Prediction Of Condenser State Variables Based On Hybrid Model

Condenser is an important auxiliary equipment of generator set,and its operation directly affects the economy and safety of power plant operation,so it is necessary to use historical data to predict the variable of condenser.At present,the research on condenser mainly uses simulation software to simulate flow field operation analysis,or establish a mechanism model according to the law of energy conservation and heat transfer,and also uses single data-driven model and optimization algorithm to study the prediction and fault detection and diagnosis of condenser vacuum and cleaning coefficient.However,in the process of establishing the mechanism model,appropriate assumptions will be made according to the boundary conditions,and some measuring point information can not be accurately obtained,which leads to the large and unadjustable error of the mechanism model,and the single data-driven model is easy to fall into the local optimal and weak generalization ability.Therefore,it is very important to achieve accurate prediction of condenser parameters.In this paper,a condenser parameter prediction model based on the mixed model is established to accurately predict the outlet temperature of circulating water.Firstly,the characteristic parameters related to the operation of the condenser in a 1000 MW unit were selected,and the characteristics of the data were analyzed,the outliers were processed,and the data were normalized.Then,the wavelet threshold denoising method was used to smooth the data,and the feature selection model based on Elastic Net CV was established to reduce the data dimension and remove the redundant data.Secondly,a Stacking framework based prediction model is established.SVM based learner,Ada Boost based learner and GBDT based learner are selected for the first layer of the model,and PSO-GRNN model,i.e.,PSO-GRNN model,is selected for the second layer of the model.After introducing the principle and mathematical description of the single model,The integrated learning method of Stacking model is expounded.Verify the validity of the PSO-GRNN model by comparing THE LSTM and PSO-SVM with the PSO-GRNN model,compare the SVM,Ada Boost,GBDT,GRNN,PSO-GRNN with the Stacking model,It is proved that Stacking multi-model fusion has better predictive performance.Then,the mechanism model of condenser tube side was established and simulated by simulink platform of Matlab software.The prediction effect of mechanism model was not ideal due to parameter acquisition and mechanism hypothesis.Then,the GBDT error compensation model based on the mechanism model is established,and the data driven model is used to compensate the mechanism error,reduce the prediction error,and make the prediction result more accurate.Finally,the prediction model based on Stacking model and mixed model was established,and the prediction effect was compared with p SO-GRNN model,Stacking model and GBDT error compensation model.In addition to the data of June,the data of March were also selected to analyze the prediction effect in different seasons to verify the robustness of the model.It has been proved that for the data set selected in this paper,the hybrid model based on Stacking framework has the lowest error and higher prediction accuracy,which proves that the hybrid model of mechanism model and multiple datadriven models can effectively improve prediction accuracy when dealing with complex systems compared to a single system.