Data Driven Based Reheater Temperature Modeling And Optimal Control

Reheater steam plays a vital role in the power plants.It can increase the efficiency of the thermal by 2%;on the other hand,it can also reduce the steam humidity and improve the safety of the final stage's blade.It is important to control reheater steam temperature within a certain range.This is because that the high reheater steam temperature can cause damage to the metal material,while the low steam temperature can reduce the efficiency of the thermal cycle.The distributed control system and the supervisory information system has stored the reheater's process data,which offer a fundamental for analysis,mining and application of reheater.From the view of data,maintaining the reheat steam temperature within a specified range is an effective means to ensure the safety and the reliability of the unit operation.Based on the reheater operation data,this thesis mainly studies the reheat steam temperature trend prediction and the control of the reheat steam temperature regulation system.The specific research contents in this thesis are as follows:1.A trend prediction model for reheat steam temperature is proposed.For the noise data during the operation of the reheater,the method of abnormal value processing and standardization processing of the operation data is established.Aiming at the nonlinearity of reheat steam temperature,a reheat steam temperature trend prediction model with deep neural network is presented.For the large inertia of reheat steam temperature,a longer historical time window is taken as the input of the model.Considering that the delay order of different input parameters of the model is uncertain,the optimal time window size by genetic algorithm is put forward.Finally,the consistency of the genetic algorithm search method in different units was verified and the accuracy of the prediction model was evaluated.2.A control strategy based on imitation learning with data is presented.For the smooth control of reheater temperature regulation system,a smoke baffle is investigated.In this method,the PID control strategy as the expert strategy for labeling,and the robustness of the neural network is used to realize the smooth control strategy of the learning system from the actual running data.The control performance of the imitation strategy under different learning tasks,different state noises and different PID control parameters is analyzed.3.The framework of the reheat steam temperature trend forecasting platform is designed.In the framework,the real-time data is transmitted to the prediction platform in the form of a data stream,and the prediction unit calculates the predicted value of the reheat steam temperature.The predicted result is stored in a database for displaying in the form of a webpage.Finally,the simulation results demonstrate the effectiveness of the steam temperature prediction model and the imitation learning control strategy,and establish a reheat steam temperature trend prediction platform.