Research On Data Mining And Its Application In Petrochemical Plants

A large amount of historical data has been accumulated in petrochemical enterprises during daily production and management.These data contain a wealth of objective laws in industrial processes as well as abundant experience of operator.If the data could be mined in a proper way,it will be of great significance to make full use of vast amount of data to improve the production,promote profitability and reduce energy consumption.Data mining technology is researched in this paper to search useful information from historical data,and then will be applied to the modelling and operation optimization of typical petrochemical units.(1)Historical development and current status of data mining is overviewed.Afterwards,main tasks of data mining are also introduced.General process of data mining is summarized as well as popular data mining techniques.Especially,applications of data mining in modelling petrochemical units are review from extensive literatures.Lastly,the main contents and structure is simply presented.(2)Development of artificial neural network is reviewed in detail,and two representative neural network are introduced,namely back propagation neural network(BPNN)and radial basis function neural network(RBFNN).Wavelet neural network(WNN),which holds strongly nonlinear approximation capacity and captures local feature,is chosen to carry out data mining.Pointed to the slow process of general WNN,Levenberg-Marquardt algorithm and batch mode are introduced to the learning of WNN,which is supposed to fasten the training.(3)The WNN model of crude oil distillation is created by data generated in its rigorous mechanism simulation of Aspen Plus.Modelling results show that the accuracy and generalization of WNN model is better than those of BPNN and RBFNN.Line-up competition algorithm(LCA)is applied to search the optimal operation parameters that maximum the sales profit with the constrains of oil specifications.(4)Fundamental knowledge of rough set is introduced firstly.When the size of problem and data increase,computation grows rapidly in traditional reduction algorithms of rough set.The novel idea that applies LCA to the reduction of large amount of data is proposed to solve the NP-hard problem of reduction.Case study shows that the optimal attribute reduction can be found fast by LCA due to its global search capability.(5)LCA is employed to implement the attribute reduction of hydrocracking data which comes from real industrial production.Results show that the proposed heuristic method can reserve the key operation variable and reduce the redundant ones efficiently by reduction calculation of hydrocracking reaction data and 4 of 12 variables are reduced.On the basis of reduction,the processed data are optimized by visualization technology,and the optimal operating point is calculated.Optimization result shows that the yield of aviation kerosene is promoted by 3.6% compared with previous state of operation.