Research On Online Modeling Method Of Ultra-supercritical Coal-fired Boiler Based On Industrial Big Data

The NO_x generated during the operation of coal-fired power station boilers is one of the main sources of NO_x in the atmosphere.With the situation of energy saving and environmental protection becoming more and more serious,as one of the main technologies of energy saving,environmental protection and efficiency improvement in thermal power generation,supercritical and ultra-supercritical unit technology has become its primary consideration.Due to system power peaking and other factors,the unit often operates under variable load conditions.When the load fluctuates greatly,the combustion process is unstable,which leads to an increase in NO_x emissions from combustion.In order to optimize the combustion process,thereby reducing NO_x production and improve combustion efficiency,it is necessary to establish a model that can effectively reflect its characteristics.Then the operation mechanism of coal-fired power station boilers is analyzed,and the combustion process is mainly discussed to determine the main variables affecting NO_x production.The variables are selected reasonably,and it also provides effective sample data for subsequent modeling work.Due to mechanism modeling complexity and a large amount of professional knowledge requirement,a modeling study on NO_x emissions of a 1000 MW ultra-supercritical coal-fired power station boiler was carried out from the perspective of data-driven modeling.For this purpose,this paper studies the problem of NO_x prediction problem based on support vector regression.Aiming at the problem that the parameters of support vector regression prediction model are difficult to determine,the parameter optimization methods based on particle swarm optimization(PSO)and grid search are studied respectively.Compared with the grid search method,the advantages of this method are demonstrated.The research results show that the method has good approximation accuracy and generalization ability.Considering the applicability in the context of industrial big data and the time information contained in the sample data,a deep learning algorithm based on Long Short-Term Memory(LSTM)is proposed for modeling.After model training and verifying it on the test set,LSTM has stronger learning ability and generalization performance compared with the prediction results of neural network and support vector regression model.In the operation of boilers,there will be equipment aging and other operating parameters changes,which will result in the inapplicability of the model to the current state.At this point,the online learning algorithm is an effective solution.By learning the real-time data,the model is continuously updated to adapt to the current state.Based on the traditional incremental online support vector regression algorithm,an adaptive update rule is proposed to update the model,which avoids the repeated learning of the trained sample points to a certain extent and improves the model update efficiency.Combined with the characteristics of field operation data and simulating the actual operation situation,an adaptive online support vector regression model for NO_x emission is established by using the boiler operation data.The results show that the model has good prediction accuracy,can track the changing operation characteristics of the boiler well,and can meet the requirements of on-line NO_x prediction.Compared with the traditional online support vector regression algorithm,the updating efficiency of the model is improved,and the accuracy of the model is also improved.In summary,this paper fully demonstrates the feasibility of online prediction of NO_x emissions from coal-fired power station boilers based on data-driven modeling.The research results have important theoretical guidance and practical application value for reducing NO_x emissions and improving the operation efficiency of power plants.