Research On The Formulation And Maintenance Of Finished Gasoline Tank Batch

In the online blending production of finished gasoline tanks,accurate and long-lasting high-quality blending formula generation is the basis for ensuring the quality of finished oil and close to the edge sticking production.Due to the batch phenomenon caused by the different origin and properties of the crude oil of the main blending ingredients and the residual oil remaining in the bottom of the tank,it is difficult for the formula generated by a single model to meet the production expectations.Therefore,driven by the improvement of the primary blending rate and the demand for trimming production in the batch blending process of finished gasoline tanks,this thesis conducts relatively systematic research on the modelling and maintenance of blending formulas.The main work is as follows:1)In order to solve the problems of excessive-quality and low primary blending rate in gasoline tank online blending,through in-depth analysis of the process mechanism and production data,based on the hybrid modelling framework of conservative and the edge fusion proposed by the previous team,a modelling scheme of two types of integrated models is proposed.The first is to establish a conservative batch formula model based on the parallel integration method to ensure a successful blending rate at one time;The second is to implant the knowledge distillation idea into the edge formula model based on the series integration method to approach the edge production;Then,a mixed formula model is formed through linear weighted fusion and online maintenance strategy to ensure that the primary blending rate is improved while approaching the production of edges.2)To solve the problem of accurate modelling of tank batch blending conservative formulations,firstly,because of the defects of the MKFCM algorithm that it is difficult to classify and difficult to select kernel parameters when the characteristics of the sample data are different,an adaptive kernel parameter calculation method is proposed,which is improved to improve the classification accuracy of batch data.Secondly,in order to ensure the modelling quality of the batch sub-models,the XGBoost with the best performance was selected to build the batch sub-models through the comparative analysis of the batch sub-models of 10 common modelling algorithms;Finally,the sub-models are fused based on the membership matrix value of the improved MKFCM algorithm.It has been verified by industrial data experiments that compared with a single model,the integrated model of multiple sub-batches in parallel can effectively improve the accuracy of the tank batch blending conservative formula model.3)To solve the problem of accurate modelling of the edge formula under small samples,considering the knowledge distillation algorithm in transfer learning,the advantage of improving the modelling ability of small samples can be improved through the existing convergent model learning.First,an edge formula modelling method based on the series integration of knowledge distillation and XGBoost(K-XGBoost)is proposed;Secondly,in view of the problem that the parameters in the K-XGBoost model are difficult to optimize,the particle swarm optimization algorithm(PSO)is used to optimize the relevant parameters,and then it is used for the predictive modelling of the batch blending edge formula of the finished gasoline tank.The simulation results show that,compared with the commonly used small sample PSO-LSSVM algorithm,both of them can be used to establish the edge formula model,and the performance of PSO-K-XGBoost is slightly better.4)Based on the above two formula models,conservative and edge,firstly,a linear weighted fusion strategy that is simple to calculate and easy to maintain online is adopted,and the two are fused to obtain a mixed formula model;At the same time,in order to deal with the uncertain factors in industrial production,two online maintenance strategies are further given.One is an adaptive algorithm for online weight correction,and the other is a periodic weight update algorithm based on PSO optimization.The experimental results show that the mixed formulation model based on the two maintenance strategies is not only beneficial to the high-efficiency production of tank batch blended gasoline,but also has stronger robustness to changes in working conditions.5)In order to apply the aforementioned formula model and maintenance strategy to engineering practice,firstly,the demand analysis of the formula generation and maintenance system was carried out,and the technologies and software platforms such as My SQL,Python,and C# were demonstrated and determined in combination with the needs of system research and development;Then,based on the aforementioned hybrid modeling scheme and online maintenance strategy,a gasoline blending formula management system with functional modules such as system login,formula prediction,formula data viewing,formula maintenance,formula data entry,alarm and event recording,and user management was designed and developed.The test results show that all the system functions have achieved the expected goals,and a proper technical exploration has been carried out for the following engineering application.