Coal-fired power plants today face the dual challenges of reducing coal energy consumption and controlling ultra-low pollutant emissions.Combustion optimization of boilers can improve the combustion state of boilers and coordinate the integrated control of boiler thermal efficiency and pollutant emission levels.Due to the nonlinear,time-varying and strongly coupled characteristics of boiler combustion system,it is difficult to establish an accurate model of the internal mechanism of boiler combustion system when using traditional methods for combustion optimization because the combustion characteristics inside the boiler system cannot be fully considered,and there are problems such as model mismatch and mismatch of input and output data.Artificial intelligence algorithms can model and learn from the actual system and use operational data to train and adjust the model parameters for prediction or optimization purposes,and have been widely used in the field of boiler combustion optimization in recent years.However,the focus of boiler combustion optimization at home and abroad is generally on the study of algorithms for the whole combustion optimization process,and the conclusions reached are often just multiple sets of optimal operating parameter solutions,which cannot be applied to field practice.This paper focuses on the development of a boiler combustion optimization system,aiming to provide boiler operators with a comprehensive solution for the analysis,operation and control of boiler combustion operation optimization adjustment.The specific research results are as follows:(1)Boiler combustion is a multivariable,nonlinear,strongly coupled complex process,and establishing a model that accurately describes the input-output relationship of the combustion system is the focus and difficulty of combustion optimization.A 300MW subcritical circulating fluidized bed thermal efficiency and NOx emission prediction model is established based on the historical operation data in DCS using Long and short-term memory neural network in deep learning,and a genetic algorithm(GA)is used to optimize the model parameters for the problem that the parameters of the Long and short term memory neural network(LSTM)model are difficult to determine.And the established prediction models are compared with other algorithmic models on the same test sample set.The NOx emission prediction model based on GA-LSTM has RMSE of 33.128,MAPE of 0.013,and R-square of 0.9962in the test sample set,and the boiler thermal efficiency prediction model has RMSE of0.271,MAPE of 0.012,and R-square of 0.9973 in the test sample set.The prediction ability and fitting accuracy of the GA-LSTM model in the test set are better than those of other models after preprocessing the data and performing parameter search by genetic algorithm(GA),and the GA-LSTM model can better complete the multivariate nonlinear fitting.(2)On the basis of the established boiler thermal efficiency and NOx emission prediction models,a multi-objective evolutionary algorithm(IDBEA)based on indicators and congestion distance is used to optimize the primary air flow,secondary damper opening,coal feed rate and furnace outlet oxygen,and to obtain the best combination of each parameter under different operating conditions.Based on the indicator-based multi-objective evolutionary algorithm,the congestion distance selection strategy is added to improve the algorithm diversity.IDBEA was used to optimize three groups of nine working conditions at low,medium and high loads with a focus on NOx emission reduction.From the optimization results,the highest NOx emission was reduced by 96.37 mg/m~3compared with the original actual operating conditions,and the average decrease of all conditions was 17.35%,which is an obvious optimization effect;meanwhile,the highest thermal efficiency was improved by 1.9%,and the average increase of all conditions was 0.88%,and the overall optimization effect is in line with the purpose of combustion optimization with NOx emission as the main focus.The multi-objective evolutionary algorithm based on IDBEA can keep the boiler thermal efficiency no less than the original actual operating level or even slightly increase,while significantly reducing the NOx pollutant emissions generated in the boiler combustion operation,which has a very good boiler combustion optimization effect.(3)Based on the established multi-objective optimization evolutionary algorithm as the underlying foundation,the boiler combustion optimization system was designed and developed.The business as well as operational performance of the system was analyzed,and an overall solution of the system was given.The boiler combustion optimization system is added to the existing control system of the power plant,which can read the boiler operation parameters collected by the original DCS control system without affecting the safety and stability of the original operating equipment of the power plant,and then use the underlying multi-objective evolutionary algorithm mathematical model of the system to achieve the purpose of combustion control optimization.The system adopts B/S architecture and is mainly developed based on Vue front-end framework and Spring boot back-end framework.The modular concept is applied to the whole system,and the system login module,operation monitoring module,prediction comparison module,system monitoring and system management module are designed and developed.The boiler combustion optimization system integrates online monitoring,operation and maintenance,and optimization management.Through the operation interface,the calculation results of the combustion multi-objective optimization mathematical model and the actual optimization effect can be clearly displayed in front of the boiler operators,which is convenient for the staff to use the boiler combustion multi-objective optimization mathematical model to control the actual production process. |