Research On Intelligent Control System Of Vapor Recovery Based On Machine Learning

Gasoline and ethanol gasoline at gas stations are volatile liquids,which is easy to produce gasoline vapor in the refueling process,because of its high saturated vapor pressure and low viscosity.Existing vapor recovery systems recovery inefficiently,leading to serious economic losses and environmental pollution.To reduce the emission of the vapor during refueling,study on vapor recovery control system must be emphasized.After a deep understanding of refueling system and physical characteristics of volatile liquids,the study is carried out as follows.First,problems of existing vapor control system,such as adjustment to its vapor-liquid ratio and the control of its vapor recovery amount,are analyzed under the physical characteristics of volatile liquid.To enhance control precision of the vapor recovery,an experimental platform corresponding to the scheme in which vacuum pump is directly controlled by frequency inverter is set up.Results show that frequency control is completely feasible,and relationship between pumping volume of the system and converter frequency can be seen as liner.Second,volatile liquid flow through T-tube,variable diameter pipe and 90°-elbow during refueling are simulated in FLUENT.Simulations show that gasoline flow rate,environment temperature,fluid temperature and pipeline conditions have important influence on the emission of gasoline vapor.On this basis,optimization of vapor recovery control system by machine learning and frequency conversion technology is proposed.A vapor recovery control system in machine learning control mode is designed,and experimental platform of the system is established.After data acquisition by Lab VIEW and data preprocessing,a dataset for machine learning is constructed.Last,based on multi-layer perception and support vector regression machine,a prediction model for vapor-liquid ratio of the vapor recovery is established.The prediction model in which machine learning predicts vapor-liquid ratio,is verified to be completely feasible by the test dataset.Further optimization by binary particle swarm optimization and genetic algorithm,show that the optimized prediction model has better generalization,and GA-SVR model is better for vapor-liquid ratio of the recovery.Comparing the vapor recovery amount and vaporliquid ratio of three vapor recovery system in different control mode,results show that vapor recovery system in machine learning control mode can more efficiently match the supposed recovery amount,and realize real-time tracking in vapor-liquid ratio,which verifies the feasibility of the new vapor recovery control system.