Research On The Control Method Of Safe Operation Of Coal Slime Flotation Process Based On Bayesian Network

Mineral processing is a key production link for obtaining mineral resources,and the utilization and recovery rate of mineral resources are affected by mineral processing performance.Flotation is one of the important mineral separation methods in mineral processing,which is widely used in coal,metallurgy,chemical industry,and other industries.Various abnormal operating conditions often occur in the coal slime flotation process,which may be mainly due to severe fluctuations in working conditions and frequent changes in raw materials.The occurrence of abnormal working conditions may not only affect product quality,but also threaten the personal safety of operators.In the actual production process,the most commonly used method after abnormal conditions occur is to manually eliminate them through operating experience,but this manual adjustment method has strong subjectivity.In order to solve the shortage of manual handling of abnormal conditions,there is an urgent need to design intelligent and safe control methods.After in-depth analysis of the coal slime flotation process,a series of studies have been conducted in this thesis:(1)Bayesian network is a knowledge representation tool that combines graph theory and probability theory,providing a powerful inference tool for industrial safety operation control.As the scale of industrial processes continues to expand and processes become increasingly complex,learning Bayesian network structures using expert knowledge becomes difficult and inefficient.Therefore,data driven methods are considered for learning Bayesian network structures.However,traditional data driven methods do not consider the weak causal relationship caused by time delays.Therefore,in order to solve the problem of inaccurate learning of Bayesian network structures caused by time delays,a method for secure operation control of Bayesian networks based on delay transfer entropy is proposed.Firstly,the likelihood part of the traditional Bayesian information criterion scoring function is replaced by transfer entropy.Then,delay parameters are added to the transfer entropy and dynamically determined through network scoring.Finally,the effectiveness of the Bayesian network security operation control method based on time-delay transfer entropy is verified by simulation experiments of coal slime flotation process.(2)Considering the complexity of modeling the coal slime flotation process,it is proposed to modularize the process and establish a distributed Bayesian network model from global to local.However,there are two problems in the distributed Bayesian network established after the modularization of long processes.One is that abnormal transmission between modules can occur in the global network,and the other is that the presence of media backflow and feedback in industrial processes can lead to cyclic in local networks.Therefore,in view of the problem of abnormal transmission between modules and the existence of local cyclic structures in local networks,a distributed dynamic Bayesian network security operation control method is proposed.This method establishes the global network in a distributed Bayesian network as a dynamic Bayesian network model,with modules connected between different time slots.During the establishment of a local Bayesian network,transfer entropy is used to identify the weakest causal relationship in the loop,placing the parent node in the weakest relationship in the previous time slice and the child node in the current time slice.Finally,the feasibility of the proposed method is verified through simulation experiments of coal slime flotation process.(3)Aiming at the problem that it is difficult to use data driven methods for structure learning of cyclic structures in coal slime flotation,a cyclic Bayesian network structure learning method based on sequence data is proposed.Firstly,a scoring search algorithm is used to learn the structure of online sequential data sets.Perform frequency statistics on the learned structures and set a threshold for the presence of directed edges.Secondly,the condition of two judgments is added to the directed edge selection.The first judgment is to determine whether there is an edge between the given variables,and the second judgment is to determine whether both directions exist based on the edge,thereby realizing the learning of the cyclic Bayesian network structure.Finally,the effectiveness of the proposed algorithm is verified through simulation experiments on the Asia network and coal slurry flotation process.The thesis includes 37 figures,46 tables and 82 references.