Research On Dynamic Adaptive Modeling Of Power Plant Boiler Combustion Based On Data Driven

It is estimated that thermal power will still occupy the dominant position of energy in the future.After various policies of energy saving and environmental protection have been put forward successively by the state,improving combustion efficiency and reducing pollutant emissions have become two important directions to optimize the combustion of power plant boilers.Especially after the intervention of new energy sources,more participation of thermal power generation in depth peak shaving is needed,and establishing an accurate and effective adaptive dynamic model is an important cornerstone of combustion optimization.Based on the historical data of boiler operation and data preprocessing,this paper establishes an adaptive dynamic model of boiler combustion process by using intelligent modeling methods of data mining and machine learning,which provides favorable conditions for combustion optimization.This paper mainly does the following work:(1)Data Preprocessing: The quality conservation characteristics of the coal intake and the amount of coal entering the coal mill are proposed to match the coal quality data and the operational data after the start-up;the determination of steady state data is realized by using steady state boundary parameters and statistical theoretical formula,and the denoising of operation data is realized by using wavelet denoising method.By using the partial mutual information method based on mutual information,the optimal feature variables of the model are screened.(2)Large Data Modeling: The model of coal calorific and fly ash carbon content are established by using the support vector machine(SVM)method,The model of smoke exhaust temperature and content are established by using the core support vector machine method.The core support vector machine combines support vector machine with computational geometry and is more suitable for large-scale data modeling,which improves the disadvantage of support vector machine(SVM)that is difficult to solve large-scale samples.At the same time,quantum genetic algorithm is used to optimize the parameters.A PMI-QGA-CVR model is proposed,the PMI variable selection reduces the complexity of the model,fast parameter optimization and the method suitable for large-scale data modeling,which greatly reduces the computing time and provides favorable conditions for the realization of online dynamic model.(3)Adaptive Dynamic Model: An adaptive dynamic model based on sample selection and incremental method is proposed and applied to boiler combustion model.The model utilizes the online processing advantages of the incremental method,and then through Lagrange multiplier,the core support vector machine and the sample selection of mutual information,the rapid identification of the newly added samples and the update of the core set are completed,The adaptive threshold judgment and the analysis and update of the sample realize the online dynamic learning of the model.The simulation results show that the prediction accuracy of the adaptive dynamic model is greatly improved compared with the offline model after the new sample has changed greatly.