CN104899762A - Trust management method based on backward inference - Google Patents

Abstract

The invention discloses a trust management method based on backward inference. The trust management method based on the backward inference comprise a trust evaluation module and a trust punishment module, wherein an operation process of the trust evaluation module comprises the following steps: generating satisfaction degree evaluation among nodes through transaction among the nodes, and adopting a multiple evaluation mechanism to calculate a direct trust value of the evaluated node; utilizing a cosine similarity formula to obtain a similar node set, and calculating an indirect trust value of the evaluated node; combining the direct trust value and the indirect trust value to calculate a comprehensive trust value of the evaluated node; comparing the comprehensive trust value with a trust threshold value, and marking the evaluated node; and generating a node sequence with a mark. The operation process of the trust punishment module comprises the following steps: inquiring the comprehensive trust value for the mark node by other nodes; according to different node marks, looking for and punishing a malicious node; punishing the malicious node to lower the reputation value of the malicious node; calculating the final reputation values of all nodes; and generating a node sequence subjected to devaluation sorting according to the reputation values. The punishment mechanism is introduced in, so that an evaluation result is more accurate.

Description

A kind of trust management method based on backward inference

Technical field

The invention belongs to the trust evaluation field of distributed system interior joint, particularly relate to for camouflage node, a kind of trust management method based on backward inference.

Background technology

In recent years, distributed system, as grid computing, general fit calculation, P2P calculating, Ad Hoc network etc., because of advantages such as its reliability are high, transmission delay is little, net interior nodes shared resource is easy, is widely used.Compare with integrated system, the distributed task that is conducive to is carried out distributing and is optimized in whole computer system, overcomes traditional integrated system and can cause the nervous defect with responding bottleneck of central host resource.On the other hand, data scatter is stored in multiple stage independently in machinery and equipment by distributed system, and do not have Centroid, but its topological structure is more complicated, thus its safety problem is also more outstanding.

Trust management is one and carrys out the process of node metric confidence level by method for evaluating trust.Trust management can be divided into the trust management of trust management based on voucher and Behavior-based control.Based on voucher trust management technology because of its subjectivity more weak, be suitable only for legacy network, effectively cannot be applied to complicated distributed network, therefore, the trust management technology of current scholar's many employings Behavior-based control carrys out the security of Logistics networks.Trust management technology, according to internodal transaction record, sets up the method for evaluating trust such as internodal direct trust, indirectly trust, with the confidence level of this critic network interior joint.

At present, obtain some progress about the trust management model of internet security and appraisal procedure thereof, but had in the work of camouflage node of deceptive practices in identification, still in research further.Patents mainly contains: 1) method for evaluating trust (CN101764821A) of user behavior " in a kind of trustable network ".The method, by historical transaction record, provides decision-making foundation for assessing.The method considers the impact of time on trust value, but does not consider the situation of the dishonest transaction of malicious node in network.2) " a kind of method for evaluating trust based on blur filter (CN104009993A) ".The method adopts multiple evaluation mechanism, by the degree of belief of blur filter assessment node, reduces the probability that node whitens, but does not mention the penalty mechanism of malicious node.3) " sensor network trust evaluating method (CN101835158B) based on nodes ' behavior and D-S evidence theory ", design wireless sensor network interior joint trust-factor strategy, utilize the degree of membership of fuzzy set theory, ensure that the security that data are transmitted, but do not associate with the method that this patent proposes.In addition correlative theses mainly contains: 1) people (the Su Z such as Zhiyuan Su, Liu L, Li M, et al.ServiceTrust:trust management in service provision networks [C] //Services Computing (SCC), 2013IEEE International Conference on.IEEE, 2013:272-279.) ServiceTrust model is proposed, multiple evaluation mechanism is adopted to calculate local trust value, and in global trusting value calculates, consider the difference of behavior between user, but ordinary node can be included in insincere set by this method by mistake, resource request is caused only to be carried out in sub-fraction trusted node, thus the load pressure of aggravation node.2) people (Zhang Shibin such as Zhang Shibin, Xu Chunxiang. the method for evaluating trust based on cloud model is studied [J]. Chinese journal of computers, 2013,36 (2): 422-431.) the trust evaluation algorithm based on cloud model is proposed, consider the factor that multiple impact is trusted comprehensively to obtain the reliability rating of entity according to Similarity Measure.The method is improved for the appraisement system that the higher trust of price is higher, prevents malicious entities from carrying out credit propagation by expensive goods, but the method is not considered to pretend entity.3) people (Zhang Lin such as Zhang Lin, Liu Jingwen, Wang Ruchuan, Deng. based on the Trust Valuation Model [J] improving D-S evidence theory. communication journal, 2013,34 (7): 167-173.) what propose that a kind of appraisal procedure obtains from local evidence storehouse calculates about evaluating the collateral information obtained in the evidence sequence of object and nominator's set the trust value evaluating object, prevent malicious node from assembling and trust attack, but the method is only for continuing the malicious node providing insincere service.4) people (Liu Yanbing such as Liu Yanbing; Gong Xuehong; Feng Yanfen. based on the method for evaluating trust [J] of Internet of things node behavioral value. communication journal; 2014; 35 (5): 8-15.) model proposed by the direct trust value of Behavior-based control feature, combine based on the recommendation trust of adjacent node and historical statistics trust value, the node administer doses of punishment mechanism low to comprehensive trust value.The method only abandons the abnormal behaviour such as grouping or a large amount of repetitive requests for node in transmitting procedure, waits behavior invalid to the untrue evaluation of malicious node feedback.

In sum, a kind of trust management system efficiently effectively must can identify malicious node, prevents malicious node from destroying network, the security of Logistics networks.Trust management system can identification camouflage node is the important indicator of evaluation model, but the large multipair common malicious node of existing trust model is effective, poor to the identification of camouflage node.

Summary of the invention

The object of this invention is to provide and effectively can identify malicious node, particularly there is the camouflage node of deceptive practices, a kind of trust management method based on backward inference.

The present invention is achieved by the following technical solutions:

Based on a trust management method for backward inference, comprise trust evaluation module and trust punishment module,

The operating process of trust evaluation module comprises following concrete steps:

Step 1.1: after recording each two node transaction, node i is at t _xmoment to the satisfaction of node j,

tr _x(i，j)＝(sat _ij，t _x)，sat _ij∈[1，5]，i，j∈N

Wherein, sat _ijfor once mutual interior joint i is to node j satisfaction, N is the sum set of system interior joint i;

Step 1.2: in time interval t, computing node i is to the direct trust value of node j:

${\mathrm{TD}}_{\mathrm{ij}}=\underset{t_x\∈t}{\mathrm{\Σ}}\mathrm{tr}(i,j)/n,i,j\∈N$

Wherein, n is the transaction count that node i and node j produce;

Step 1.3: utilize cosine similarity to try to achieve and node i behavior similar node collection;

Step 1.4: computing node i is to the indirect trust values of node j:

${\mathrm{TI}}_{\mathrm{ij}}=\frac{{\mathrm{\Σ}}_{m\∈S}{\mathrm{TD}}_{\mathrm{mj}}\·{\mathrm{RV}}_m}{\left|S\right|},i,j\∈N$

Wherein, | S| is and node i behavior similar node collection, RV _ifor the credit value of node i;

Step 1.5: in conjunction with direct trust value and indirect trust values, weighted calculation node i is to the comprehensive trust value of node j:

TC _ij＝α·TD _ij+β·TI _ij

Wherein, weight factor alpha+beta=1;

Step 1.6: compared with trust threshold by comprehensive trust value, if comprehensive trust value is less than trust threshold, is labeled as incredible by node j, otherwise is labeled as believable by node j;

Step 1.7: repeat step 1.1 ~ step 1.7, generates the sequence node of tape label;

Inputed to by the sequence node of trust evaluation CMOS macro cell tape label and trust punishment module, the operating process of trusting punishment module comprises following concrete steps:

Step 2.1: the difference marked according to the sequence node interior joint of tape label, adopts corresponding backstepping mode, find and punish malicious node, obtains the final punishment number of times of malicious node;

Step 2.2: the credit value reducing malicious node according to the punishment number of times of malicious node;

Step 2.3: according to the comprehensive trust value of other nodes to present node and the credit value of present node self, calculates the final credit value of present node;

Step 2.4: repeat step 2.3, calculate the final credit value of all nodes, generates the sequence node pressing the sequence of credit value depreciation.

A kind of trust management method based on backward inference of the present invention, can also comprise:

1, according to the difference that the sequence node interior joint of tape label marks, adopt corresponding backstepping mode, find and punish that the method for malicious node is:

Step a1: the node a in the sequence node of select tape mark, the mark of query node, if the mark of node a is believable, forwards step a2 to, otherwise forwards step a4 to;

Step a2: query node k comprehensive trust value to node a, the size of more comprehensive trust value and trust threshold, if be greater than threshold value, give up node k, forwards step a3 to; Otherwise the punishment number of times of node k is added 1;

Step a3: judge have the node of historical trading whether to inquire about with node a complete, if so, carry out step a6, otherwise another node k is next node, repeats step 2.2 ~ step 2.3, until travel through all nodes;

Step a4: query node k comprehensive trust value to node a, the size of more comprehensive trust value and trust threshold, if be less than threshold value, give up node k, forwards step a5 to, otherwise the punishment number of times of node k is added 1;

Step a5: judge have the node of historical trading whether to inquire about with node a complete, if so, carry out step a6, otherwise another node k is next node, repeats step a4 ~ step a5, until travel through all nodes;

Step a6: the final punishment number of times obtaining all nodes, punishment number of times be not 0 node be malicious node.

2, according to the method for the credit value of the punishment number of times reduction malicious node of malicious node be:

The punishment number of times of query node m, if punishment number of times is not 0, reduces the credit value of node m, obtains node m by the credit value after punishing:

${{\mathrm{RV}}_m}^{\′}={\mathrm{RV}}_m\·e^{{-\mathrm{CN}}_m}$

Wherein, CN _mfor the punishment number of times of node m.

3, according to the credit value of other nodes to the comprehensive trust value of present node m and present node m self, the final credit value calculating present node m is:

RV _m ^(k+1)＝RV ₁TC _1m ^(k)+…+RV _jTC _jm ^(k)+…+RV _nTC _nm ^(k)，n，m，j∈N。

Beneficial effect:

The trust management system based on backward inference that the present invention proposes and method thereof, the beneficial effect had comprises: no longer only rely on the transaction record of node to carry out trust evaluation, introduces penalty mechanism, makes assessment result more accurate; After preliminary trust evaluation, node is divided into two classes, by the behavior of different research of sorts nodes, reduces the otherness of similar nodes ' behavior, ensure the accuracy of assessment; Adopt the thought of trusting backstepping, the efficient identification malicious node of energy, particularly has the camouflage node of deceptive practices.The present invention is on the basis setting up both sides' trusting relationship, and with the angle of evaluated entity, reverse searching malice assesses node.The method effectively can identify malicious node, particularly has the camouflage node of deceptive practices.

Accompanying drawing explanation

Fig. 1 is the trust management model frame diagram based on trusting backstepping;

Fig. 2 is trust evaluation block flow diagram;

Fig. 3 is for trusting punishment block flow diagram;

Fig. 4 is symbol description table.

Embodiment

Below in conjunction with accompanying drawing, the present invention is described in further details.

The trust management system based on backward inference that the present invention proposes and method thereof, break through traditional trust evaluation thinking, adopt backstepping thought, according to the reverse searching malicious node of the trust value of node, the method effectively can identify malicious node, particularly has the camouflage node of deceptive practices, and reduces their trust value, prevent malicious node and other nodes of camouflage node cheating, thus ensure the security of network.

The model that the present invention proposes is divided into trust evaluation module and trusts punishment module, as shown in Figure 1.1) for trust evaluation module, produce internodal satisfaction evaluation by transaction between node, adopt multiple evaluation mechanism to calculate the direct trust value of evaluated node; Utilize cosine similarity formula to try to achieve similar node collection, calculate the indirect trust values of evaluated node; In conjunction with direct trust value and indirect trust values, weighted sum calculates the comprehensive trust value of evaluated node; The size of more comprehensive trust value and trust threshold, mark is evaluated node; Generate markd sequence node, for trust punishment module afterwards.2) for trust punishment module, the comprehensive trust value of other nodes to flag node is inquired about; According to the difference of vertex ticks, adopt different backstepping methods, find and punish malicious node; Punishment malicious node, reduces its credit value; According to the credit value of node to the trust value of other node and node self, the final credit value of all nodes of iterative computation; Generate the sequence node pressing the sequence of credit value depreciation, comprise trusted node sequence and insincere sequence node.

The symbol description related in the inventive method as shown in Figure 4

1, trust evaluation module

In trust evaluation module, have the trading activity of request resource between node and node, when node i is to node j request resource, node j can send a file to node i, and node i is according to the authenticity obtaining file, and each node j that gives feeds back a satisfaction, sat _ij.The scope of satisfaction is divided into " 1 " to arrive " 5 " these 5 numerical value, and " 1 " represents that transaction content and expection do not meet, and very dissatisfied, " 5 " represent that this transaction is very satisfied, and intermediate value represents transaction satisfactory value in various degree.Here adopt multiple evaluation mechanism, avoid the satisfaction evaluation in the past only having polarization.Module was added up within the t time interval, the transaction record of node.

The concrete steps of trust evaluation module are as follows, see Fig. 2:

1) each closing the transaction, record node i is at t _xmoment, formula was to the satisfaction evaluation of node j:

tr _x(i，j)＝(sat _ij，t _x)，sat _ij∈[1，5]，i，j∈N (1-1)

2) direct trust value calculates.Be defined in the t time interval, entity i and entity j creates n time and concludes the business, then i is to the direct trust value of j, TD _ijcomputing formula be:

${\mathrm{TD}}_{\mathrm{ij}}={\mathrm{\Σ}}_{t_x\∈t}\mathrm{tr}(i,j)/n,i,j\∈N---(1-2)$

3) the similar node of behavior is found.Utilize the behavior similarity S of cosine similarity formulae discovery node A and Node B _aB, or then S _aB=0.With Node B from cosine similarity principle, S _aBget over 1, represent that the behavior of A and B is more close.S _aBcomputing formula be:

$S_{\mathrm{AB}}=\frac{{\mathrm{\Σ}}_{k=1}^n{\mathrm{TD}}_{\mathrm{Ak}}\·{\mathrm{TD}}_{\mathrm{Bk}}}{\sqrt{{\mathrm{\Σ}}_{k=1}^n{{\mathrm{TD}}_{\mathrm{Ak}}}^2}\·\sqrt{{\mathrm{\Σ}}_{k=1}^n{{\mathrm{TD}}_{\mathrm{Bk}}}^2}},A,B,k\∈N---(1-3)$

4) indirect trust values calculates.By reference to the set of node similar to node i to the direct trust value computing node i of node j to the indirect trust values of node j, | S| represents the node set similar to node i, TI _ijcomputing formula be:

${\mathrm{TI}}_{\mathrm{ij}}=\frac{{\mathrm{\Σ}}_{m\∈S}{\mathrm{TD}}_{\mathrm{mj}}\·{\mathrm{RV}}_m}{\left|S\right|},i,j\∈N---(1-4)$

5) comprehensive trust value computing.In conjunction with direct trust value and indirect trust values, weighted calculation node i to the comprehensive trust value of node j, wherein alpha+beta=1, TC _ijcomputing formula is:

TC _ij＝α·TD _ij+β·TI _ij(1-5)

6) comprehensive trust value compares with trust threshold.If comprehensive trust value is less than trust threshold, to step 7).Otherwise to step 8);

7) node is marked as incredible, to step 9);

8) node is marked as believable;

9) sequence node of tape label is generated.

2, punishment module is trusted

Trust punishment module and adopt backstepping thought, from being evaluated node, the reverse evaluation node finding malice, and depreciation punishment is carried out to malicious node.In trust evaluation module, according to comprehensive trust value, node is marked, for trusted node i, if there is other nodes j to give node i low evaluation of estimate, then think that node j has the tendency of malicious node; For insincere node i, if there is other nodes k to give node i high evaluation value, then think that node k has the tendency of malicious node.According to the punishment number of times CN added up _m, to the depreciation punishment having the node m of malice tendency to carry out respective degrees, the method reaches the effect reducing malicious node credit value.

The concrete steps of trusting punishment module are see Fig. 3:

1) sequence node of the tape label generated in trust evaluation module is input as;

2) select node i, the mark of query node, if believable, to step 3), otherwise to step 8);

3) query node j is to the comprehensive trust value of node i;

4) size of more comprehensive trust value and trust threshold, if be greater than threshold value, to step 5), otherwise to step 6);

5) this node is given up, to step 7);

6) the punishment number of times of node j, CN _jadd 1;

7) judge and node i whether complete, if then arrive step 13 if having the node of historical trading all to inquire about), otherwise arrive step 3), inquire about next node;

8) query node k is to the comprehensive trust value of node i;

9) size of more comprehensive trust value and trust threshold, if be greater than threshold value, to step 10), otherwise to step 11);

10) this node is given up, to step 12);

11) the punishment number of times of node j, CN _kadd 1;

12) judge and node i whether complete, if then arrive step 13 if having the node of historical trading all to inquire about), otherwise arrive step 8), inquire about next node;

13) the punishment number of times CN of query node m _mif, CN _mbe not 0, to step 14), otherwise to step 17);

14) punish node m, reduce the credit value of node m, RV _m' be the new credit value after punishment, node punishment formula is:

${{\mathrm{RV}}_m}^{\′}={\mathrm{RV}}_m\·e^{{-\mathrm{CN}}_m}---(2-1)$

15) according to the credit value of other nodes to the trust value of node m and node m self, the final credit value of iterative computation node m, computing formula is:

RV _m ^(k+1)＝RV ₁TC _1m ^(k)+…+RV _jTC _jm ^(k)+…+RV _nTC _nm ^(k)，n，m，j∈N (2-2)

16) sequence that node presses credit value descending sort is generated;

17) terminate.

Fig. 2 is trust evaluation block flow diagram, and a specific embodiment is as follows:

1) record the time after each two node transaction and satisfaction, first node i applies for resource to node j, and node j sends resource to node i after receiving application, then after node i receives resource, think and meet needed for it, then the satisfaction evaluation 5 provided, be then recorded as:

tr(i，j)＝(5，3)

2) direct trust value calculates.Within 0 ~ 10s time interval, entity i and entity j creates 5 times and concludes the business, and transaction record is respectively tr (i, j)=(4,1), (5,3), (5,5), (4,6), (5,8), then i is to the direct trust value of j, TD _ijcomputing formula be:

TD _ij＝Σ(4+5+5+4+5)/5

3) the similar node of behavior is found.The transaction record of node A and node k is: TD _ak={ 4,5,5,4,5}, the transaction record of Node B and node k is: TD _bk={ 4,4,5,5,4}, S _aBcomputing formula be:

$S_{\mathrm{AB}}=\frac{\mathrm{\Σ}(4*4+5*4+5*5+4*5+4*4)}{\sqrt{\mathrm{\Σ}(4^2+5^2+5^2+4^2+5^2)}\·\sqrt{\mathrm{\Σ}(4^2+4^2+5^2+5^2+4^2)}}$

4) indirect trust values calculates.The node set similar to node i | S|={m, n}, wherein similar node to the direct trust value of node j is: TD _mj=4.6, TD _nj=4.4, the credit value of similar node self is: RV _m=0.6, RV _n=0.7 node i is to the indirect trust values TI of node j _ijcomputing formula be:

TI _ij＝Σ(4.6*0.6+4.4*0.7)/2

5) comprehensive trust value computing.In conjunction with direct trust value and indirect trust values, node i is to the direct trust value TD of node j _ij=4.5, indirect trust values TI _ij=2.92, weight factor α=0.7, β=0.3, then node i is to the comprehensive trust value TC of node j _ijcomputing formula be:

TC _ij＝0.7*4.5+0.3·2.92

6) comprehensive trust value is compared with trust threshold.Node i is to the comprehensive trust value TC of node j _ij=4.026, trust threshold is 3.5, and comprehensive trust value is greater than trust threshold, then node j is labeled as believable.

Fig. 3 is for trusting punishment block flow diagram, and a specific embodiment is as follows:

1) sequence node of the tape label generated in trust evaluation module is input as;

2) select node i, the mark of query node, is labeled as believable;

3) query node j is to the comprehensive trust value TC of node i _ji;

4) size of more comprehensive trust value and trust threshold, TC _ji=3.24, trust threshold is 3.5, and comprehensive trust value is less than trust threshold;

5) the punishment number of times CN of node j _jbe initialized as 0, add 1, CN now _j=1;

6) there is the node of historical trading also not inquire about with node i complete, except node j also has node k, therefore inquire about next node k;

7) query node k is to the comprehensive trust value TC of node i _ki;

8) size of more comprehensive trust value and trust threshold, TC _ki=4.31, trust threshold is 3.5, and comprehensive trust value is greater than threshold value;

9) this node is given up;

10) node of historical trading is had all to inquire about with node i complete;

11) the punishment number of times CN of query node j _j, CN _j=1;

12) punish node j, reduce the credit value of node j, RV _jfor initial credit value, RV _j=0.5, RV _j' be the new credit value after punishment, node punishment formula is:

RV _j′＝0.5*e ^-1

13) according to the credit value of other nodes to the trust value of node i and node i self, the final credit value of iterative computation node i, has node j and node k, RV with node i is related _j=0.184, TC _ji=3.24, RV _k=0.5, TC _ki=4.31, the final credit value of node i is:

RV _i ^(k+1)＝(0.184*3.24) ^k+(0.5*4.31) ^k

A kind of trust management model and method thereof based on trusting backstepping proposed by the invention, node is labeled as credible and insincere two classes by trust value, according to different classes of nodes ' behavior situation, adopt the thought of backward inference, from being evaluated node, the reverse evaluation node finding malice, and reduce the node credit value having malice tendency, thus ensure the safety of network.The method energy efficient identification malicious node that the present invention proposes, particularly has the camouflage node of deceptive practices.

Claims (4)

1. based on a trust management method for backward inference, it is characterized in that: comprise trust evaluation module and trust punishment module, the operating process of trust evaluation module comprises following concrete steps:

Step 1.1: after recording each two node transaction, node i is at t _xmoment to the satisfaction of node j,

tr _x(i，j)＝(sat _ij，t _x)，sat _ij∈[1，5]，i，j∈N

Wherein, sat _ijfor once mutual interior joint i is to node j satisfaction, N is the sum set of system interior joint i;

Step 1.2: in time interval t, computing node i is to the direct trust value of node j:

${\mathrm{TD}}_{\mathrm{ij}}=\underset{t_x\∈t}{\mathrm{\Σ}}\mathrm{tr}(i,j)/n,i,j\∈N$

Wherein, n is the transaction count that node i and node j produce;

Step 1.3: utilize cosine similarity to try to achieve and node i behavior similar node collection;

Step 1.4: computing node i is to the indirect trust values of node j:

${\mathrm{TI}}_{\mathrm{ij}}=\frac{{\mathrm{\Σ}}_{m\∈S}{\mathrm{TD}}_{\mathrm{mj}}\·{\mathrm{RV}}_m}{\left|S\right|},i,j\∈N$

Wherein, | S| is and node i behavior similar node collection, RV _ifor the credit value of node i;

Step 1.5: in conjunction with direct trust value and indirect trust values, weighted calculation node i is to the comprehensive trust value of node j:

TC _ij＝α·TD _ij+β·TI _ij

Wherein, weight factor alpha+beta=1;

Step 1.6: compared with trust threshold by comprehensive trust value, if comprehensive trust value is less than trust threshold, is labeled as incredible by node j, otherwise is labeled as believable by node j;

Step 1.7: repeat step 1.1 ~ step 1.7, generates the sequence node of tape label;

Inputed to by the sequence node of trust evaluation CMOS macro cell tape label and trust punishment module, the operating process of trusting punishment module comprises following concrete steps:

Step 2.1: the difference marked according to the sequence node interior joint of tape label, adopts corresponding backstepping mode, find and punish malicious node, obtains the final punishment number of times of malicious node;

Step 2.2: the credit value reducing malicious node according to the punishment number of times of malicious node;

Step 2.3: according to the comprehensive trust value of other nodes to present node and the credit value of present node self, calculates the final credit value of present node;

Step 2.4: repeat step 2.3, calculate the final credit value of all nodes, generates the sequence node pressing the sequence of credit value depreciation.

2. a kind of trust management method based on backward inference according to claim 1, is characterized in that: the difference that the described sequence node interior joint according to tape label marks, adopts corresponding backstepping mode, finds and punishes that the method for malicious node is:

Step a1: the node a in the sequence node of select tape mark, the mark of query node, if the mark of node a is believable, forwards step a2 to, otherwise forwards step a4 to;

Step a2: query node k comprehensive trust value to node a, the size of more comprehensive trust value and trust threshold, if be greater than threshold value, give up node k, forwards step a3 to; Otherwise the punishment number of times of node k is added 1;

Step a3: judge have the node of historical trading whether to inquire about with node a complete, if so, carry out step a6, otherwise another node k is next node, repeats step 2.2 ~ step 2.3, until travel through all nodes;

Step a4: query node k comprehensive trust value to node a, the size of more comprehensive trust value and trust threshold, if be less than threshold value, give up node k, forwards step a5 to, otherwise the punishment number of times of node k is added 1;

Step a5: judge have the node of historical trading whether to inquire about with node a complete, if so, carry out step a6, otherwise another node k is next node, repeats step a4 ~ step a5, until travel through all nodes;

Step a6: the final punishment number of times obtaining all nodes, punishment number of times be not 0 node be malicious node.

3. a kind of trust management method based on backward inference according to claim 1, is characterized in that: the method that the described punishment number of times according to malicious node reduces the credit value of malicious node is:

The punishment number of times of query node m, if punishment number of times is not 0, reduces the credit value of node m, obtains node m by the credit value after punishing:

${{\mathrm{RV}}_m}^{\′}={\mathrm{RV}}_m\·e^{-{\mathrm{CN}}_m}$

Wherein, CN _mfor the punishment number of times of node m.

4. a kind of trust management method based on backward inference according to claim 1, is characterized in that: described according to the credit value of other nodes to the comprehensive trust value of present node m and present node m self, the final credit value calculating present node m is:

RV _m ^(k+1)＝RV ₁TC _1m ^(k)+…+RV _jTC _jm ^(k)+…+RV _nTC _nm ^(k)，n，m，j∈N。