Soft Sensing Of Mill Load Based On Multi Source Feature Extraction And Machine Learning

One of the most important equipment of grinder as grinding ore material widely used in chemical,mineral,energy and building materials industries.Mill load is an important parameter of grinding mill,grinding is directly related to the stable operation,which affects the production efficiency and product quality of mill.Due to the mill is a serious nonlinear,multivariable coupling system,we often use some indirect means to measure the mill load,despite a certain effect,but due to mining hidden information in the sample data,unable to obtain satisfactory measurement results.This paper presents a feature extraction and machine learning of mill load soft sensing method based on,make full use of mill data contains useful information,and according to two kind of mill respectively,using different sample characteristics extraction methods and modeling methods of measurement,achieved good accuracy.According to the characteristics of the wet mill,the soft sensor of the support vector machine based on kernel selection is proposed.According to characteristics of wet mill,mill power,vibration and noise as the external input of mill,the material ball ratio,medium filling rate and grinding concentration as the output,respectively,said mill load,ball load and water load.Specific process is as follows: first,data collection,select the external characteristics of the mill,the sensor was used to collect the corresponding external characteristic information,then feature extraction,kernel function of the metric characteristics and energy entropy of related concepts,the use of ultra sphere to describe the realization of kernel function selection by means of;the is modeling and soft measurement,using access to prior information by support vector machine modeling to realize the soft measurement.According to the operating characteristics of dry mill,the soft sensing of mill load based on tensor machine is put forward.The first step,according to the dry mill running characteristics and existing knowledge collected correlative information of the external characteristics of the mill,such as mill current,grinding outlet temperature and inlet pressure and cylinder vibration and so on;the second step,according to the grinding mechanism and the sample data characteristics,of the mill load of grading,improve the efficiency of measurement.Finally after grading data batch use support tensor machine modeling,in order to achieve the dry grinding mill load measurement.Through the experiment,the different methods for different types of the mill to make full use of the sample prior information,improves the precision of mill load measurement software,it has strong generalization ability.The research content of this paper enriches the mill load measurement method for future mill automatic control provides the conditions.