"基于LSTM和随机森林算法的核电汽轮机组出力优化方法研究"

"Based on LSTM and Random Forest algorithm, this paper proposes a method for optimizing the output of a nuclear power turbine unit. LSTM neural network can accurately predict seasonal time series data, while Random Forest algorithm is insensitive to outliers and has strong generalization ability, making it widely used in classification and regression problems. In this study, LSTM neural network is used to predict sea water temperature time series, and Random Forest algorithm is applied to establish a regression model on the relationship between sea water temperature, power setpoint, high pressure regulator valve opening, and thermal power. By combining these two models, an optimized curve of power setpoint for the next 24 hours is obtained, which allows operators to adjust the unit's output accordingly. The effectiveness of this method is validated using historical operational data of a nuclear power unit. Furthermore, a web application for optimizing the unit's output is developed using Flask framework and deployed in the local area network of the nuclear power plant using Flask Tornado Nginx. By setting the unit's output based on the optimized power setpoint curve, the unit's output in summer can be effectively increased while ensuring that the operational parameters of the unit do not exceed the limits, thus improving the unit's economic efficiency. Keywords: nuclear power turbine unit; insufficient output; Long Short-Term Memory neural network; time series; Random Forest; web application"