relationship between the seawater temperature and electric power set value on the opening of the high-pressure regulating valve and heat power. The two models were combined to obtain the optimized curve of the electric power set value in the next 24 h, and the unit operator can adjust the output of the unit according to the optimized curve. Through the historical data of the nuclear power plant, the effectiveness of the method was verified. Using the electric power set value optimization curve to set the unit output will effectively increase the unit output in summer and improve the unit economy under the condition that the unit operating parameters are not exceeded.