| In order to accept more new energy sources and improve the flexibility of peak and frequency regulation of power grid,China's energy structure has changed.Coal-fired power generating units will operate more in low load and variable load conditions,which will cause significant changes in unit performance and control mode.In-depth analysis of the operating characteristics of coal-fired thermal power units under global operating conditions is helpful to research and develop new optimal control strategies.The strategy can tap the energy saving potential of the unit and realize the safe,stable,energy saving and environmental protection operation under a wide load range.This strategy has also become an important measure for thermal power generation to actively adapt to the role change and enhance market competitiveness under the new situation.The optimization of boiler combustion involves three aspects:safety,economy and environmental protection.The current research focuses on a single aspect and lacks comprehensive consideration of these three aspects.Firstly,the influence of slag state on boiler combustion heat transfer model is considered from the aspect of safety.On this basis,an online identification system is designed to identify slagging rate.Then,the prediction model of boiler combustion process and SCR denitration system based on deep learning neural network was established by usinglarge data information in DCS system.On this basis,the precise control strategy of ammonia injection volume was proposed to avoid the mismatch between ammonia injection volume and NOx emission in SCR system and ensure its environmental protection.Finally,according to the actual needs of the field operation,the theory is combined with the actual engineering,the offline optimization is combined with the online optimization,the online real-time boiler optimization is realized,and the boiler combustion economy is improved.The main research contents are as follows:1.Aiming at the slagging problem in complex combustion process,a theoretical model of boiler combustion heat transfer in slagging state is established.Based on the characteristic model and the adaptive golden section method,a recognition system for slagging of the heating surface was designed.Combining the on-line identification with the boiler combustion heat transfer simulation module based on CFD,a new control strategy is provided for the slagging condition in the combustion process which cannot be measured by the equipment in real time.A identification method is provided for adjusting boiler operating parameters reasonably,helping operators to know the safety state of combustion in time and preventing accidents caused by slagging.2.A dynamic model for NOx emission prediction of coal-fired boiler based on hybrid LSTM and CNN neural network is constructed.The original combustion data sample is decomposed into a smooth approximation component and a series of detail components by using wavelet transform(WT)signal processing technique.A dynamic model with approximate components was established by using LSTM deep network to predict the overall trend of NOx emission.At the same time,three CNN neural networks were used to carry out dynamic modeling for multiple detail components respectively to predict the characteristic information of NOx emission.Finally,the two prediction models are fused to get the final NOx emission model.Simulation results show that this method can achieve accurate and stable modeling and good prediction performance.Compared with the typical modeling methods,this model has better versatility and repeatability.3.In order to make full use of historical information and future information,the influence of input variables on output is considered comprehensively,and the delay time of each input variable is estimated by dynamic joint mutual information(DJMI).Bi-LSTM deep learning algorithm is used to predict NOx emission from SCR system outlet of coal-fired boiler,and the prediction accuracy is improved.The NOx model for predicting t+3 time in the next 3min is also established.The simulation results confirm that the prediction model is significantly ahead of the current time waveform,and the advance time can fully meet the requirements of actual ammonia injection control.The model can be used to adjust the amount of ammonia injection in time,which has a guiding significance for reducing pollutant emission and the cost of coal-fired units.4.The boiler combustion model and SCR system model are integrated through the NOx emission at SCR inlet to form an integrated dynamic model for predicting NOx emission at SCR outlet.The intelligent feedforward control system is constructed by using it as the intelligent predictive feedforward signal to control the SCR ammonia injection accurately.The simulation results show that the integrated intelligent feedforward predictive control method is effective and the ammonia injection control is stable.This control method can meet the control requirements of industrial objects with large inertia and large delay.5.An on-line boiler combustion optimization method based on grey relational theory(GR-CBR)is proposed.The global optimization algorithm was used to set up the optimization case base offline.Combining subjective and objective factors,genetic algorithm was used to optimize CBR feature weight.This optimization measure improves the retrieval accuracy and can retrieve the case matching with the target case from the huge case base.At the same time of ensuring the stable combustion of the unit,considering the boiler combustion efficiency and NOx emission concentration,the second and third damper baffle opening instruction and the fixed value of oxygen are reasonably given.This will be able to achieve stable and economical combustion of the boiler.The whole system is applied to a 350 MW coal-fired generating unit,which simplifies the process of optimization calculation,has the advantages of short optimization time and high stability,and is suitable for online real-time optimization. |
PDF Full Text Request