Research On Abnormal Condition Identification For Electro-fused Magnesium Furnace Based On Bayesian Network Parameters Transfer Learning

The fused magnesia furnace is one of the main equipment for producing fused magnesia.The heat generated by the arc between the electrode and the charge causes the charge to melt,thereby obtaining a higher purity fused magnesia crystal.The abnormal working condition identification of the fused magnesia furnace at the smelting site mainly relies on the operator with long-term operation experience to observe the fluctuation of the current value of the threephase electrode and the operating state of the on-site melting equipment.The smelting process of fused magnesium furnace is a complex non-linear,multi-uncertainty and multi-variable coupling process.Due to limited experience or operational lag,field operators often cannot adjust the control variables in time and accurately,which ultimately leads to product poor quality,wasted resources,difficult to meet production requirements and even pose a great threat to the safety of production equipment and operators.Existing scholars use Bayesian networks to identify abnormal conditions in the smelting process of fused magnesium furnaces.The learning of Bayesian networks relies on sufficient data.However the melting process of single furnaces does not produce enough abnormal data.Effective learning cannot achieve good learning results.Aiming at this problem,this thesis uses the traditional learning method to extract the smelting information of other furnaces or other time to learn Bayesian network parameters,thereby improving the network learning effect,so as to guide the actual production and improve production,efficiency and the goal of maximizing economic efficiency.The main contents of this thesis are as follows:(1)For the fused magnesia Bayesian network model of multi-source information fusion,when the data in the target is insufficient,the established model cannot accurately and effectively analyze the target problem.The Bayesian network parameter migration learning method based on the target domain expert knowledge is proposed to be consistent with the target structure,which improves the accuracy of abnormal state recognition.(2)Most existing factories still only use the current information to judge the abnormal working conditions.Therefore,in order to use these data in Bayesian network parameter migration learning,a target domain expert knowledge based parameter migration method is proposed when the Bayesian network structures are inconsistent.The Bayesian network parameter migration learning method combined with structural knowledge improves the accuracy of abnormal state recognition.(3)The current information,the image information and the sound information are input as evidence to the Bayesian network after the migration learning,and the level of the abnormal working condition of the fused magnesium furnace is inferred.The level of anomalies is determined based on the inferred results to assist the operator in making decisions.Finally,the validity of migration learning for Bayesian network reasoning is verified by test data.