| To respond to the construction of smart mines and realize the high-quality development of the coal mine industry,related research is carried out on the subsystem of smart mine —intelligent ventilation system.Intelligent mine ventilation means that the mine ventilation system has the accurate perception ability of the holographic ventilation data,makes use of the analysis,judgment,and decision-making ability of the ventilation brain software platform to analyze the status of the ventilation system in real-time,executes the commands of ventilation brain through the remote controlled ventilation equipment and facilities,realizes the precise linkage regulation of ventilation equipment and facilities,and forms an information closed loop by feeding back the data immediately.Normalization of system optimization,the precision of fault diagnosis,intelligent decision scheme,and automation of facility management and control are realized,and the entire process of intelligent ventilation,including knowledge accumulation,updating,and reusing,is completed.The optimal decision during the normal ventilation period is an important guarantee for an intensive and reliable ventilation system.A good perception system provides the basic for accurately decision-making.Therefore,perception and decision,as two inseparable core links in intelligent ventilation,are the primary problems.Ventilation perception includes data perception and abnormal perception.The perception of wind speed and air volume is important to ventilation data perception.This paper studies a real-time network algorithm independent of wind resistance.The accurate global air volume calculation is realized using some sensor data with errors.An adjustment algorithm is proposed to solve the turbulence,observation,and node air-volume-balance errors common in ventilation systems and obtain more accurate sensor monitoring data.Based on the accurate sensor data,the real-time network algorithm and the adaptive wind-velocity-sensor layout method are studied,the influence of different sensor layouts on the robustness of the algorithm is analyzed,and different solving methods are given to solve the problem that the traditional network calculation relies on wind resistance data and cannot update the wind velocity and wind volume data accurately and in real-time.Abnormal perception is an important means to maintain the stability and safety of the ventilation system.In this regard,the resistance and variable ventilation network fault diagnosis in ventilation network abnormal perception was studied,and a generalized large-scale ventilation network diagnosis architecture composed of random walk,prior topology weight feature aggregation,and posterior data weight feature aggregation was realized.It solves the problem of low accuracy of traditional diagnosis methods in large-scale ventilation networks and can accumulate,transfer and generalize fault diagnosis knowledge to new ventilation networks.Regarding ventilation decision-making,two kinds of decision algorithms are proposed:swarm evolution and reinforcement learning decision-making.The improved backboneparticle-swarm optimization algorithm was used to construct the dimensionless ventilation economy and safety goals.The constraint penalties were set for the ventilation power,air demand,and wind velocity to ensure the algorithm converged to a reasonable position.By initializing the optimized particle swarm in logarithmic space,the initializing sampled particle swarm has higher information entropy.The standard deviation contraction factor beneficial to population diversity is studied,and the elastic mirror mechanism is proposed to deal with the transgressive particles to avoid the waste of computing resources.This paper studies the perceptual decision algorithm and its process applied to the production environment,designs the parallel computing architecture of large-scale network decisions,and greatly improves the accuracy and efficiency of ventilation optimization decisions.However,swarm evolutionary decision-making still has shortcomings,such as the need for repeated calculation in each decision-making,difficulty in reusing the existing decision-making experience,unable to achieve the accumulation of decision-making knowledge,and large networks still needing to consume a certain amount of time.Therefore,multi-agent reinforcement learning decisionmaking is proposed.Reinforcement learning decision-making learns from the perceived observation of the ventilation system,the action of ventilation equipment,and the feedback of the ventilation system,and trains the value neural network used to evaluate the ventilation state and the action of ventilation equipment,as well as the policy neural network that can guide the action of ventilation equipment.An intelligent ventilation multi-agent reinforcement learning architecture based on "actor-critic" is constructed on this basis.The architecture pre-trains the decision-making model,realizes the reuse of decision-making experience,and can realize efficient and accurate ventilation decisions.The above two decision algorithms belong to the category of perceptual decision-making.Swarm evolutionary decision-making can complete the decision under imperfect data and experience collection,but each decision needs to be calculated,and the speed is slow.Reinforcement learning decision-making is a strategy model based on the training of historical experience and simulation experience.It is fast to calculate but needs a lot of decision-making experience.According to the current construction progress of intelligent ventilation systems,the swarm evolutionary decision algorithm can be applied in engineering and accumulate experience to reinforce learning decisions,gradually transition and achieve high-speed and accurate ventilation optimization decisions.Based on perception and decision algorithm,an intelligent ventilation system composed of the ventilation brain software system,holographic sensing system,precise regulation system,and intelligent ventilation knowledge base is developed.Among them,the ventilation brain software system is the central system,responsible for intelligent ventilation system perception,decision-making,and other workflow scheduling and resource allocation;the Holographic sensing system is responsible for collecting,analyzing,and calculating ventilation data.The precision control system is the execution system responsible for the precision control and feedback of ventilation equipment;the knowledge base is a special database for knowledge management.The intelligent ventilation system is constructed in stages in the production mine,and the intelligent ventilation perception and decision are preliminarily realized,which has certain significance for the high quality and intensive development of the mine ventilation field.The datasets are available on Gitee. |
