Finished Gasoline Blending Formula Evaluation,Maintenance And Intelligent System Development

In the production process of finished gasoline tank blending,the merit of blending formulation determines the production efficiency of enterprise to a certain extent.However,at present,many domestic refineries still rely on linear formulation guided by manual experience,and pre-evaluation and modification of formulation are also done by manual.This is not only very demanding for production implementers,but also vulnerable to human factors.To ensure the quality of finished gasoline,only conservative blending formulas are used,resulting in a serious excess of product quality.Therefore,there is an urgent need for a more objective and intelligent " ante-hoc " evaluation and maintenance technology for blending formulation to improve production efficiency and intelligence of enterprises.In view of this,the thesis focuses on formulation evaluation model,establishment of the evaluation system and maintenance technology,etc.The following research work has been carried out:1)To solve the problem of intelligent evaluation and maintenance of finished gasoline blending formulation,a tank batch blending formulation evaluation and maintenance scheme is proposed by carefully sorting out the tank blending process mechanism and deeply analyzing the actual industrial production data.The scheme includes: formulation quality prediction modeling,formulation quality evaluation criterion establishment,defective formulation causation analysis for short-term maintenance,and quality formulation generation for long-term maintenance.Formulation quality prediction evaluation and causation analysis constitute "ante-hoc" evaluation and modification framework of blending formulation,combined with formulation quality evaluation criterion,to ensure that satisfactory formulation is used in actual production.Formula generation of long-term maintenance is to consider the tank batch finished gasoline blending process to provide a refined and efficient formulation generation model for product quality change and multi-working condition factor,so as to lay a foundation for the realization of intelligent evaluation and maintenance of formulation.2)To solve the problem that finished gasoline blending formulation needs "ante-hoc" evaluation and modification,a machine learning framework for quality prediction and causation analysis of blending formulation is proposed by combining the Light GBM model with SHAP.Secondly,improved genetic algorithm(IGA)is introduced for hyperparameter search of the Light GBM model to establish a model that can predict both performance and environmental indicators of finished gasoline,and formulation quality evaluation criterion is combined with gasoline National VIA standard and actual production of enterprises to realize the "ante-hoc" evaluation of formulations.Further,the SHAP-based global and local causation analysis of defective formulations provides easy-to-operate univariate qualitative correction recommendations.Through the experimental study using industrial data,the prediction model of finished gasoline quality index based on IGA＿Light GBM is more comprehensive and accurate.The modification suggestion given based on the causation analysis of the defective formulation with SHAP is reasonable and in line with the actual production,so short-term maintenance of formulation can be carried out accordingly.3)To solve the problem of accurate modelling of finished gasoline tank batch blending formulation,first,the Feature-Reduction Fuzzy C-Means clustering algorithm is combined with the Category Boosting model to propose a batch hierarchical integrated modeling scheme.Secondly,in view of unsatisfactory effect of FRFCM algorithm in division of nonspherical data,and the defects such as correlation of harmonic component is not considered,Mahalanobis distance is proposed to improve the FRFCM algorithm,in order to maximize the division of component batch types.Then,inspired by addition order and quantity of blending component,the correlation between blending component and oil quality is analyzed by grey correlation method,and blending component is classified into main and auxiliary ingredients categories,and the idea of sub-formulation grading modelling is proposed.Finally,by comparing the preferred Cat Boost as the batch sub-recipe grading modelling algorithm,optimizing its hyperparameters with a cuckoo search,and then fusing sub-grading model based on improved FRFCM affiliation matrix values,a refined modelling of tank batch blending recipe is achieved.The simulation results show that the proposed hierarchical integration model has better precision and generalization ability than single or un-hierarchical model,and can be used for long-term maintenance of intelligent formulation generation.4)To solve the problem of application of formulation evaluation and maintenance,first,function modules of system are determined by analyzing the requirements of intelligent system for finished gasoline blending formulation evaluation and maintenance.Secondly,considering practical application background and underlying algorithm,the technical scheme of using B/S as system architecture,My SQL as database and Django as front development layout of system is determined.Then,user login module,formulation quality evaluation module,formulation operation and maintenance function module,historical formulation view function module,alarm and event record module and system maintenance module of intelligent system are designed and developed.Finally,"black box" test method is used to test each function.The test results show that the system achieves expected functional goals and is easy to use and maintain,which is useful exploration for engineering application of blending technology.