1 Introduction
Handling uncertainty is heavily studied in many real applications such as approximate reasoning and decision making. The first step to handle uncertain information is reasonable description or modelling of the uncertain information. One of the most used theory is Dempster Shafer evidence theory. Since first proposed by DempsterDempster1967 and then developed by ShaferShafer1976 , has been paid much attentions for a long time and continually attracted growing interestsyang1994evidential ; yager1996aggregation ; denoeux2004evclus ; elouedi2004assessing ; dempster2006dempster ; yang2006evidential ; yang2006application ; jousselme2006measuring ; cuzzolin2007two ; cuzzolin2008geometric ; klir2008remarks ; dempster2008generalization ; utkin2009computing ; dezert2012hierarchical ; denoeux2012evidential ; kang2012evidential ; Wei20132564 ; yager2012dempster ; masson2011ensemble ; zhang2012assessment ; deng2011risk ; deng2011modeling ; deng2011new ; nguyen2012belief ; Huang2013 ; liu2013evidential ; denoeux2013maximum ; liu2014credal .
One open issue of evidence theory is the conflict management when evidence highly conflicts with each other. A famous example is illustrated by Zadehzadeh1986simple . Since then, hundred of methods are proposed to address this issue murphy2000combining ; lefevre2002belief ; yong2004combining ; liu2006analyzing ; haenni2005shedding ; SebastDesterckeToward2013 ; schubert2011conflict ; yang2013discounted ; roquel2012new ; lefevre2013preserve ; sarabi2013information ; yang2013evidential .
The normalization step is questionable in Dempster combination rule and is deleted to assign the conflict to the whole setsyager1987dempster , under the closedworld assumption that the frame of discernment is exhaustive. The transferable belief model (TBM) is presented to represent quantified beliefs based on belief functionssmets1994transferable ; smets2005decision
. TBM was constructed by two levels, namely the credal level where beliefs are entertained and quantified by belief functions and the pignistic level where beliefs can be used to make decisions and are quantified by probability functions. Dubois and Prade proposed a new rule of combination which is better adapted and more specific than Yager’s rule of combination concerning the assignment of the conflicting mass
dubois1988representation . Lefevre et.al reviewed the previous works and made a unified model to handle conflict evidence. The main idea of their method is to address that the conflict will be assigned to which hypothesis and how to assign the conflictlefevre2002belief . However, to assign the conflict to different hypothesis is questioned by Haenni haenni2002alternatives . A model X with method Y has an intuitive result Z. Haenni argued to modify the method Y is not reasonable since that the intuitive result may be caused by model X. Some methods are coincide with Haenni’s opinion. For example, Murphy averages the conflict evidence and then combines the averaged evidence itself several timesmurphy2000combining . A weighted average method is proposed by Dengyong2004combining with a better convergence ability. In the weighted average method, the evidence distance function presented in jousselme2001new , instead of the commonly used conflict coefficient in evidence theory, is used to model the conflict degree. Liu pointed out that the commonly used conflict coefficient in evidence theory is not reasonable to represent the conflict degree between two pieces of evidenceliu2006analyzing . To address this issue, a two dimension conflict measure, combined with the classical conflict coefficient and pignistic betting distance, is proposed. In addition, the adoptive is discussed. Three cases are proposed according to the two dimension conflict measure.Generally speaking, there are two main reasons to cause the evidence conflicts with each other. One is the questionable sensor reliability caused by any disturbance or the maintain condition of the equipment. The other is that the system is in open world where our knowledge is not completed. For example, target recognition is widely used in military application. We know that the enemy has three types of jet fighter, namely . As a result, one of the three types will be recognized, given by some collected evidence from different radars. However, we do not know that a new type of jet fighter, namely is developed secretly. When the target is flying near the radars, one radar may sent the report that the target is and the one radar may sent the report that the target is , which is very similar to Zadeh’s countintuitive example.
The rest of this paper is organized as follows. Section 2 the background knowledge are brief introduced, including DempsterShafer theory, pignistic probability transform and Jousselme’s evidence distance proposed. In Section 3, the proposed generalized evidence theory is presented, mainly including GET, generalized evidence distance, generalized combination rule (GCR) and its application. The discuss of , generalized conflict model and its application are presented in Section 4. Finally, conclusions are given in Section 5.
2 Preliminaries
2.1 DempsterShafer theory
For completeness of the explanation, a few basic concepts about DempsterShafer theory are introduced as follows.
For a finite nonempty set , is called a frame of discernment (FOD) when satisfying
(1) 
Let be the set of all subsets of , namely
(2) 
is called the power set of . For a FOD , a mass function is a mapping from to , formally defined by:
(3) 
which satisfies the following condition:
(4)  
(5) 
In DempsterShafer theory, a mass function is also called a basic probability assignment (BPA). Given a BPA, the belief function is defined as
(6) 
The plausibility function is defined as
(7) 
where . These functions and express the lower bound and upper bound in which subset has been supported, respectively.
Given two independent BPAs and , Dempster’s rule of combination, denoted by , is used to combine them and it is defined as follows
(8) 
with
(9) 
Note that the Dempster’s rule of combination is only applicable to such two BPAs which satisfy the condition .
2.2 Pignistic Probability Transform
In the transferable belief model(TBM)smets2005decision , pignistic probabilities are used for decision making.
Let be a BPA on the frame of discernment . Its associated pignistic probability function is defined as
(10) 
Where is the cardinality of subset
. The process of pignistic probability transform (PPT) is that basic probability assignment transferred to probability distribution. Therefore, the pignistic betting distance
Smets1994191 can be easily obtained by PPT. Let and be two BPAs on the frame of discernment , the pignistic probability functions are and , respectively. The pignistic betting distance (, for short) between and are given as follows.(11) 
indicates that the support degree by BPAs.
2.3 Jousselme’s evidence distance
Jousselme et al.jousselme2001new proposed a new distance of measure the conflicting of two bodies of evidence. That is evidence distance.
Let and be two BPAs defined on the same frame of discernment , containing mutually exclusive and exhaustive hypotheses. Let represents the distance of two bodies of evidence.
(12) 
Where and are respective two BPAs. is an matrix whose elements with from and , respectively.
2.4 Liu’s evidence conflict model
In traditional DempsterShafer evidence, the conflict coefficient represents the degree of conflicting between two bodies of evidence. Liu pointed out that the traditional conflict coefficient can not efficiently measure the disagreement between two bodies of evidenceliu2006analyzing . The conflict model proposed by Liu in liu2006analyzing indicates that the pignistic betting distance and coefficient are united to represent the degree of conflict.
Let two BPAs and are on the same frame of discernment , the conflict model proposed by Liu is shown as follow.
(13) 
Where is the classic conflict coefficient of Dempster combination rule, and the is the pignistic betting distance in Eq.(11). When both and , and are regard as conflict, where is the threshold of conflict tolerance. Because there does not exist an ”absolute meaningful threshold” of conflict tolerance satisfying all pairs of BPAsayoun2001data , the value of is subjective and not fixed in different applications. Generally speaking, the value of more closer to 1, the greater the conflict tolerance is.
3 The Generalized Theory
DempsterShafer theory has many merits in information fusion, which is a big improvement, it is relative to the Bayesian theory. However, there are still some drawbacks in DempsterShafer theory, including but not only as follows. Computational complexity is heavy with the increasing elements in the set, which limits the application in the area of high realtime. In addition, while the evidence are high conflicting, the counterintuitive results will present. In the frame of discernment, the follow two may be the main reasons that lead to highly conflicting. One is the incomplete of the frame of discernment. For example, for military applications, if there are three targets , and on the frame of discernment. Then the sensors can only recognize the different unions of this three targets. However, if there existing a new unknown target , the sensors can not distinguish whether it is one of the previous three targets. In this situation, the recognition results will be multifarious, after combination, the incorrect results will present. Another is the reliability of sensors itself. Work condition, disturbance, etc. will influence the judgment result. There are some alternatives to overcome these shortcomings. Preprocessing the information or approximation algorithmvoorbraak1989computationally ; haenni2002resource are used to solve the computational complexity. Researchers have paid great effort to solve the high conflicting problem in recent years. The two typical solution is transferable belief model (TBM)smets1994transferable and DezertSmarandache theory (DSmT)dezert2006dsmt . The characteristic of TBM model is that the interesting concepts of close world and open world. However, to our best knowledge, the application of TBM is still under the close world. DSmT provides a new idea of solution the high conflicting problem, but the computation complexity is heavier, and when under the condition of low conflicting, the result inferior to DempsterShafer theory. In summary, the solution for high conflicting still needs a more reasonable theory model. The new model should be capable of the incomplete of the fame of discernment, and the computation complexity should not larger than traditional DempsterShafer theory. Taking these into consideration, a new evidence theory, called generalized evidence theory (GET) is proposed in this paper.
It should be point out that traditional DempsterShafer evidence is based on the frame of discernment, and constraint the BPA must equal to 0. That is the classic DempsterShafer theory can only function in a close world. In this paper, close world means that the elements in the frame of discernment is exhaustive and complete. There are many cases in reallife. For example, the points of a dice only contain the six possibility: 1, 2, 3, 4, 5 and 6. However, in some application, due to lack of complete knowledge, it is possible to obtain a partial frame of discernment, but not a complete frame of discernment. As mentioned above, we know that enemy’s targets are ”a”, ”b” and ”c”, whereas there may exist a secret undeclared target ”d”. And none of the sensors can recognize the target ”d”, efficiently. In this situation, the discernment {a, b, c} is incomplete. This is the situation of open world. Along with the enriched knowledge, open world is absolute and close world is relative. Another example of SARS, before its appearances, the frame of discernment of pneumonia is sure not complete. Obviously, the traditional DempsterShafer evidence can only represent and function uncertain information in the close world, which limits the application of DempsterShafer evidence.
3.1 Basic Concepts of GET
Supposed is a frame of discernment of open world, the power set, denoted , is composed with the propositions, . A mass function is a mapping, , which satisfies the following conditions:
(14) 
Then, is the generalized basic probability assignment (GBPA) of the frame of discernment . The different between GBPA and traditional BPA is the Eq.(5). Notice that the is not necessary in GBPA. In other words, the empty set can also be a focal element. If , the GBPA degenerates as a traditional BPA.
The is used to model open world in GET. It should be emphasized that the in GET is not a common empty set, which means that a focal element or the unions of focal elements that out of the given frame of discernment. The in Definition 3.1 indicates that the is the focal elements outside of the frame of discernment, not the empty set in traditional BPA. Likewise, in Eq.(14) means GBPA assign some probability to the propositions, beyond the frame of discernment. For simplicity, the following of this paper without special explanation, BPA on behalf of GBPA, and the mass function simplifies as .
Similar to DempsterShafer theory, the generalized belief function (GBF) and generalized plausible function (GPL) in GET are defined as follows. Suppose a GBPA, the generalized belief function (GBF) is that GBel: , satisfied these conditions:
(15)  
(16) 
Suppose a GBPA, the generalized belief function (GPF) is that GPl: , satisfied these conditions:
(17)  
(18) 
Note that in Definition 3.1 and Definition 3.1, the values of and equal to , this is logical. Because is a proposition, beyond the fame of discernment, can not be supported by these propositions within the frame of discernment, and also unknown whether agreed with these propositions beyond the frame of discernment. GBF and GPF can be regarded as generalized lower bound and upper bound in which subset has been supported, respectively. It is obviously that
(19) 
Suppose there is a frame of discernment of , the GBPA is given as follows.
In this example, the GBPA assign just to these nonempty sets, that is =0. In this case, GBPA is the same to BPA.
These results show that while =0, the values of GBF and GPL in GBPA is the same to Bel and Pl in traditional BPA.
Suppose there is a frame of discernment of , the GBPA is given as follows.
In this example, the GBPA assign some value to the focal element . The GBF and GPL are as follows.
3.2 Generalized combination rule (GCR)
The classic Dempster’s combination rule can combine two BPAs and yield to a new BPA . Based on the classic Dempster’s combination rule, the generalized combination rule (GCR) is defined as follws. In generalized evidence theory, means that the intersection between empty set and empty set is still an empty set. Given two GBPA and , the generalized combination rule (GCR) are defined as follows.
(20)  
(21)  
(22)  
(23) 
The Characteristics of GCR, composed of Eqs.(20) (23), are summarized as follows.
(1) When , the GCR degenerate to Dempster’s combination rule.
(2) The combination result of two empty sets can be obtained by multiplied their GBPAs values.
(3) The factor in Eq.(20) is a normalized process that reassigned the GBPA values after deducting the obtained from Eq.(22). In other words, the process is to multiply these GBPA, which intersections are not empty, and accumulate, at last amplify times.
(4) Because of , the GBPA of conflicting coefficient is obtained after superposing Eq.(9) and Eq.(22).
There partial properties of generalized evidence theory (GET) are shown as follows. When
, GBPA degenerates to traditional BPA. More generally, if GBPA just assigns in the single elements, GBPA degenerates to the probability of probability theory.
For the GCR of GET, if
, then GCR degenerates to classic Dempster’s combination rule. More generally, when GBPA just assigns in the single elements, the results of GCR is the same to Bayesian probability.
The same as Dempster’s combination rule, GCR satisfies commutativity and associativity. This is that the combination results by GCR is unrelated to the orders of combination.
3.3 Generalized evidence distance
The generalize evidence distance in GET is defined as follows. Let and are two GBPAs on the frame discernment , the generalized evidence distance between GBPA and is defined as
(24) 
where is a matrix, the elements of is that
(25) 
The detail calculation process is that
and ,
is the inner product of the two vectors:
3.4 Application of GCR
In this subsection, numerical examples are used to show the applications of GCR. Assume a frame of discernment , two GBPAs are given as:
The process of calculation is that:
and
then
In this example, , the GCR is the same to the classic Dempster’s combination rule.
Suppose the frame of discernment , two GBPAs are given as:
The conflicting coefficient in GET is calculated as:
and
then
Thus, the final results are as follows
It is sure that, traditional Dempster’s combination rule is not suitable for this situation, because of . It is clearly that, after ascertained the value of , the rest probability is redistributed to other nonempty sets, in GCR. In this example, the probability of is zero, because the single set is not supported by any one of the two BPAs, in the frame of discernment. That is to say, the probability of single sets and are raised, owing to both of them are more or less supported by the two GBPAs.
Suppose the frame of discernment , two GBPAs are given as follows:
The conflicting coefficient in GET is calculated as:
thus,
From the view of GCR, we can first obtained the = 0.56. However, since the rest two propositions are not supported by each other, thus the rest probability can not assign to any one of them, and reassigned to . Therefore, the gets the twice assigned values, and = 0.56 + 0.44 = 1. We believe that this is reasonable. While the two GBPAs are highly conflicting and not supported by each others, the frame of discernment should be considered as an incomplete.
Suppose the frame of discernment , two GBPAs are given as follows:
The result is that
This example is similar to Example 3.4. Colloquially, can be viewed as a number 0, and any numbers multiply 0 is still 0. If a GBPA assigns the total probability to the set , the frame of discernment is incomplete.
4 Application and Discussion
4.1 The in DempsterShafer evidence and GET
In classic DempsterShafer theory, is indispensable. In view of the DempsterShafer theory, is a proposition without any supported by other propositions. That is non physical meaning of in DempsterShafer. The first observes and assigns the value to is the TBM by Smetssmets1994transferable . In TBM, if the conflict between two bodies of evidence is highly, the conflicting BPAs are assigned to . Basic belief assignment (BBA) in TBM is used to distinguish the traditional BPA in DempsterShafer theory. However, the BBA and BPA are both assigned on these nonempty sets, essentially. In other words, the restriction condition of is still necessary. Such an approach is questionable in logic. Now that the problem is in open world, TBM why not assign the value to at the step of generate BBAs? In addition, the approach of dealing with conflict is humble. Yager proposed a method that when the evidences are highly conflicting, the conflicting values are assigned to the whole set yager1987dempster . Yager’s method is contentious in the following research. In GET, while the GBPAs are generated, is permissible. And this is easy to understand, that is there may exist some hypothesises or propositions beyond the fixed frame of discernment. The value of indicates the open world degree of frame of discernment. Classic DempsterShafer theory is an extension of Bayes probability, and DempsterShafer theory can express and deal with uncertain information than Bayes probability, efficiently. And GET proposed in this paper, is an extension of classic DempsterShafer theory, and can express and deal with more uncertain information in the open world than DempsterShafer theroy in the close world.
4.2 Modified Liu’s conflict model
As mentioned in Subsection 2.4, Liu analysed the shortage of classic conflict coefficient , and proposed a conflict model. The main idea is that construct a two tuples by pignistic betting distance in TBMSmets1994191 and the classic conflict coefficient in DempsterShafer theoryDempster1967 ; Shafer1976 , and the union value to represent the conflict degree. That is when the value of is large, the two evidence can not be regarded as conflicting. While the value of is large, the two evidence can not be viewed as conflicting, also. Whether the evidences is conflict just ascertained by union with . The conflict model proposed is very interesting, since it provides a new thought of how to express the conflict between two bodies of evidence. There are some numerical examples by Liu’s conflict model. Let a frame of discernment of , and the two sensors and report that:
The classic conflict coefficient is that . The pignistic betting distance between this two evidence is that . The two couples values of Liu’s conflict model is that , and can be regard as non conflict. This result is the same to DempsterShafer theory. However, it is clearly that the information of is richer than . And the difference between the two evidence is given by . That is the weights of two evidence is different.
Let a frame of discernment of , there are two BPAs as and ,
Since and , , and the two BPAs are regarded as not conflict. This is irrational obvious, because the two BPAs provide different information, and the is more definite, and is total ignorance.
From the above two examples, we know that ”the same probability of occurrence” is the same to ”the total ignorance of system” in Liu’s conflict model. However, the real situation is not so simple. When we known nothing about the system, that is . It indicates that , , and so forth are possible, and the probability of uncertainty of system is the biggest. Thus pignistic betting distance can not distinguish the same probability and total ignorance, and not be a well represent model of conflict. Based on Jousselme’s evidence distancejousselme2001new , we propose a modified evidence model of conflict.
4.3 Conflict model of GET
In the previous subsection, we modify Liu’s conflict model by Eq.(26), and it is more reasonable. However, the conflict model is still in the close world, and can not be applied to the open world, which frame of discernment may be incomplete. In view of this, based on GET, the new conflict model of GET is proposed. Assume there are two GBPAs and on the frame of discernment , then the generalized conflict model is given as
(27) 
Where is the generalized conflict coefficient by Eq.(21), and is the generalized evidence distance by Eq.(24).
Compared with these existing methods, this new proposed generalized conflict model can handle with the information on the incomplete frame of discernment. When the frame of discernment is complete, the generalized conflict coefficient degenerates to classical coefficient in DempsterShafer theoryEq.(9), and the Eq.(27) degenerates to Eq.(26)
The below examples are used to show the application of generalized conflict coefficient and generalized conflict model. There are to GBPAs and on the frame of discernment. The first group GBPAs is varied, begin with and , and decreases progressively 0.1 each step , progressively increases 0.1 each step . The second group GBPAs is also varied, begin with and , and decreases progressively 0.1 each step , progressively increases 0.1 each step . Then the generalized conflict coefficient between the two GBPAs is shown in Fig.1.
Fig.1 indicates that, when ; and ; , the generalized conflict coefficient is the smallest, and the proposition is absolutely supported by the system and the frame of discernment is complete. While decreases progressively 0.1 each step, progressively increases 0.1 each step, the generalized conflict coefficient is also gradually increased, and this situation indicates the incomplete degree of the frame of discernment is bigger and bigger. According to GCR, when appeared in anyone of these GBPAs, the generalized conflict coefficient achieves the maximum value 1.
Suppose there are two GBPAs on the frame of discernment. The first group GBPAs is varied, begin with and , and progressively increases 0.1 each step, decreases progressively 0.1 each step. The second group GBPAs is also varied, begin with and , and progressively increases 0.1 each step , decreases progressively 0.1 each step. Then the generalized evidence distance between the two GBPAs is shown in Fig.2.
From Fig.2, we can see that the generalized evidence distance on axis keeps 0, because the two GBPs are the same.
Let two GBPAs on the same discernment, and the two GBPAs are given as
Fig.3 indicates that there is just one class on the frame of discernment. The two dotted triangles represent the classes a and c, respectively, and both of them do not appeared on this frame of discernment. In fact, the sensors and belong to classes and , respectively. If generalized evidence distance is considered merely, the error result will be obtained. This example points out that generalized conflict coefficient to measure the conflict is better than generalized evidence distance on the incomplete frame of discernment. The two below examples are applied to the complete frame of discernment with .
Suppose a frame of discernment , the first fixed BPA is that: . The second varied BPA begin with . And progressively increases 0.1 each step , decreases progressively 0.1 each step. Then the evidence distance and pignistic betting distance between and are shown in Fig.4.
From Fig.4, When , the pignistic betting distance between two BPAs is 0, which indicates no conflict between BPAs. However, at this time step, the value of evidence distance is 0.5, which indicates that there exists conflict between BPAs. It is obviously that, when the frame of discernment is complete, the measure of conflict model should be evidence distance but not the pignistic betting distance.
Let a frame of discernment of , and the two BPAs are defined as follows
where is a varied subset of , which increments one more element at a time, and begins with A=, ending with Case 20 when . The evidence distance and classic conflict coefficient between two BPAs are given in Table 1 and Fig.5.
Cases  

A={1}  0.7916  0.6 
A={1,2}  0.7024  0.6 
A={1,2,3}  0.6  0.6 
A={1,…,4}  0.6782  0.6 
A={1,…,5}  0.7211  0.6 
A={1,…,6}  0.7483  0.6 
A={1,…,7}  0.7982  0.6 
A={1,…,8}  0.8  0.6 
A={1,…,9}  0.8083  0.6 
A={1,…,10}  0.8149  0.6 
A={1,…,11}  0.8202  0.6 
A={1,…,12}  0.8246  0.6 
A={1,…,13}  0.8283  0.6 
A={1,…,14}  0.8315  0.6 
A={1,…,15}  0.8343  0.6 
A={1,…,16}  0.8367  0.6 
A={1,…,17}  0.8388  0.6 
A={1,…,18}  0.8406  0.6 
A={1,…,19}  0.8423  0.6 
A={1,…,20}  0.8438  0.6 
From Table 1 and Fig.5, we can see that, no matter how the subset changed, the classic conflict coefficient keeps the value 0.6. This is irrational. The generalized evidence distance is changed along with subset , and it can measure the conflict efficiently. That is the the generalized evidence distance to measure the conflict is better than generalized conflict coefficient in the case of complete frame of discernment.
These numerical examples indicate the applications of the new conflict model. When the frame of discernment is incomplete and , the twotuples of conflict model to measure the conflict should depend on generalized conflict coefficient mainly. On the other hand, when the frame of discernment is complete and , the twotuples of conflict model to measure the conflict should depend on generalized evidence distance mainly.
5 Conclusions
In this paper, a novel theory called generalized evidence theory is systematically proposed. Some key points can be concluded as follows.
First, evidence theory is an efficient tool to handle the information fusion under uncertain environment.Under the condition that we definitely know that we are in close world, the Dempster combination rule is enough to combine the evidence from different sources. If the evidence conflicts with each other, with more and more evidence are collected, the conflict evidence combination can be well addressed by taking the reliability of each evidence into consideration.
Second, generalized evidence theory (GET)is an extension of DempsterShafer theory, since the strict restriction condition of is abandoned. In GET, is regards as a element, which means unknown but not just empty, with all of the property of other elements. That is GET can deal with more uncertain information fusion in the open world than DempsterShafer theory. GET inherits all the merits of DempsterShafer theory, and expands the application of DempsterShafer theory, from close world to open world. When the frame of discernment is complete, and , GET degenerates to DempsterShafer theory. And the generalized combination rule (GCR) is provided in GET, the distinction between GCR and Dempster combination rule is .
Third, the generalized conflict model is given based on GET. Compared with these existing conflict model, the new conflict model is generalized. Under different conditions, how to apply the generalized conflict model is provided. That is when the frame of discernment is incomplete, the measure of generalized conflict should take advantage of generalized conflict coefficient in main. When the frame of discernment is complete, the measure of generalized conflict should take advantage of evidence distance in main.
It should be pointed out that, besides the conflict management in open world, the exclusive condition in Dempster Shafer evidence theory should also be paid great attention. One of the ongoing works is a new theory called D numbers theory which focuses on the mutual exclusion in evidence theory. In classical evidence theory, it is assumed that each hypothesis is exclusive to each other. However, it is not reasonable in real world. For example, given two linguistic values Good and Very Good, the is not accepted under the framework of Dempster Shafer evidence theory since that the two linguistic values are not exclusive with each other. To address this limitation, a novel theory called D numbers theory is proposed Deng2012 . Both the D numbers theory and the proposed GET in this paper are generalization of the evidence theory, providing a more flexible and reasonable way to handle the uncertainty in our real world deng2014supplier ; deng2014environmental ; liu2014failure .
Acknowledgements
The author greatly appreciates Professor Shan Zhong, the China academician of the Academy of Engineering, for his encouragement to do this research. The author also greatly appreciates Professor Yugeng Xi in Shanghai Jiao Tong University for his support to this work. Professor Sankaran Mahadevan in Vanderbilt University discussed many relative topics about this work. Dr. Deqiang Han and Dr. Wen Jiang discussed the topic of conflict management in this paper. My Ph.D students in Shanghai Jiao Tong University, Peida XU and Xiaoyan Su, do some numerical experiments of this work. The author’s Ph.D students in Southwest University, Xinyang Deng, Ya Li and Daijun Wei, graduate students in Southwest University Yajuan Zhang, Bingyi Kang, Xiaoge Zhang, Shiyu Chen, Yuxian Du, Cai Gao, have discussed the conflict evidence management. Mr. Hongming mo did some editorial works of this submission. This work is partially supported by National Natural Science Foundation of China, Grant Nos. 30400067, 60874105, 60904099, 61174022, Chongqing Natural Science Foundation, Grant No. CSCT, 2010BA2003, Program for New Century Excellent Talents in University, Grant No.NCET080345, Shanghai RisingStar Program Grant No.09QA1402900, the Chenxing Scholarship Youth Found of Shanghai Jiao Tong University, Grant No.T241460612, Doctor Funding of Southwest University Grant No. SWU110021, the open funding project of State Key Laboratory of Virtual Reality Technology and Systems, Beihang University (Grant No.BUAAVR14KF02).
References
 (1) A. Dempster, Upper and lower probabilities induced by a multivalued mapping, Annals of Mathematics and Statistics 38 (2) (1967) 325–339.
 (2) G. Shafer, A Mathematical Theory of Evidence, Princeton University Press, Princeton, 1976.
 (3) J.B. Yang, M. G. Singh, An evidential reasoning approach for multipleattribute decision making with uncertainty, IEEE Transactions on Systems, Man and Cybernetics 24 (1) (1994) 1–18.
 (4) R. R. Yager, On the aggregation of prioritized belief structures, IEEE Transactions on Systems, Man and Cybernetics, Part A: Systems and Humans 26 (6) (1996) 708–717.
 (5) T. Denœux, M.H. Masson, EVCLUS: evidential clustering of proximity data, IEEE Transactions on Systems, Man, and Cybernetics, Part B: Cybernetics 34 (1) (2004) 95–109.
 (6) Z. Elouedi, K. Mellouli, P. Smets, Assessing sensor reliability for multisensor data fusion within the transferable belief model, Systems, Man, and Cybernetics, Part B: Cybernetics, IEEE Transactions on 34 (1) (2004) 782–787.
 (7) A. P. Dempster, W. F. Chiu, DempsterShafer models for object recognition and classification, International Journal of Intelligent Systems 21 (3) (2006) 283–297.
 (8) J.B. Yang, Y.M. Wang, D.L. Xu, K.S. Chin, The evidential reasoning approach for mada under both probabilistic and fuzzy uncertainties, European journal of operational research 171 (1) (2006) 309–343.
 (9) B.S. Yang, K. J. Kim, Application of DempsterShafer theory in fault diagnosis of induction motors using vibration and current signals, Mechanical Systems and Signal Processing 20 (2) (2006) 403–420.
 (10) A.L. Jousselme, C. Liu, D. Grenier, E. Bosse, Measuring ambiguity in the evidence theory, IEEE Transactions on Systems, Man and Cybernetics, Part A: Systems and Humans 36 (5) (2006) 890–903.
 (11) F. Cuzzolin, Two new bayesian approximations of belief functions based on convex geometry, Systems, Man, and Cybernetics, Part B: Cybernetics, IEEE Transactions on 37 (4) (2007) 993–1008.
 (12) F. Cuzzolin, A geometric approach to the theory of evidence, Systems, Man, and Cybernetics, Part C: Applications and Reviews, IEEE Transactions on 38 (4) (2008) 522–534.
 (13) G. J. Klir, H. Lewis, Remarks on “Measuring ambiguity in the evidence theory”, IEEE Transactions on Systems, Man and Cybernetics, Part A: Systems and Humans 38 (4) (2008) 995–999.

(14)
A. P. Dempster, A generalization of Bayesian inference, in: Classic Works of the DempsterShafer Theory of Belief Functions, 2008, pp. 73–104.
 (15) L. Utkin, S. Destercke, Computing expectations with continuous pboxes: Univariate case, International Journal of Approximate Reasoning 50 (5) (2009) 778–798.
 (16) J. Dezert, D. Han, Z. Liu, J.M. Tacnet, Hierarchical proportional redistribution for bba approximation, in: Belief Functions: Theory and Applications, Springer, 2012, pp. 275–283.
 (17) T. Denoeux, M.H. Masson, Evidential reasoning in large partially ordered sets, Annals of Operations Research 195 (1) (2012) 135–161.
 (18) B. Kang, Y. Deng, R. Sadiq, S. Mahadevan, Evidential cognitive maps, KnowledgeBased Systems 35 (2012) 77–86.
 (19) D. Wei, X. Deng, X. Zhang, Y. Deng, S. Mahadevan, Identifying influential nodes in weighted networks based on evidence theory, Physica A: Statistical Mechanics and its Applications 392 (10) (2013) 2564–2575.
 (20) R. R. Yager, DempsterShafer structures with general measures, International Journal of General Systems 41 (4) (2012) 395–408.
 (21) M.H. Masson, T. Denoeux, Ensemble clustering in the belief functions framework, International Journal of Approximate Reasoning 52 (1) (2011) 92–109.
 (22) Y. Zhang, X. Deng, D. Wei, Y. Deng, Assessment of ECommerce security using AHP and evidential reasoning, Expert Systems with Applications 39 (3) (2012) 3611–3623.
 (23) Y. Deng, R. Sadiq, W. Jiang, S. Tesfamariam, Risk analysis in a linguistic environment: a fuzzy evidential reasoningbased approach, Expert Systems with Applications 38 (12) (2011) 15438–15446.
 (24) Y. Deng, W. Jiang, R. Sadiq, Modeling contaminant intrusion in water distribution networks: A new similaritybased DST method, Expert Systems with Applications 38 (1) (2011) 571–578.
 (25) Y. Deng, F. T. Chan, Y. Wu, D. Wang, A new linguistic MCDM method based on multiplecriterion data fusion, Expert Systems with Applications 38 (6) (2011) 6985–6993.
 (26) H. T. Nguyen, On belief functions and random sets, in: Belief Functions: Theory and Applications, Springer, 2012, pp. 1–19.
 (27) S. Huang, X. Su, Y. Hu, S. Mahadevan, Y. Deng, A new decisionmaking method by incomplete preferences based on evidence distance, KnowledgeBased Systems (56) (2014) 264–272.

(28)
Z.g. Liu, Q. Pan, J. Dezert, Evidential classifier for imprecise data based on belief functions, KnowledgeBased Systems 52 (2013) 246–257.

(29)
T. Denoeux, Maximum likelihood estimation from uncertain data in the belief function framework, IEEE Transactions on knowledge and data engineering 25 (1) (2013) 119–130.

(30)
Z.g. Liu, Q. Pan, J. Dezert, G. Mercier, Credal classification rule for uncertain data based on belief functions, Pattern Recognition 47 (7) (2014) 2532–2541.
 (31) L. A. Zadeh, A simple view of the DempsterShafer theory of evidence and its implication for the rule of combination, AI magazine 7 (2) (1986) 85.
 (32) C. K. Murphy, Combining belief functions when evidence conflicts, Decision support systems 29 (1) (2000) 1–9.
 (33) E. Lefevre, O. Colot, P. Vannoorenberghe, Belief function combination and conflict management, Information fusion 3 (2) (2002) 149–162.
 (34) D. Yong, S. WenKang, Z. ZhenFu, L. Qi, Combining belief functions based on distance of evidence, Decision Support Systems 38 (3) (2004) 489–493.
 (35) W. Liu, Analyzing the degree of conflict among belief functions, Artificial Intelligence 170 (11) (2006) 909–924.
 (36) R. Haenni, Shedding new light on Zadeh’s criticism of Dempster’s rule of combination, in: 2005 7th International Conference on Information Fusion, Vol. 2, 2005, pp. 879–884.
 (37) S. Destercke, T. Burger, Toward an axiomatic definition of conflict between belief functions, IEEE Transactions on Cybernetics 43 (2) (2013) 585–596.
 (38) J. Schubert, Conflict management in DempsterShafer theory using the degree of falsity, International Journal of Approximate Reasoning 52 (3) (2011) 449–460.
 (39) Y. Yang, D. Han, C. Han, Discounted combination of unreliable evidence using degree of disagreement, International Journal of Approximate Reasoning 54 (8) (2013) 1197–1216.
 (40) A. Roquel, S. Le HégaratMascle, I. Bloch, B. Vincke, A new local measure of disagreement between belief functions–application to localization, in: Belief Functions: Theory and Applications, Springer, 2012, pp. 335–342.
 (41) E. Lefèvre, Z. Elouedi, How to preserve the conflict as an alarm in the combination of belief functions?, Decision Support Systems 56 (2013) 326–333.
 (42) A. SarabiJamab, B. N. Araabi, T. Augustin, Informationbased dissimilarity assessment in dempster–shafer theory, KnowledgeBased Systems 54 (2013) 114–127.
 (43) J.B. Yang, D.L. Xu, Evidential reasoning rule for evidence combination, Artificial Intelligence 205 (2013) 1–29.
 (44) R. R. Yager, On the DempsterShafer framework and new combination rules, Information sciences 41 (2) (1987) 93–137.
 (45) P. Smets, R. Kennes, The transferable belief model, Artificial intelligence 66 (2) (1994) 191–234.
 (46) P. Smets, Decision making in the tbm: the necessity of the pignistic transformation, International Journal of Approximate Reasoning 38 (2) (2005) 133–147.
 (47) D. Dubois, H. Prade, Representation and combination of uncertainty with belief functions and possibility measures, Computational Intelligence 4 (3) (1988) 244–264.
 (48) R. Haenni, Are alternatives to dempster’s rule of combination real alternatives?: Comments on about the belief function combination and the conflict management problem —lefevre et al, Information Fusion 3 (3) (2002) 237–239.
 (49) A.L. Jousselme, D. Grenier, É. Bossé, A new distance between two bodies of evidence, Information Fusion 2 (2) (2001) 91–101.
 (50) P. Smets, R. Kennes, The transferable belief model, Artificial Intelligence 66 (2) (1994) 191–234.
 (51) A. Ayoun, P. Smets, Data association in multitarget detection using the transferable belief model, International Journal of Intelligent Systems 16 (10) (2001) 1167–1182.
 (52) F. Voorbraak, A computationally efficient approximation of dempstershafer theory, International Journal of ManMachine Studies 30 (5) (1989) 525–536.
 (53) R. Haenni, N. Lehmann, Resource bounded and anytime approximation of belief function computations, International Journal of Approximate Reasoning 31 (1) (2002) 103–154.
 (54) J. Dezert, F. Smarandache, Dsmt: A new paradigm shift for information fusion, arXiv preprint cs/0610175.
 (55) Y. Deng, D numbers: Theory and applications, Journal of Information and Computational Science 9 (9) (2012) 2421–2428.
 (56) X. Deng, Y. Hu, Y. Deng, S. Mahadevan, Supplier selection using AHP methodology extended by D numbers, Expert Systems with Applications 41 (1) (2014) 156–167.
 (57) X. Deng, Y. Hu, Y. Deng, S. Mahadevan, Environmental impact assessment based on D numbers, Expert Systems with Applications 41 (2) (2014) 635–643.
 (58) H.C. Liu, J.X. You, X.J. Fan, Q.L. Lin, Failure mode and effects analysis using d numbers and grey relational projection method, Expert Systems with Applications 41 (10) (2014) 4670–4679.
Comments
There are no comments yet.