Quantitative credibility evaluation of Global Energy Interconnection data

1 Introduction

With the construction of the Global Energy Interconnection (GEI) Digital Research Platform,the data collection and storage capabilities of the Global Energy Interconnection Development and Cooperation Organization (GEIDCO) have been greatly improved [1].A comprehensive data center has been established,wherein the following six major systems are included:energy,electricity,economy,politics,environment,and finance.Though such a resource of abundant data provides substantial support for the development of various research,it brings new challenges [2],[3].In particular,methods of accurately describing and scientifically evaluating the credibility of the massive amounts of GEI data need to be developed.

Domestic and foreign scholars have conducted extensive research on data credibility evaluation.Currently,such data credibility analysis and evaluation methods are mainly divided into two categories:subjective trust analysis and objective trust analysis.Subjective trust analysis refers to the cognitive phenomenon in which the subject of trust subjectively judges the specific behavior or characteristics of the object of trust.Different subjective trust evaluation methods have been proposed based on probability theory and fuzzy set theory [4-10].Objective trust analysis is primarily based on evidence theory:evidence is used to define the trust relationship between different items,and description,verification,and reasoning are then conducted.E.g.,D-S evidence theory for credibility calculation was used to evaluate data credibility [11-17].Reference [18] proposed a trust propagation model based on a Bayesian conditional probability formula and the trust propagation model; however,this model has a strong dependence on the trust propagation model.Reference [19] proposed a trust worthiness calculation method based on user contexts in social network services by referring to the process of trust generation between people in sociology and psychology.Reference [20] proposed a credible metric model based on a dynamic Bayesian network.Therein,the following interactions were highlighted:if the comprehensive credibility remains unchanged,it declines according to a certain time rule; if the entity is operated illegally,it must punish the comprehensive credibility; if the context is constantly updated,it needs to update the comprehensive credibility of the target entity.The disadvantage of this is that the model only evaluates the credibility between data sources,without measuring the credibility of the data overall.Reference [21] proposed a group trust algorithm based on a social network; two interactive groups are abstracted as group nodes based on their characteristics,and,to simplify calculation,the multi-to-multi trust relationship is transformed into a one-to-one relationship.

In general,subjective trust analysis is based on the relative independence of the subject from the characteristics and behavior of the subject,with ambiguity,uncertainty,and inability to accurately describe,verify,and reason.Though objective trust analysis is more reasonable than subjective trust,it does not consider the impact of timeliness and malicious recommendations,and lacks flexibility.There has yet to be a suitable model for evaluating the credibility of GEI data,as these data have a wide range of sources,complex classification,and a large volume.

To address this lack of a credibility evaluation method,this paper proposes a GEI data credibility quantitative evaluation method based on the Cloud Model.The evaluation indicator system of this method evaluates the data credibility from multiple perspectives to ensure the comprehensiveness and impartiality of the evaluation results.Additionally,the combination of ambiguity and randomness is achieved via the Cloud Model.The hard division of comments is also abandoned,making up for the subjective randomness defects of the traditional methods and ensuring the objectivity and accuracy of the evaluation results.

2 Cloud Model

In nature,the main influencing factors of the conversion relationship between qualitative concepts and quantitative values are randomness and ambiguity.Randomness is a form of contingency,i.e.the uncertainty reflected by each event in a set of probabilistic events.Ambiguity refers to the uncertainty in judgment caused by the unclear division of things.To achieve the uncertainty conversion between qualitative concepts and quantitative values,Li De-yi (member of the China Engineering Academy) proposed the Cloud Model based on traditional fuzzy mathematics and probability statistics [22].In the process of conversion,the Cloud Model takes into account both the ambiguity of qualitative concepts and the randomness of quantitative data; thus,the essential characteristics of the event are maintained.The Cloud Model has been successfully applied to natural language processing,data mining,decision analysis,intelligent control,image processing,etc.

2.1 Basic concepts

·Definition

Suppose U={x} is a domain represented by a numerical value.The domain can be one-dimensional,two-dimensional,or multi-dimensional; T is a qualitative concept which is described by language,and the concept is associated with U.For any x in domain U,if there is a membership random number u(x) which has a stable tendency for the qualitative concept T,then the distribution of x on the domain U is called a cloud.Each (x,u(x)) is called a cloud drop.The Cloud Model has the following characteristics:

1) For any x in domain U,the mapping of x to the membership interval [0,1] is a one-to-many relationship.The membership of x for the qualitative concept T is a probability distribution,not an exact value.

2) Each cloud drop is a mapping of the membership relationship between the value of the domain U and the qualitative concept T.The overall shape of a cloud composed of a large number of cloud drops reflects the basic characteristics of the qualitative concept.

3) The greater the probability of a cloud drop appearing,the greater the probability that a quantitative value belongs to a qualitative concept.

·Digital features

The Cloud Model contains three digital features:expected value Ex,entropy En,and hyperentropy He.These features can reflect the quantitative characteristics of the qualitative concept as a whole.Their meaning is as follows:

Ex is the point that best represents this qualitative concept in the domain.It is the most typical sample point after this concept is quantified.

En represents a measurable granularity of a qualitative concept; usually,the larger the entropy,the more macroscopic the concept.Entropy also reflects the uncertainty of the qualitative concept,indicating the range of values that can be accepted by the qualitative concept in the space of the domain,i.e.,the degree of ambiguity.

He is a measure of the uncertainty of entropy.It reflects the randomness of the sample representing the qualitative concept value.It reveals the association between ambiguity and randomness.

2.2 Cloud generator

The cloud generator can be divided into two types:the forward cloud generator and the reverse cloud generator (Fig.1).The forward cloud generator generates several cloud drops (x,u(x)) from the digital features (Ex,En,He) of a cloud.The reverse cloud generator obtains three digital features of the cloud through several cloud drops.

Fig.1 Two types of the cloud generator:(a) forward cloud generator; (b) reverse cloud generator

·The algorithm of the forward cloud generator is as follows:

Input:Three digital features (Ex,En,He) of the cloud and the number N of cloud drops to be generated.

Output:N cloud drops (x,u(x)).

1) Take En as the expectation and He as the standard deviation to produce a normal random number Enʹ;

2) Take Ex as the expectation and Enʹ as the standard deviation to produce a normal random number x;

3) Obtain the result (x,u(x)),a cloud drop (refer to (1));

4) Repeat steps 1-3 until N cloud drops are produced.

·The algorithm of the reverse cloud generator is as follows:

Input:N cloud drops x[N] = {x1,x2, …,xN}.

Output:Three digital features (Ex,En,He) of the cloud.

1) Calculate the average via (2);

2) Calculate the first order sample absolute center L via (3);

3) Calculate the sample variance S2 via (4);

4) Obtain Ex via (5),En via (6),and He via (7).

3 Data credibility quantitative evaluation model

The GEI data credibility quantitative evaluation model is a multi-level comprehensive evaluation model.First,the evaluation indicator system of data credibility is established and the data evaluated from different perspectives; second,the weights of different evaluation indicators are determined; third,single-indicator evaluation clouds are generated according to the evaluation results of different evaluation indicators; finally,a comprehensive evaluation cloud is generated for comprehensive evaluation.The whole process is shown as follows (Fig.2).

Fig.2 Process of data credibility quantitative evaluation model

3.1 Evaluation indicator system

When evaluating data credibility,different indicators are used to evaluate the data from multiple perspectives.

·Integrity evaluation

Integrity refers to whether data information is missing or not.Missing data involves two cases:the entire data record is missing,or the record of a field in the data is missing.As the value of data which are incomplete can be greatly reduced,data integrity serves as a basic evaluation indicator.

The integrity of data is relatively easy to evaluate and it can be evaluated by the recorded values and statistics results.

·Consistency evaluation

Consistency refers to whether the values of the same attribute of an entity are consistent across different systems or data sets.Usually,data has standard coding rules.Consistency evaluations are relatively simple:they involve considerations of the data’s satisfaction of standard encoding rules.

·Accuracy evaluation

Accuracy refers to whether the data is abnormal or incorrect.Unlike consistency,data with accuracy issues does not only involve inconsistencies in rules.Data accuracy errors,such as garbled data,are more common.Moreover,an unusually large or small amount of data is also considered inaccurate.

Generally,data is normally distributed.If there are some errors in data with lesser proportions,the method of abnormal value detection can be used to evaluate the accuracy of data.

·Validity evaluation

Validity refers to both format validity and numerical validity.Before performing format validity analysis,all valid formats of the data must be predetermined.Then,the record is compared with the valid format,one by one.If the expression format of a record matches a valid format,the record can be considered valid; otherwise,the record is considered unrecognizable.E.g.,the line loss rate includes three valid formats,namely a decimal,“%” and “/”.Numerical validity is usually analyzed to judge whether the size of a record is within a certain accepted range of values.

·Timeliness evaluation

Timeliness considers the “freshness” of the data.Timesensitive data can reflect the latest features in a timely manner,with a high availability and reference value.The timeliness of data can be judged according to the degree of data delay.

3.2 Weights of evaluation indicators

The analytic hierarchy process (AHP),based on mathematics and psychology,is a structured technique for organizing and analyzing complex decisions.It is used to categorize the weights of different evaluation indicators.

Rather than prescribing a “correct” decision,the AHP helps decision makers identify a decision that best suits their goal and their understanding of the problem.The AHP provides a comprehensive and rational framework for structuring a decision problem,representing and quantifying its elements,relating those elements to the overall goals,and evaluating alternative solutions.

The AHP first decomposes the decision problem into a hierarchy of more easily comprehended sub-problems,each of which can be analyzed independently.Once the hierarchy is built,the decision makers systematically evaluate its various elements by comparing them with each other,two at a time,with respect to their impact on an element above them in the hierarchy.The AHP converts these evaluations to numerical values that can be processed and compared over the entire range of the problem.A numerical weight or priority is derived for each element of the hierarchy,allowing diverse and often incommensurable elements to be compared to one another in a rational and consistent manner.In the final step of the process,numerical priorities are calculated for each of the decision alternatives according to the importance scale outlined in Table 1.These numbers represent the alternatives’relative ability to achieve the decision goal,allowing a straightforward consideration of the various courses of action.The AHP avoids directly assigning weights to various factors,instead comparing the importance of different factors between them,with higher accuracy and objectivity.

Table1 Importance scale for assigning numerical priorities

Importance scale Meaning 1 The former factor is as important as the latter factor.3 The former factor is slightly more important than the latter factor.5 The former factor is obviously more important than the latter factor.7 The former factor is strongly more important than the latter factor.9 The former factor is extremely more important than the latter factor.2,4,6,8 Intermediate value of the above adjacent judgments.reciprocal If the importance scale of element i to element j is aij,then the importance scale of element j to element i is aji=1/aij.

The algorithm for determining multi- index weights based on the AHP is as follows:

Inputs:A comparison matrix of five evaluation indicators.

Outputs:Weight values of five evaluation indicators [w1,w2,w3,w4,w5].

1) The feature vector method,least square method,approximate calculation method,or another method is used to solve the weight vector and the maximum eigenvalue of the comparison matrix.

2) The consistency is tested,and the weight vector’s conformation to the logic law checked.The judgment indicator of the consistency test is:

where RI is the average random consistency indicator,which can be obtained via the RI table (Table 2); CI is the consistency indicator,which is calculated as follows:

where λmax is the maximum eigenvalue of the comparison matrix.

3) If CR ＜ 0.1,the comparison matrix is considered to have a strong consistency,and the weight vector is valid.The value of the weight vector is the weight value of the indicator.If CR≮0.1,the comparison matrix needs to be appropriately modified.

Steps 1-3 are repeated until CR ＜ 0.1.

Table2 Random consistency indicator (RI) of comparison matrix with different orders

n 1 2 3 4 5 6 7 8 9 RI 0 0 0.58 0.9 1.12 1.24 1.32 1.41 1.45

3.3 Comprehensive evaluation cloud

Let A (Ex,En,He) be a Cloud Model and k be a constant,then the product of A and k has the following relationship.

Let A1(Ex1,En1,He1) and A2(Ex2,En2,He2) represent two different clouds,respectively.Their relationship is represented as follows.

where A is the comprehensive cloud of A1 and A2.Let the evaluation cloud corresponding to the evaluation indicator Ai be Ai(Exi,Eni,Hei),and the indicator weight be wi,where i=1,2,...,N (N is the number of indicators in the evaluation indicator system).The comprehensive evaluation cloud A(Ex,En,He) composed of all single-indicator evaluation clouds can then be obtained via (10) and (11).

4 Case analysis

4.1 Test data

We selected the electricity consumption data in the comprehensive database of GEIDCO for analysis.The data contains 11 categories of consumption data,including steel industry electricity consumption,petrochemical industry electricity consumption,non-ferrous metals industry electricity consumption,machinery industry electricity consumption,energy industry electricity consumption,transportation equipment electricity consumption,highway transportation electricity consumption,railway transportation electricity consumption,etc.The statistical years of the data range from 1990 to present; the data sources were the IEA and the UN.Some of the data are as follows (Tables 3 and 4).

Table3 Steel industry electricity consumption of IEA (Unit:GWh)

2014 2013 2012…1990 Hong Kong0.00 0.00 0.00…554.00 Mainland China and Hong Kong 579560.00 570423.00 522052.00 … 56124.00 Mainland China 579560.00 570423.00 522052.00 … 55570.00 Chile 398.00 464.00 549.00…377.00………………Albania 188.00 178.00 153.00… 0.00

Table4 Steel industry electricity consumption of UN (Unit:GWh)

2014 2013 2012…1990 China 626439.30 617392.00 565894.00 … 0.00 Chile 398.00 464.00 549.00…377.00 Jordan 0.00 0.00 0.00…0.00 United Kingdom 3786.00 3804.00 3376.00 … 9071.00………………Albania 188.25 178.16 153.03…0.00

4.2 Calculation of the weights of indicators

The relative importance of indicators is obtained by expert scoring.Based on the results of this expert scoring,the comparison matrix A is constructed as shown below (Tables 5-8),where each of the five evaluation indicators (integrity,consistency,accuracy,validity,timeliness) are represented in turn.

The weight vector is as follows:

The maximum eigenvalue of matrix A is λmax=5.137.The value of the consistency indicator CI is as follows:

The value of the consistency judgment indicator CR is as follows:

Since CR=0.0306 ＜ 0.1,the comparison matrix satisfies the consistency requirement,i.e.the weight vector obtained via (18) is valid.In summary,the weights of integrity,consistency,accuracy,validity,and timeliness are 0.2586,0.0546,0.5049,0.0546,and 0.1274,respectively.

4.3 Data credibility evaluation

Data credibility evaluation mainly includes singleindicator evaluation and comprehensive evaluation.

·Single-indicator evaluation results

According to the evaluation indicator system,the 11 categories of electricity consumption data from the IEA and UN are evaluated and analyzed.The results are shown in Tables 5-8.

Table5 Single-indicator evaluation results of IEA electricity consumption data

Evaluation index steel petrochemical nonferrous…line loss Integrity 43.97 48.82 36.33 … 62.23 Consistency 99.32 99.32 99.31 … 99.56 Accuracy 97.23 95.4 95.82 … 97.75 Validity 100 100 100…100 Timeliness 89.29 89.29 89.29 … 89.29

Table6 Single-indicator evaluation results of UN electricity consumption data

Evaluation index steel petrochemical non-ferrous … line loss Integrity 42.3 46.22 26.62 … 90.32 Consistency 100 100 100 … 100 Accuracy 98.08 95.55 95.75 … 97.63 Validity 100 100 100 … 100 Timeliness 89.29 89.29 89.29 … 89.29

According to Tables 5 and 6,the inverse cloud generator can be used to generate the single-indicator evaluation clouds.The digital features of the clouds are shown in Tables 7 and 8.

Table7 Digital features of the evaluation clouds for IEA

Evaluation index Exi Eni Hei Integrity 47.6264 18.7902 8.8563 Consistency 99.38 0.1162 0.0386 Accuracy 96.4627 1.4889 0.1288 Validity 100 0 0 Timeliness 89.29 0 0

Table8 Digital features of the evaluation clouds for UN

Evaluation index Exi Eni Hei Integrity 46.8955 25.3269 9.3332 Consistency 100 0 0 Accuracy 96.70818 1.2968 0.4401 Validity 100 0 0 Timeliness 89.29 0 0

·Comprehensive evaluation results

According to Tables 7 and 8,the digital features of the comprehensive evaluation clouds for IEA and UN data can be obtained.The results are shown in Table 9 and Fig.3.

Table9 Digital features of comprehensive evaluation clouds of different data sources

Data source Ex En He IEA 83.27 4.76 1.82 UN 83.24 5.89 2.14

Fig.3 Comprehensive evaluation clouds of IEA and UN

4.4 Results analysis

·Single-indicator evaluation results analysis

It can be seen from Tables 7 and 8 that,in terms of integrity,both the IEA data and the UN data have room for improvement.Additionally,the integrity of the different categories of electricity consumption data varies greatly.In terms of consistency,the result of UN data is better.The IEA data includes a small amount of inconsistent data.In terms of accuracy,different categories of electricity consumption data from the IEA and UN are all highly accurate.In terms of the validity,there is no problem with either the IEA or UN data.In terms of timeliness,both the IEA and the UN data have room for improvement; this is because neither the electricity consumption data from the IEA nor the UN contain the records of the most recent three years in the comprehensive database of GEIDCO.

·Comprehensive analysis

As can be seen from Table 9,the IEA data has a slightly higher expected value than that of the UN data,indicating that the score of the IEA’s electricity consumption data is relatively better.At the same time,the entropy and superentropy of the IEA data are lower,indicating that the stability of the IEA’s electricity consumption data is better,and that the data credibility deviation between different categories is smaller.

From the above analysis,it can be seen that the main factor affecting the credibility of IEA and UN electricity consumption data is lack of integrity.At the same time,the integrity between different categories varies greatly.In the next step of data collection,it is necessary to increase the collection intensity for categories with serious data missing,and obtain relevant data from multiple channels to improve the overall data credibility.Additionally,all categories of the IEA and the UN electricity consumption data in our database lack data from the last three years; thus,in the subsequent database construction process,the frequency of data updating should be increased and data credibility improved by improving the timeliness of the data.

5 Conclusion

We proposed a GEI data credibility evaluation model,namely,the quantitative evaluation method based on the Cloud Model.First,integrity,consistency,accuracy,validity,and timeliness were selected to construct the evaluation indicator system.On the basis of this,the data set was processed with the same measurement to solve the dimensional differences of different fields.Additionally,AHP was used to determine the weight of evaluation indexes according to the relative importance of indexes.Second,the quality evaluation model of GEI data based on the cloud model was processed; this model combines fuzziness and randomness,abandons the hard division of comments,makes up for the defects of subjective randomness of traditional methods,and ensures the objectivity and accuracy of evaluation results.Finally,through a case analysis of the power consumption data in the comprehensive data center of GEIDCO,the validity of the evaluation model was verified,and suggestions for improving data center construction and data quality are given.

Acknowledgements

This work was supported by the State Grid Science and Technology Project (No.52450018000H).