Energy efficiency evaluation based on DEA integrated factor analysis in ethylene production☆

Shixin Gong,Cheng Shao,Li Zhu*

Institute of Advanced Control Technology,Dalian University of Technology,Dalian 116024,China

1.Introduction

To keep competitive and achieve sustainability,ethylene industry,which constitutes a large portion of the petrochemical industry,is required to improve its energy efficiency[1].Therefore,the increment of energy efficiency for ethylene production has attracted growing attention worldwide,especially in China.The 2014 national bureau of statistics reports that the average comprehensive energy consumption of ethylene in China is as high as 816.6 kgoe·t-1(kilograms of oil equivalent per ton),1.63 times higher than the average level of Middle-East region,the international advanced level[2].In 2015,the comprehensive energy consumption per ton of ethylene has only a 0.1%drop compared with that in 2014.

The above figures suggest that ethylene production industry in China demands energy conservation.Energy conservation implies improvement of energy efficiency and energy efficiency evaluation is the precondition of implementing energy conservation[3],namely,energy saving is realized scientifically through the comprehensive evaluation of energy efficiency level in production process.It is obvious that scientific and reasonable energy efficiency evaluation of ethylene production process will be of great economic and environmental benefits for the development of petrochemical industry.

As a favored method for energy efficiency evaluation,data envelopment analysis(DEA)is extensively used to measure energy efficiency of a production unit[4].DEA algorithm can be used in the energy efficiency evaluation in ethylene production due to the relative efficiency defined by DEA.A method for analyzing the energy efficiency in ethylene production at the whole plant level is proposed based on DEA[5].But the traditional DEA shows poor resolution and many optimal values of DEA efficiency facing up to the complex operation and huge data in the large-scale chemical production process directly.Therefore,the researches aimed at the improvement of DEA have been carried out a lot.

To overcome the influence caused by the dimension of input and output data and the inappropriate indicators,DEA integrated PCA is presented[6].The DEA model combined with the fuzzy mathematics is presented to distinguish the performance of decision making units(DMUs)and the improvement of relative ineffective DMUs is provided[7].DEA integrated Interpretation Structure Model is proposed to find the dominant factors influencing the energy consumption of ethylene production process and surmount the problem of evaluation difficult caused by plenty of DMUs[8].Energy efficiency of ethylene production

DEA is a method based on the concept of relative efficiency and uses the convex analysis and linear programming as the tool to estimate effectiveness of DMUs according to the input and output data[10].The main advantage of DEA is that it allows for efficiency evaluation of multiple inputs and outputs without assigning weights and specifying any function form.

Two DEAmodels named CCR(Charnes,Cooper and Rhodes)and BCC(Banker,Charnes and Cooper)are used to evaluate overall technical efifciency and pure technical efficiency respectively[11].The relationship of them is that the overall technical efficiency equals the product of pure technical efficiency and scale efficiency.

There arensupposed DMUs and each DMU has the sameminputs andsoutputs.The input and output vectors can be expressed as follows.virtual benchmarking method based on dependent function analytic hierarchy process modelis proposed according to technologies,scales and data distribution of energy consumption[9].

Above all the literatures,the improvement of DEA model itself is the focus but the complex productive technology of ethylene production is ignored.The operational condition of ethylene production process is complex.Energy consumption shows obvious differences in the different conditions and the differences are huge.It is not accurate of energy efficiency assessment to lose sight of the complex ethylene productive technology and the improvement strategies of energy consumptions based on energy efficiency evaluation cannot meet the requirements of different working conditions and achieve optimizing the development of petrochemical industry.

Therefore,the energy efficiency evaluation is based on the operation classification in this paper.DEA algorithm is applied respectively into the different working conditions of ethylene production and improvement strategies of energy consumption based on the multi-working conditions are provided.The data screening procedure is applied before DEA for the low resolution of DEA efficiency problem caused by excessive input indicators.This evaluation strategy with respect to the operation classification realizes an in-depth,accurate and comprehensive analysis of energy efficiency in ethylene production process.

There's one point needed to be attention to is that the values calculated by DEA are based on the concept of relative efficiency,which has different physical explanations according to the selection of input and output data.The relative efficiency means comprehensive energy consumption for unit output of product when the inputs are energy resources and the outputs are the product yield while the relative efficiency is explained as“energy”utilization efficiency when the input and output parameters selected are all energy resources.The former,that is comprehensive energy consumption for unit output of product,is the definition of energy efficiency in this paper.

The remainder of the paper is organized as follows.Section 2 reviews the DEA model,factor analysis andk-mean clustering algorithm respectively.Section 3 expounds the central idea of the proposed method.Section 4 gives a case study about energy efficiency evaluation based on the multi-working conditions in an actual ethylene production.Finally,Section 5 gives the concluding remarks.

2.Methodology

2.1.Data envelopment analysis

wherexmjis them-th input of thej-th DMU,ysjis thes-th output of thej-th DMU.

So for thej-th evaluated DMU,the efficiency value obtained by the CCR-DEA model can be expressed as the fractional programming in Model(1)[12].

where u=(u1,u2,…,us)T,v=(v1,v2,…,vm)Tare the weight coefficients of theminputs andsoutputs.

For convenience of calculation and following evaluating work,the Model(1)can be transformed as an equivalentproblem linearprogramming model and non-Archimedean infinitesimal ε is introduced.So the CCR-DEA with non-Archimedean infinitesimal ε is shown as Model(2).

whereθ,λjare the dualvariables;e-,e+are themandndimension unit vectors;andS+,S-are the slack variables.

And the BCC-DEA with non-Archimedean infinitesimal ε is shown as Model(3)[12].

whereθ,λjare the dualvariables;e-,e+are themandndimension unit vectors;andS+,S-are the slack variables.

CCR model is based on the constant returns to scale(CRS)for the efficiency while BCC model is variable returns to scale(VRS)[12].In the practical application of ethylene production energy efficiency evaluation,the annual ethylene yield and production scale of the domestic ethylene production enterprise that is an example of energy efficiency evaluation are fixed.In addition,this paper studies the comprehensive energy efficiency evaluation of ethylene production process and the overall technical efficiency,namely,the absolute DEA effective should be considered.Therefore,the concept of CCR efficiency that is in accordance with the actual existing state of the stable operation of ethylene production is more appropriate than BCC model in this paper.

CCR model shows that the inputX0should be tried to ensure to increase or decrease in the same proportion when the outputY0of thej0-th DMU remains constant.So the judgments of the CCR dual model with non-Archimedean in finitesimal ε can be obtained[12]:if θ0＜ 1,the evaluated DMUs are relatively ineffective;if θ0=1,the evaluated DMUs are at least the weak relatively effective.

2.2.Factor analysis

In the actual energy efficiency evaluation,input and output indicators are so many and there is a complex relationship among the indicators.The performance of DEA model depends on the input and output data[13].If the number of input data can be reduced by some means,the performance of DEA can be improved.Factor analysis is such a data screening method to solve this problem.

Factor analysis was presented in 1900s by K.Pearson and C.Spearman.Factor analysis refers to a kind of statistical technique which extracts the common factors from the variable group.The characteristics of this method are:the number of factors is less than the number of original factors and the analysis of the factor variable can reduce the workload of the calculation;the common factors canreflect the most in formation of original factors;the actual situation of all the evaluation units can be evaluated comprehensively and objectively[14].

Supposing that there arepinput and output indicators,Xp,and the mean value is μ and the covariance matrix is σ:

whereF1,F2,… ,Fmare common factors extracted fromXp;ε1,ε2,… ,εmare specific factors,which are used to account for the partofXp;A=(aij)is the factor loading matrix.

According to the above introduction,the modeloffactor analysis can be established[15].And the common factors can be found after factor extract,factor rotation and factor scores.

2.3.K-means clustering algorithm

Clustering algorithm was proposed by Mchalski in 1980s to realize the classification of samples or indicators.K-means clustering algorithm,which was presented in 1967 by J.B.Macqueen,is an influencing and effective algorithm for scientific and industrial applications to solve problem of data clustering.Thek-means clustering algorithm is an iterative partitioning algorithm that realizes the property of compaction in the class and the property of independency among the class,by determining a set ofkpoints,which are considered as the centers of the resulting clusters[16].K-means clustering divides the data samples into different categories through the iteration process,making criterion function of clustering performance to achieve the optimal result.The processes ofk-means clustering algorithm can refer to the review in detail[17].

Compared to other clustering algorithm,k-means algorithm has been proven to have the performance of efficient and good stability,spectral clustering effect and fast hierarchical clustering to deal with big data,especially the industrial continuous data[18].This algorithm is mainly used to realize the division of working conditions of ethylene production according to the corresponding data in this paper.The ethylene production data relating to the working condition are complex and continuous.In order to obtain a better working condition classification result,k-means algorithm,which is mature and effective,is applied to recognize working conditions.And the validity of the division of working conditions based on thek-means algorithm is verified in the following section.

3.DEA Integrated Factor Analysis with Respect to Operation Classification

3.1.Energy boundary of ethylene production

In the ethylene production process,raw materials are processed through the high temperature cracking furnaces,cooling unit,compressing and separation unit and ethylene,propylene and other byproducts are produced.The main energy types used include the water,power,steams,fuels,N2and compressing air.According to the statistical analysis of the energy consumption of ethylene production process[19],the proportion of consumption of fuel,steam,electricity,water and other gases is 72.6%,10.9%,4.7%,14.6%and 1.1%metering from the view of energy medium types.

In order to guarantee the comprehensive and objective analysis for energy efficiency of ethylene production process,all the energy consumption data need to be considered,including fuel gas;steams;electricity;water including the recycle water,the industrial water and the desalted water;nitrogen and compressed air.In addition,the energy mediums generated,such as medium pressure steam,low pressure steam and waste water,etc.should be also considered,namely,the energy produced from the input energy mediums need to be removed.

3.2.Data preprocessing

The precision of results of the DEA depends on the input and output indicators[13].The ethylene data has complex nonlinearfeature and includes noise and abnormal data,etc.Thus,consistency check,uniform dimension and normalized disposal are needed[20].The data could be processed according to the Grubbs rule shown as Eq.(5):ifT＞T(n,α),thexishould be eliminated.

whereandSdenote the mean and variance of the data;xiis thei-th data;the value ofT(n,α)is obtained by the table of Grubbs rule,nis the number;α is the significance level.

In addition,different energy mediums have different energy qualities and levels,which are not comparable.So the consumptions of energy mediums cannot be used directly as the inputs of the energy efficiency evaluation.The general method to express the level is to convert measurement units of energy consumption parameters of fuel,electricity,water,steam,etc.into uniform kgoe·t-1based on the General Principles for Calculation of the Comprehensive Energy Consumption(GB/T 2589-2008)[21].

3.3.DEA integrated factor analysis with respect to operation classification

Ethylene production is a complicated process with different kinds of parameters and there exists tight coupling among these parameters.Energy efficiency has a strong relationship with operational parameters and is also affected by the parameters such as the load rate,ethylene yield,productive technology,operation time,raw materials properties,etc.[22].So energy distribution medium compositions and conversion efficiency are different with different parameters.In other words,a certain type of working condition has its inherent amount of energy efficiency,which is incomparable in different working conditions.So it is necessary to adoptclassified working condition and proceed to evaluate energy efficiency afterwards considering productive technology.

However,research on efficiency evaluation by DEA considering the actual productive operation and technology is very lacking.The previous researches pay more attention to the improvement of the DEA algorithm[7,8].Even though the same property of DMUs is met,which are selected only by the time interval,the features of productive technology are weakened.If energy efficiency evaluation lacks productive technology analysis,the current energy efficiency level and production situation cannot be explained fundamentally and make the further energy efficiency increase in the ethylene production process difficult.If the data constituting DMUs are divided into several subsets in a certain feature considering the productive technology,then we can respectively analyze for each subset,the results are analyzed independently or synthetically again.So DMUs can be confirmed combining with parameters affecting energy efficiency.

Besides,the number of the input and output indicators of DMUs may influence the performance of DEA[13].The energy mediums needed in the ethylene production are various.The improper or excessive input and output indicators would lead to the poor resolution of DEA efficiency.Thus screening the appropriate input and output indicators is particularly important.

Therefore,efficiency evaluation in ethylene production with respect to operation classification based on DEA integrated factor analysis is proposed.Operation classification refers to the different working conditions here,which is confirmed according to the actual operation and productive technology.The purpose of our study is to evaluate the energy efficiency more scientifically and provide the more effective improving strategies of energy saving.So the typical working conditions are conifrmed by the operational parameters and the cross DEA with non-Archimedean in finitesimal ε is used as an example analysis model to analyze the energy efficiency of ethylene process according to the input and output data in different working conditions.Furthermore,due to the complexity and coupling of ethylene production data,factor analysis is used to screen the appropriate input and output indicators.The improving direction of the ineffective DMUs for different working conditions can be obtained by the slack variables of the DEA model.

The energy efficiency evaluation procedure for ethylene production based on DEA integrated factor analysis with respect to operation classification is described as follows.

Step 1:Select energy consumption data and carry out data

preprocess.

Step 2:Analyze the factors of energy efficiency and confirm the

number of the typical working conditions considering the produc

tive technology.

Step 3:Recognize the working conditions of data byk-means

clustering algorithm.

Step 4:Obtain the common factors from the input indicators by

factor analysis as the inputs of DEA.

Step 5:Analyze the data of the same working conditions by DEA.

Step 6:Estimate the rationality of typical working conditions according

to the result of DEA.

Step 7:If the division of working conditions is reasonable,give the

improving scheme of energy efficiency to different working condi

tions;or return to Step 2.

Daily stable production data of nine months in 2014 in a domestic ethylene production unit are used as the analysis object in this paper.The data include the ones used to evaluate energy efficiency and the ones used to determine the working conditions.First,according to the above energy boundary of ethylene production process and the definition of energy efficiency determined before,taking effects of different feeds on fuel,steam,water and electricity consumption into consideration,the sum of consumption is used as the model input while ethylene yield is the model output.And these data used to evaluate energy efficiency need to be preprocessed for data rejection,abnormal data removal and data screening.Second,after the working conditions are determined by the corresponding data stated in the next section,the different working condition data are analyzed by the DEA model respectively,making analysis results more accurate and objective.

So the energy efficiency evaluation with respect to operation classification is realized by DEA model integrated factor analysis.The specific flow of energy efficiency evaluation in ethylene production with respect to operation classification is shown in Fig.1.

4.Energy efficiency Evaluation

4.1.Division of typical working conditions

Working condition means the normal operational condition of industrial production process[23],which is influenced by productive technological parameters.Furthermore,different working conditions correspond to different energy efficiency levels.Thus,the division of working condition and the confirmation of working condition of ethylene production data are the preconditions of energy efficiency evaluation with respect to operation classification.

Fig.1.Flow diagram of energy efficiency evaluation.

Ethylene production itself is complicated and accompanied by the energy transfer and material transfer in the process with many operating procedure and operating parameters.In general,the main factors influencing energy efficiency of ethylene production process include raw materials,production scale and technology,operating parameters,ethylene yield,load rate,energy consumption,waste energy loss,etc.[22].The energy consumption and ethylene yield are the input and output of DEA model.Production scale and technology are settled when the boundary of energy efficiency evaluation is fixed.

Raw materials are the fundamental of ethylene production and the main factor affecting the product yield[24].In addition,raw material cost accounts for more than half of the operation cost in the ethylene production[25].Raw materials include lightweight and heavyweight material.Lightweight material is an ideal cracking raw material,but the price is expensive.Generally,different raw materials are mixture to be cracked in the actual ethylene production.Therefore,different raw material compositions have different impact on the energy efficiency.

Raw material composition mainly have three kinds of circumstances according to the actual production of one domestic ethylene plant,including

1.Raw materials are HTO,AGO,NAP,LH and HC5;

2.Raw materials are HTO,AGO,NAP and HC5;

3.Raw materials are HTO,AGO and NAP.

The operating parameters are various and the parameters related to cracking process should be considered first.Reaction temperature,dilution ratio,residence time and cracking depth are the most related parameters.Reaction temperature,dilution ratio and residence time are constant generally,so cracking depth is considered in this paper.

According to the actuality of ethylene production,the cracking depth value is real-time changing and reflects the degree of reaction.Here the ratio of propylene and ethylene yield denotes cracking depth,which can reflect the product structure.And the cracking depth of cracking furnace in the actuality is divided into low and high cracking depth,0.4 and 0.5 respectively.

The energy efficiency also has a relationship with load rate[26].According to the requirement of the actual production,the load rate is divided into 0.7,0.8 and 0.9.

Therefore,raw materials composition,operating parameters and load rate are considered in this paper as the basis for classification of the type of working condition.The typical working conditions are shown in the Table 1.

Table 1The typical working conditions

According to the typical working conditions,k-means clustering algorithm is used to classify the working condition of data.The process ofk-means clustering algorithm is as follows:

1)Determine the initial centers according to the 8 typical working conditions in the Table 1;

2)Assign each data used to evaluate energy efficiency to the most similar classes again according to the average of every object in the dataset;

3)Update the average of class by calculating the average of the new object class;

4)Repeatstep 2),3)until the criterion function of clustering meets the

precision requirement.

The historical energy consumption data of ethylene production for nine months in 2014 in a row are classified into eight typical working conditions byk-means clustering algorithm.The classification results are shown in Table 2.The validity of the division of the working conditions is verified in the following part.

Table 2Results of working condition division by k-means

4.2.Factor analysis for input and output indicators

The types of energy medium needed in the ethylene production process are various.If the DEA model is applied directly,the precision and speed of the algorithm will be influenced[27].In addition,some energy mediums impact on improvement of energy efficiency little due to the little consumption.Therefore,the input and output indicators need to be filtrated and the most representative parameters are used to consist of the sample space.So in this paper,factoranalys is is used to screen the representative inputindicatorsto prepare for the energy efficiency evaluation based on DEA.

According to the actual ethylene production process,energy mediums mainly include fuel gas;steams including the high pressure steam,the middle pressure steam,the low pressure steam and the dilution steam;water including the recycle water,the industrial water,the boiled water,the desalted water and the domestic water;nitrogen,compressed gases,decoking air and electricity,etc.15 kinds of energy mediums are selected to analyze by factor analysis to determine the dominant factors,where the principal component analysis mode is used to handle factors.Finally, five factors are extracted as common factors due to the reasons that eigenvalues of the five variables are greater than 1 and the cumulative contribution rate is 88.249%.Therefore,the input indicators of DEA model are confirmed.Factor 1 mainly reflects the information of MS,DS and purified compressed gas.Factor 2 mainly reflects the information of LS,the desalted water and decoking air.Factor 3 mainly reflects the information of HS,the recycle water and the domestic water.Factor 4 mainly reflects the information of fuel gas,the industrial water and nitrogen.Factor 5 mainly reflects the information of the boiled water,unpurified compressed gas and electricity.

The common factors obtained after factor analysis may be negative,which will influence the subsequent evaluation.Thus the common factors need to be standardized so that the energy data are realized to be screened to the appropriate indicators for energy efficiency evaluation based on DEA.

4.3.Energy efficiency evaluation

The method proposed in this paper is applied into the energy efficiency evaluation and analysis of one domestic ethylene production unit.According to the above determined energy boundary and the analysis result of factor analysis, five inputindicators and one outputindicator have been confirmed.The result of the efficiency of production obtained by DEA model and DEA integrated factor analysis with respect to operation classification is shown in Fig.2.The red line is the result of FA-DEA with respect to operation classification proposed in this paper,the blue line is the result of the previous DEA and the green line is the result of DEA with respect to operation classification.

Fig.2.The efficiency values of a specific ethylene production.

As shown in Fig.2,the trends of efficiency values obtained by the three methods are the same.However,because of too much input indicators,the results of DEA and DEA with respect to operation classification are unsatisfactory.The pros and cons of part of DMUs cannot be identified,which lead to high efficiency of some DMUs and the low validity of the analysis results.For example,the DMUs from 37th to 41th,from the 240th to 244th in Fig.2 belong to the 2nd working condition.The Comprehensive Energy Consumption of Ethylene(CECE)of these sample points are 1048.99 kgoe·t-1,992.78 kgoe·t-1,1034.90 kgoe·t-1,1052.47 kgoe·t-1,880.52 kgoe·t-1,837.62 kgoe·t-1,1001.93 kgoe·t-1,1127.14 kgoe·t-1,965.33 kgoe·t-1and 755.78 kgoe·t-1respectively.The relative efficiency values of these DMUs are all 1 obtained by DEA with respect to operation classification while the results of DEA are 0.770,0.789,0.904,1,1,0.754,0.882,1,1,1 respectively,which does not conform to the actuality.And the efficiency values of FA-DEA with respect to operation classification proposed in this paper are 0.900,0.920,0.923,0.924,0.936,0.910,0.90,0.890,0.903 and 0.92,which conforms to the trend of the CECE.The same phenomenon also appears in the other working conditions.

In order to explain the validity of the two methods and verify the rationality of the confirmed working conditions,the results of the FA-DEA with respect to operation classification are compared to the values of DEA model and the CECE.

According to CCR-DEA model,the method is used to evaluate the same working condition of DMUs based on the relative efficiency.And the input indicators of DMUs are the consumption of energy mediums and outputindicator is the ethylene yield in this article.So the efficiency value obtained by DEA is the reciprocal of the comprehensive energy consumption per ton of ethylene.So if the confirmed working conditions are effective,the curve should be inverse trend with the comprehensive energy consumption per ton of ethylene,namely,when the comprehensive energy consumption per ton of ethylene declines,the curve of the relative efficiency increases.

The compared results are shown in Fig.3.We can see that the curves of two methods are roughly inverse trends with CECE.In other words,the efficiency values obtained by DEA and FA-DEA with respect to operation classification are both valid.But more specifically,the results of the proposed method are guaranteed while the values are more reasonable,such as the DMUs from85th to 173th.The fluctuating ranges of the efficiency values of these DMUs obtained by DEAare higher,which does not conform to the actual changes.Therefore,the proposed method is more reasonable in terms of the division of working condition and the evaluation results.

Fig.3.Verified model for energy efficiency.

Atthe same time,the precision ofenergy efficiency values calculated by the above two methods can be appraised by Root Mean Square Error(RMSE)and Mean Absolute Error(MAE).The formulas of the RMSE and MAE are defined as Eqs.(6)and(7):

wherenis the number of data,xiis thei-th efficiency value,xis the comprehensive energy consumption per ton of ethylene.The comparison analysis results are shown in Table 3.

The result of DEA with respect to operation classification is also compared to the other two methods.From Table 3,it can be seen that energy efficiency evaluation with respect to operation classification,namely,inallusion to different working conditions,canreflect the energy efficiency level of the ethylene production more accurately.And the performance of DEA integrated factor analysis is better than that of DEA and DEA with respect to operation classification.

Table 3The comparison analysis results of two methods

Because of the evaluation without respect to the working conditions,the property of the DMUs in term of raw material composition,load rate and cracking depth will be different and the reliability of result of DEA is influenced to a great degree.Therefore,the accuracy of evaluation is relatively low and the actual energy efficiency level of ethylene production cannot be reflected scientifically and reasonably.By contrast,the evaluation based on DEA with respect to operation classification is improved.Data belonging to the same working condition have consistency.The impact on the calculation results of DEA model caused by material composition,load rate and cracking depth is eliminated to a great extent,which makes the energy efficiency more close to the actuality.In addition,the hybrid evaluation method integrated factor analysis reduce effectively the poor performance of DEA caused by overmuch input indicators.

The energy efficiency is influenced by many factors and the potential of energy efficiency in the different working conditions is different.So the opportunity and strategy of improvement on energy consumption should be different according to different working conditions.The improving strategy of ineffective DMUs in allusion to different working conditions,namely,the improving direction of energy consumption,is given according to the input slack variables of the CCR-DEA model.

In order to verify the effectiveness of the direction of the input indicators obtained by the method presented in this paper,the comprehensive energy consumption per ton of ineffective DMUs is updated according to the input and output slack variables.The results are shown in the Fig.4.The blue line is the updated energy efficiency based on the method proposed in this paper,the red line is the result based on DEA and the green line is the original energy efficiency value,CECE.So we can see that the CECE is improved more obviously by the improving advices obtained by the FA-DEA with respect to operation classification while the value is declined unobvious by DEA.

The energy efficiency values improved based on the method proposed in this article varies roughly from 500 kgoe·t-1to 700 kgoe·t-1in most situations while the energy efficiency values are higher than 700 kgoe·t-1at some points.After analysis,these points belong to the second and third working condition.The energy efficiency range of the second working condition is from 705.95 kgoe·t-1to 755.31 kgoe·t-1and the range of the third working condition is from 630.65 kgoe·t-1to 701.23 kgoe·t-1.Similarly,other working conditions can obtain the energy efficiency range.The conclusion is shown in the Fig.5.

Fig.4.The improving advices of energy.

Fig.5.Energy efficiency of different working conditions.

According to the division of working conditions,it can be seen that the best energy efficiency of different working conditions is different,especially different raw material composition,the difference is large.So the high energy efficiency doesn't have to be blindly pursued without regarding to the constraints of working condition.

5.Conclusions

With respect to the complex ethylene production conditions and the production data,the traditional DEA model shows poor resolution and many optimal values of DEA efficiency.This paper studies a way to evaluate the energy efficiency in the ethylene production process more objectively and comprehensively,which is based on DEA model considering the operation classification.

For the complex operational condition of ethylene production process,the typical working conditions are confirmed firstly according to the ethylene productive technology parameters.Secondly,on the basis of the working conditions,DEA model is applied to evaluate the energy efficiency of different working conditions and the opportunity and direction for input energy indicators of the ineffective DMUs are provided based on the multi-working conditions.

By applying this method to an actual energy efficiency evaluation in a domestic ethylene production unit,it is proved that the typical working conditions presented in this paper are reasonable and scientific,which conform to the actual production.And the evaluation method integrated data screening can over the poor resolution of traditional DEA.The more targeted improvement directions of energy efficiency are more satisfactory for different working conditions and the effectiveness is verified.In addition,the impacts on raw material composition,load rate and cracking depth on energy efficiency of different working conditions are analyzed.The idea of energy efficiency evaluation with respect to operation classification can be applied into other chemical production process.

[1]Siyu Yang,Qingchun Yang,Qian Yu,A composite efficiency metrics for evaluation of resource and energy utilization,Energy61(2013)455-462.

[2]National Bureau of Statistics,Annual Research and Consultation Report of Panorama Survey and Investment Strategy on China Industry,Zero Power Intelligence Group,Shenzhen,2015.

[3]Chenxi Song,Mingjia Li,et al.,Research on energy efficiency evaluation based on indicators for industry sectors in China,Appl.Energy134(2014)550-562.

[4]Zhanxin Ma,Data Envelopment Analysis Model and Application,Science Press,Beijing(in Chinese),2013.

[5]Cui Peichong,Xuelei Wang,Zengliang Ma,An energy efficiency analysis system for ethylene production based on DEA,Comput.Appl.Chem.27(9)(2010)1183-1186.

[6]Qunxiong Zhu,Chen Xi,Yanlin He,Lin Xiaoyong Gu Xiangbai,Energy efficiency analysis for ethylene plant based on PCA-DEA,CIESC J.01(2015)278-283(in Chinese).

[7]Han Yongming,Geng Zhiqiang,Qunxiong Zhu,Qu Yixin,Energy efficiency analysis method based on fuzzy DEA cross-model for ethylene production systems in chemical industry,Energy83(2015)685-695.

[8]Han Yongming,Zhiqiang Geng,Gu Xiangbai,Qunxiong Zhu,Energy efficiency analysis based on DEA integrated ISM:A case study for Chinese ethylene industries,Eng.Appl.Artif.Intell.45(2015)80-89.

[9]Geng Zhiqiang,Zhu Qunxiong,Gu Xiangbai,Dependent function analytic hierarchy process model for energy efficiency virtual benchmark and its applications in ethylene equipment,CIESC J.62(8)(2011)2372-2377(in Chinese).

[10]Chen Liang,Guozhu Jia,Environmental efficiency analysis of China's regional industry:A data envelopment analysis(DEA)based approach,J.Clean.Prod.1(45)(2016)1-8.

[11]Chenxi Song,Mingjia Li,et al.,A data envelopment analysis for energy efficiency of coal- fired power,Energy Convers.Manag.102(2015)121-130.

[12]Zhanxin Ma,Data Envelopment Analysis Model and Method,Science Press,Beijing,2010(in Chinese).

[13]Han Yongming,Geng Zhiqiang,Liu Qiyu,Energy efficiency evaluation based on data envelopment analysis integrated analytic hierarchy process in ethylene production,Chin.J.Chem.Eng.22(2014)1279-1284.

[14]Zhao Zhonggai,Liu Fei,Factor analysis and its application to process monitoring,CIESC J.58(4)(2007)970-974.

[15]Kim Jae-On,W.Mueller Charles,Factor Analysis:Statistical Methods and Practical Issues,Truth and Wisdom Press,Shanghai,2012(in Chinese).

[16]Qunxiong Zhu,Shi Xiaoyun,Gu Xiangbai,Tian Qing,Time-series data fusion and its application to energy efficiency value for ethylene plants,CIESC J.61(10)(2010)2620-2626(in Chinese).

[17]Anil K.Jain,Data clustering:50 years beyond K-means,Pattern Recogn.Lett.31(8)(2010)651-666.

[18]Rehab Duwairi,Mohammed Abu-Rahmeh,A novel approach for initializing the sphericalK-means clustering algorithm,Simul.Model.Pract.Theory54(2015)49-63.

[19]Yuehua Xu,Chemical Plant Energy Saving Technology and the Case Analysis,China Petrochemical Press,Beijing,2009(in Chinese).

[20]Geng Zhiqiang,Han Yongming,Zhang Yuanyuan,Shi Xiaoyun,Data fusion-based extraction method of energy consumption index for the ethylene industry,Lect.Notes Comput.63(29)(2010)84-92.

[21]General principles for calculation of the comprehensive energy consumption GB/T 2589-2008,Committee of National Standardization Management of China,2008(in Chinese).

[22]Han Yongming,Energy efficiency Evaluation Method and Application in Ethylene Industry,PhD Thesis,Beijing University of Chemical Technology,2014(in Chinese).

[23]Qunfeng Niu,Li Wang,Working condition recognition of screw compressor using wavelets theory,Proceedings of the 7th World Congress on Intelligent Control and Automation.Chongqing,2008.

[24]XueleiWang,The dynamic benchmarking method for energy consumption of ethylene production process,Comput.Appl.Chem.27(2010)1165-1170.

[25]Mian Yang,Dalia Pati?o-Echeverri,Fuxia Yang,Eric Williams,Industrial energy efficiency in China:Achievements,challenges and opportunities,Energ.Strat.Rev.6(2015)20-29.

[26]Yin Yang,Ethylene Energy efficiency Management Expert System,PhD Thesis,Beijing University of Chemical Technology,2015(in Chinese).

[27]Yu Chenping,Gu Xiangbai,Geng Zhiqiang,Ethylene industry efficiency analysis based on EPI,CIESC J.09(2012)2931-2935(in Chinese).