Study On Energy Consumption Abnormality Diagnosis In Energy System Of Papermaking Process Based On Data-driven Technique

Condition monitoring and abnormity diagnosis in energy system of papermaking process have important industrial value,and directly relate to the stable operation of the equipment,the quality of the products and the energy consumption level.The papermaking process has the characteristic of nonlinearity,multi-variable and strongly-coupling.It is difficult to establish the accuracy mathematical model.With the rapid development of the measurement,sensor and automation technology,the papermaking process has accumulated lots of operating data.How to mine out the potential rules and patterns from the operating data to improve the ability of monitoring and diagnosis is critical problem in both academia and industry.Based on the operating data obtained from Energy Management System(EMS)in papermaking process,condition monitoring and abnormity diagnosis in energy system of papermaking process are studied in a systemic way by applying the data mining theory,multivariate statistical analysis and intelligent optimization algorithm.The association rules algorithm is used to obtain the inherent laws between the energy consumption and the process variables.The genetic neural network is brought for energy consumption control chart pattern recognition.The running state and changing regularity of the energy system are analyzed from both global and local perspective respectively,achieving good results.This dissertation is organized as follows:(1)The energy consumption characteristics in paper industry are reviewed.The current research state in energy system of papermaking process is described.The influence of industrial big data on the development of the global industrial chain is analyzed.The existing fault diagnosis methods are emphatically introduced.(2)Based on the operating data obtained from EMS,the energy flows in different unit processes are quantified,and the influence of key operation parameters and environment parameters on energy consumption is discussed.The unreasonable links are found and the suggestions for improvement are proposed.(3)The association rules algorithm is used to obtain the inherent laws between the energy consumption and the process variables.The continuous data are first discretized and graded,and then the frequent item sets are represented by using the Apriori algorithm.The strong association rules are obtained by screening.Combining technics knowledge of papermaking process,the meaningful rules are stored to provide guidance.(4)The genetic neural network is brought for energy consumption control chart pattern recognition.In order to make the quality characteristics of the sample data the similar as actual production data,the sample data of the six basic patterns and the six mixed patterns are obtained through Monte Carlo method.BP neural network is used for control chart pattern recognition,and the average recognition rate is 91.75%.Because the BP neural network is easy to fall into local optimum,a genetic optimization algorithm that can search for the optimum parameters of BP neural network is introduced.The genetic neural network is used for control chart pattern recognition,and the recognition rate can reach 96.08%.Combined with the practical production data,the genetic neural network is used for energy consumption control chart pattern recognition,and the changing regularity of the energy consumption is analyzed from the global perspective,providing the basis for abnormity diagnosis.(5)Because the papermaking process has the characteristic of nonlinearity,multi-variable and strongly-coupling,this study introduces the kernel method,and establishes the multivariable abnormity diagnosis model based on the kernel principal component analysis(KPCA)and the kernel slow feature analysis(KSFA).The four different abnormal patterns are designed to verify the validity of the model.By comparing the study results,the performance of KPCA is much better than that of PCA and KSFA.KPCA possesses better ability to detect the abnormalities and identify abnormal process variables in time.(6)Based on the platform of Visual Studio 2010 and using the object-oriented programming language C++,the software is developed for online monitoring and abnormity diagnosis.The developed software integrates the matrix computing environment,which can realize complicated operation.The online application indicates that the software is effective.The contributions of this dissertation are as follows:(1)an association rule method based on the 3? criterion is proposed,which can discretize and classify the continuous data.By applying the proposed method,the strong association rules between the energy consumption and the process variables are obtained from the large number of data,which are meaningful to provide guidance for abnormity diagnosis.(2)The energy consumption control chart pattern recognition model is established.The established model takes the raw data in control chart as feature set,and takes the genetic neural network as classification tool.It can provide a strong theoretical support and judging basis of energy consumption abnormity diagnosis.(3)The condition monitoring and abnormity diagnosis method in energy system of papermaking process based on KPCA and KSFA is proposed.This method can extract nonlinear features from the sample set,reduce the correlation between the variables,and achieve a good diagnosis performance.