Modeling Research Of Medium Speed Mill Coal Pulverizing System Based On Data Mining

The establishment of effective dynamic model which can accurately predict the outputs of pulverizing system will be beneficial for the control optimization,economic and safe operation of the power plant.Existing pulverizing system models disregard the heat transfer effect on pulverized coal mass flow.In some models,the raw coal moisture is disregard or is soft sensed with various empirical parameters and might be inaccurately sensed in some circumstance.This research proposes a medium coal mill model that considers the fluctuation of coal moisture by a built-in moisture soft sensing method and the heat transfer effect on coal balance in mill.The proposed model realizes the dynamic calculation of raw coal moisture and can predict the outputs of pulverizing system with high accuracy.The specific contents of this paper are as follows:Black box models of primary air temperature and primary air mass flow based on long and short-term memory network are proposed and compared with mathematic grey box model.Results show that on test dataset,the ratio of data which the deviation between prediction and measured primary air mass flow is smaller than 0.5 kg/s is increased by 23.14%;the ratio of data which the deviation between prediction and measured primary air temperature is smaller than 5? is increased by 18.75%.A new coal mill model is established based on the principles of conservation laws.A dynamic soft sensing method of coal moisture considering the internal energy of coal mill is included in the model.The model considers the heat transfer effect on coal mass flow by correlating the outlet temperature with outlet pulverized coal mass flow.The model is tested on large amount of historical data and compared with previous coal mill models.Results show that the proposed model respectively increases the prediction accuracy of differential pressure by 5.58% and 0.95% on two datasets,and the mean absolute error of outlet temperature is around 0.05%.The model covers the whole operation of coal mill from starting up to shutting down.Dynamic simulation test show that the proposed model can reflect the dynamics of coal mill well.A coal particle images detection method based on fully convolutional neural networks is proposed.Compared with traditional threshold-based method,the proposed method has better detection accuracy of small particles with diameter smaller than four pixels on simulated images,calibration images and experimental images.In this research,an accurate model of medium speed coal mill pulverizing system is established and an accurate holographic coal particle images detection method is proposed.The work can be applied for advanced control system design and fault diagnosis of pulverizing system,and for measuring large-scaled pulverized coal fineness with high precision by on-line coal fineness measurement device based on digital holography.