| The state has encouraged the comprehensive development and utilization of the associated resources of the coal.At present,the circulating fluidized bed(CFB)technology is the best industrial method for comprehensive utilization of coal slime.Moreover,mix-burning low-cost coal slime is also an important means to improve the economy of CFB units.However,the fluctuation of control parameters in mix-burning process has brought challenges to the operation of units.How to adjust control parameters to achieve economic optimization under the premise of ensuring safe and stable operation is of great research significance.The characteristics of large inertia,large delay and strong coupling of mix-burning coal slime CFB units make it difficult to model the mechanism,let alone form a universal control optimization system.The deep integration of big data,artificial intelligence and power generation industry is an important means for the application and promotion of intelligent power plants.At present,the overall modeling and optimization research on mix-burning coal slime CFB units based on actual process data are rare.Therefore,it is of great significance to mine process data in depth and put forward an operation assistant information system to fill the gaps in this kind of research.Based on the process data stored in the distributed control system,the stability,economy and safety of mix-burning coal slime CFB units are regarded as the objectives.By using data-driven modeling,data mining optimization,expert system guidance and state intelligent early warning technology,the overall operation supervision and intelligent early warning schemes of mix-burning coal slime CFB units are innovatively put forward.Based on the above contents and ideas,the research is carried out from the following aspects:(1)Establishment of comprehensive economic model in operation supervision systemThe sum of the fuel cost,the desulfurization and denitrification cost and the auxiliary power cost is taken as comprehensive economic indicator.On the basis of data preprocessing and feature selection based on partial mutual information method,the traditional data-driven algorithms such as Elman neural network,support vector machine and least squares support vector machine are used to establish black-box models between control variables and comprehensive economy respectively,which are made a comprehensive comparative analysis.On the basis of the best-performed least squares support vector machine algorithm,an improved strategy is proposed:the improved grid search method and model updating strategy are used to improve the prediction accuracy and adaptive ability of the model.The trend and scope of comprehensive economy are further analyzed by fuzzy information granulation.(2)Construction of operation supervision systemThe operation supervision system of mix-burning coal slime CFB unit is composed of the operation database,the model algorithm database and the expert knowledge base.Based on the comprehensive economic model,the off-line expert knowledge base is composed of typical steady-state conditions optimized by genetic algorithm.The improved algorithm of fuzzy association rules mining introduces the"utility cost",an evaluation index of association rules.Further,the data in expert knowledge base are fuzzified and the association rules are mined.The association rules between variables under steady states with optimal comprehensive economy are selected and input to the fuzzy logic controller.After receiving the load instructions,the on-line fuzzy logic controller outputs the target value of control variables under steady states with optimal comprehensive economy,which provides operation guidance and information reference for the unit operation.(3)Establishment of state prediction model in intelligent early warning systemThe state prediction model is the model foundation of the intelligent early warning system of equipment state.The difference information between the accurate normal state prediction model and the observed state implies the early fault characteristics.The multivariate time series prediction,the fuzzy inference prediction,the multivariate state estimation technology and the improved multivariate state estimation technology are used to predict the normal state parameters of equipment,and the models are compared with each other.In terms of model prediction accuracy,the improved multivariate state estimation technique adopting probability density to construct process memory matrix and the multivariate time series presdiction method are superior to the other two methods;In terms of model complexity and operation speed,the improved multivariate state estimation technique using state vector as operation unit does not need to model one by one parameter,which is far superior to the other three methods.(4)Construction of intelligent early warning systemThe intelligent early warning system realizes its function through state prediction,state judgment and variable positioning.In the research of state judgment,the state judgment methods based on the adjustable smoothing parameter,the K-means clustering and the sliding window similarity,are compared.In comparison,the proposed sliding window similarity method has the most advantages in the accuracy,timeliness and simplicity of early warning.Based on the output of the state prediction model,the sliding window similarity function describes the similarity between the normal states and the observed states,by using the inverse function of Euclidean distance between states.The weights of fault information obtained by analytic hierarchy process is assigned to the variables in similarity function,and the sliding window method is used to eliminate random repeated early warning errors.The proportion of abnormal variables markers calculated after early warning,is designed to locate and diagnose fault variables,and the fault causes are obtained by combining variable information and on-site maintenance.Finally,the influence of sliding window parameters on early warning sensitivity is analyzed and discussed. |
PDF Full Text Request