Study On Data Mining And Prediction Method Of Energy Consumption Used In Energy Management System Of Paper Process

Paper industry is one of the nine high-energy-consumption industries in China, andnowadays paper mills are facing enormous pressure to save energy under the severe situationsof environment, policy and market. Energy saving methods can be generally classified intothe following three categories:①saving energy with structural adjustment;②saving energywith technical renovation;③saving energy with management reform. And managementmethods not only provide important support to structural and technical methods, but alsocontinuously improve energy efficiency.EMS (Energy Management System) takes a key rolein the management energy-saving methods. EMS is a new management-control integrationtechnology on energy-saving which combines automation and information technology, itapplies dynamic monitoring on energy conversion, utilization and recycling within the mills,improves and optimizes energy balance and realizes the mill-wide energy-saving.Guangzhou Paper mill (Haizhu district) was taken as the object to carry out our research,GE Intelligent Platform was taken as the software platform, and then the core modules ofMEOP (Mill Energy Optimization Platform)——Data Integration and Real-Time Calculationof Energy Information was developed to achieve the centralized and meticulous managementof energy system in the studied paper mill. Then the specific interaction techniques in EMSwere employed to realize the integration of MEOP and functional models based onpapermaking process. The established MEOP realized the accurate, transparent and real-timeaccess to the mill-wide energy information. Taking the2009PM1specific energyconsumption indicator for an example, the RMSE (Root Mean Square Error) between MEOPand manual statistic is0.47GJ/t, MPE (Mean Percentage Error) is6.79%, and in addition,MEOP realized the real-time calculation of this indicator with interval of1minute, whosetimeliness was much higher than manual statistic of1month. Apart from the specific energyconsumption and such other energy efficiency indicators, MEOP also supported to access theenergy flow data based on material flow, this rich real-time energy information provided adeep understand to the process energy consumption and energy flow. Based on dataintegration and energy information extraction, the integration of MEOP and papermaking process functional models was achieved by data interaction techniques in EMS, and thenenergy monitoring, energy analysis and such other basic EMS functions were fulfilled,moreover rich process data and energy information provided fundamental data for furtherenergy factors analysis and mining, and energy consumption prediction.There are a mass amount of process data in papermaking process, but comprehensiveunderstanding of these data is beyond of human’s capability, and operators are always trappedin the trouble of “rich data but poor information”. Based on the EMS, the research took themassive process data as foundation and energy consumption prediction as final goal, thenadopted data mining techniques to analyze and mine the energy influencing factors inpapermaking process. According to the process data characteristics, such as multivariable,multi-correlation among variables and massive data, PCA (Principal Component Analysis),PLS (Partial Least Square) and such multivariate transformation and selection methods wereintroduced and combined to analyze and mine the importance of energy influencing factors inpapermaking process. As a result, the important energy influencing factors were obtained andthe importance of energy influencing factors was ranked. Then according to the variablesimportance analysis result, important energy influencing factors were selected, and energyinfluencing factors with no obvious correlation to energy consumption were removed. Theresearch shows that the combination of proper data preprocessing and PLS can realize theimportance analysis of multi-correlation energy influencing factors and provide importantsupport to energy-saving technical retrofits, abnormal energy consumption analysis andenergy consumption prediction, etc.Based on the PLS analysis results of energy influencing factors, and according to theadvantages and characteristics of PLS and ANN, a novel PLS-ANN prediction model wasdeveloped to predict process energy consumption. For improving the accuracy, thepapermaking process energy consumption was classified based on the energy consumptioncharacteristics of each process procedure, and according to each energy consumption divisiontype, appropriate prediction model——PLS-ANN or ANN, was taken to realize the energyconsumption prediction of each process procedure. Based on the energy consumptionprediction of each process procedure, the effective1hour energy consumption prediction ofthe entire papermaking process was achieved. The results show that for the power consumption of papermaking process, the prediction RMSE is9.02kWh/t and the correlationcoefficient between predicted data and measured data is0.95, and for the steam consumptionof papermaking process, the prediction RMSE is0.03t/t and the correlation coefficientbetween predicted data and measured data is0.88, it shows the proposed prediction methodhas good precision. The effective energy consumption prediction of papermaking process canprovide important support to optimize the operation in energy conversion section and reduceenergy waste, then benefit to realize the dynamic balance of process energy system.This thesis focuses on the study of methods and applications, and its features andinnovation points are:1) The real-time energy information calculation based on material flow and materialthermodynamic properties is introduced to the EMS of paper process, and accurate,transparent and real-time energy information are obtained;2) According to the characteristics of papermaking process data, such as multivariable,multi-correlation among variables and massive data, PCA, PLS and suchmultivariate transformation and selection methods are employed to analyze andmine the energy influencing factors in paper process. Appropriate data preprocessingmethods and PLS are combined to realize the transformation and importanceanalysis of energy influencing factors, and to identify the primary energy influencingfactors;3) The division and classification of energy system of paper process is introduced, andaccording to each energy consumption division type, appropriate prediction model isestablished to improve the energy consumption prediction accuracy. And aPLS-ANN prediction model is developed which combines the PLS advantages indata analysis and interpretation and the ANN advantages in nonlinear modeling,self-learning and adaptability.