Water Supply Flow Prediction And Control Optimization Of Supercritical Unit Based On Deep Learning

Supercritical thermal power units have significant advantages in energy utilization,power generation efficiency,environmental pollution,and grid load regulation.Therefore,they have become a hot spot and research direction in major industrial countries.Supercritical thermal power units are different from subcritical units that use steam drums to separate steam and water.Because of the internal use of once-through boilers,the feedwater control is relatively complicated.Therefore,it is of great significance to predict the feed water flow of the boiler,which can provide decision-making basis for the prediction and real-time adjustment of the feed water.This article first introduces the working principle and structural characteristics of the once-through boiler of a supercritical unit from a macro perspective,and takes the feedwater control system of a 350 MW supercritical CFB unit in a power plant in Shanxi as the research object,and the process flow and control logic are analyzed in detail.It is found that there is a strong coupling relationship between the feed water flow data and the operating parameters such as the main steam flow rate,the enthalpy value of the middle point and the unit operating load,and summarize the relevant theories of predictive modeling,which provides a theoretical basis and preparation for the predictive modeling of feed water flow.Then,the collected historical data of unit operation of the power plant from August 1 to 8,2020 is divided into a training set and a test set.After preprocessing the original data,feature selection is used as the input of the prediction model to predict the boiler feedwater flow in the short-term and medium-and long-term,and use the test set to test the model training effect.The short-term prediction uses three kinds of neural network algorithms: BP neural network,PSO-BP neural network and GA-BP neural network.The medium and long-term prediction uses five kinds of deep learning algorithms:RNN algorithm,LSTM algorithm,GRU algorithm,Bi-LSTM algorithm and EMD-LSTM Algorithms.After comparing MAE,RMSE and MAPE of the test set,it is found that EMD-LSTM algorithm has the best comprehensive prediction performance.Finally,based on the analysis of the control logic of the water supply system,the control model of the water supply system is simplified,and the particle swarm algorithm is improved by introducing the shrinkage factor,and the transfer function of the control model of the water supply system is identified.The predicted value of boiler feed water flow output by EMD-LSTM algorithm is taken as guidance and the input of the set value of the control model,a control scheme of the water supply system based on fuzzy PID control is designed.The simulation results show that the fuzzy PID control system has better stability,dynamic response performance and anti-disturbance performance than conventional PID control,and can improve the quality of feedwater control.