CN101415256A - Method of diagnosing wireless sensor network fault based on artificial immunity system - Google Patents

Abstract

The invention discloses a fault diagnosis method based on an artificial immune system in a wireless sensor network. The fault diagnosis is realized by collecting a characteristic data schema of the wireless sensor network and applying the autologous/ non-autologous principle of the artificial immune identification. The whole artificial immune system comprises four parts of an immune object, a database, immune detection and immune computing. The immune object consists of characteristic schema data. The data base is used for storing schema data in normal state and known fault state. The immune detection part completes the fault diagnosis and the immune computing part is used for realizing the cloning and the mutation of antibodies, optimizing the distributing space of the antibodies and distinguishing fault types. The method has the advantages of rapid abnormal recognition, good adaptability, dynamic balance and strong information processing capability and can effectively solve difficult diagnosis problems of numerous fault modes, coexistence of numerous fault modes, fault unpredictability, and the like in the wireless sensor network, thus improving the rapidness and the accuracy of the diagnosis.

Description

Wireless sensor network fault diagnosis method based on artificial immune system

Technical field

The present invention relates to a kind of wireless sensor network fault diagnosis method, belong to artificial immune system and wireless sensor network fault diagnostic field based on artificial immune system.

Background technology

Wireless sensor network (WSNs) by a large amount of freely distribute, have calculate and the sensor node of communication function by the communication for coordination of self-organizing mode to finish the intelligent network system of specific function, have highly reliable, easily dispose and advantage such as can expand, being widely used in fields such as national defense and military, safe anti-terrorism, environmental monitoring, traffic administration, health care and industrial production manufacturing, is 21 centurys four one of big new and high technologies.U.S. DARPA starts SensIT (Sensor Information Technology) plan, sets up a network system in order to physical quantitys such as monitoring optics, temperature, humidity; California, USA university uncle gram Intel laboratory, branch school and Atlantic Ocean institute have united in the Da Yadao deploy a multi-level sensor network system, are used for monitoring the life habit of petrel on the island; (structure healthy monitoring is to utilize sensor network to monitor the safe condition of building SHM) to the building condition monitoring; The JPL of U.S. NASA (Jet PropulsionLaboratory) laboratory development Sensor Webs carries out technology preparation preferably for the mars exploration technology.Studies show that more than wireless sensor network is one of emphasis direction of current international research, have tangible application potential.Yet because the wireless sensor network operational environment is comparatively complicated usually, uncertain factor is a lot, and sensor node is subjected to external interference easily and produces fault, influences operate as normal, and serious also can be accidents caused.NASA once yielded the space shuttle Launch Program, and its reason is exactly to find that a wireless sensor node that is installed on the space shuttle breaks down, and therefore, it is very necessary that wireless sensor network is carried out failure diagnosis.

From the nineties later stage, carried out the research in this field abroad.Marzullo has proposed data-centered sensing node fault detect thought first, yet because number of nodes is huge, back under attack data are easily lost, so it can not be applied in extensive sensing network environment; It is the multi-modal sensing node fault detection mechanism of cost with the consume significant energy that Farinaz has proposed a kind of, and the wireless sensing node energy constraint, this mechanism also is not suitable for radio sensing network; Recently, Krishnamachari etc. have proposed a kind of sensing node fault detection method based on shortest path spanning tree structure, and it is to come the decision node performance by the feature of judging bunch head, and is very strong to the leader cluster node dependence, and it is bigger to implement difficulty.In view of this, searching is a kind of accurately, real-time, method for diagnosing faults has crucial meaning for actual the applying of engineering of wireless sensor network reliably.

Summary of the invention

Technical problem: the technical problem to be solved in the present invention is to propose a kind of wireless sensor network fault diagnosis method based on artificial immune system at the defective that prior art exists.

Technical scheme: the present invention is based on the wireless sensor network fault diagnosis method of artificial immune system, comprise the steps:

A.) start wireless sensor network, initialization sensor node, adopt TinyOS to send acquisition;

B.) sensor node is carried out acquisition and the data and the consumption information of sensor node own of the monitoring target of gathering is fed back to leader cluster node;

C.) leader cluster node becomes packet with the data fusion of the described monitoring target of step B, and leader cluster node also sends to computer with packet and the consumption information of the described sensor node of step B own through the base station;

D.) adopt the size of monitoring target data value in the described packet of computer read step C and the packet size information that constitutes with the own consumption information of monitoring target data message, sensor node that reads and monitoring target data fusion makes up database, database comprises that two kinds of patterns are normal input pattern and fault input pattern, and the structure of database is as follows:

The sensor node number of wireless sensor network is N, the pattern count of normal input pattern is a, the pattern count of known fault input pattern is b, packet size during normal input pattern is that the data of re, monitoring target are rt for rp, sensor node energy consumption, packet during known fault input pattern size is that the data of fe, monitoring target are ft for fp, sensor node energy consumption, and then normal input pattern and known fault input pattern are expressed as respectively:

${\mathrm{IN}}_a={\left[\begin{array}{cccc}{\mathrm{rp}}_{11}& {\mathrm{rp}}_{12}& \&CenterDot;\&CenterDot;\&CenterDot;& {\mathrm{rp}}_{1N}\\ {\mathrm{re}}_{11}& {\mathrm{re}}_{12}& \&CenterDot;\&CenterDot;\&CenterDot;& {\mathrm{re}}_{1N}\\ {\mathrm{rt}}_{11}& {\mathrm{rt}}_{12}& \&CenterDot;\&CenterDot;\&CenterDot;& {\mathrm{rt}}_{1N}\\ \&CenterDot;& \&CenterDot;& \&CenterDot;& \&CenterDot;\\ \&CenterDot;& \&CenterDot;& \&CenterDot;& \&CenterDot;\\ \&CenterDot;& \&CenterDot;& \&CenterDot;& \&CenterDot;\\ {\mathrm{<figure-callout\; id="1"\; label="rp\; a"\; filenames="A200810236291E00131.png"\; state="\{\{state\}\}">rp</figure-callout>}}_a& {\mathrm{rp}}_{a2}& \&CenterDot;\&CenterDot;\&CenterDot;& {\mathrm{rp}}_{\mathrm{aN}}\\ {\mathrm{re}}_{a1}& {\mathrm{re}}_{a2}& \&CenterDot;\&CenterDot;\&CenterDot;& {\mathrm{re}}_{\mathrm{aN}}\\ {\mathrm{rt}}_{a1}& {\mathrm{rt}}_{a2}& \&CenterDot;\&CenterDot;\&CenterDot;& {\mathrm{rt}}_{\mathrm{aN}}\end{array}\right]}_{3a\&times;N},$ ${\mathrm{IN}}_b={\left[\begin{array}{cccc}{\mathrm{fp}}_{11}& {\mathrm{fp}}_{12}& \&CenterDot;\&CenterDot;\&CenterDot;& {\mathrm{fp}}_{1N}\\ {\mathrm{fe}}_{11}& {\mathrm{fe}}_{12}& \&CenterDot;\&CenterDot;\&CenterDot;& {\mathrm{fe}}_{1N}\\ {\mathrm{ft}}_{11}& {\mathrm{ft}}_{12}& \&CenterDot;\&CenterDot;\&CenterDot;& {\mathrm{ft}}_{1N}\\ \&CenterDot;& \&CenterDot;& \&CenterDot;& \&CenterDot;\\ \&CenterDot;& \&CenterDot;& \&CenterDot;& \&CenterDot;\\ \&CenterDot;& \&CenterDot;& \&CenterDot;& \&CenterDot;\\ {\mathrm{<figure-callout\; id="1"\; label="fp\; b"\; filenames="A200810236291E00131.png"\; state="\{\{state\}\}">fp</figure-callout>}}_b& {\mathrm{fp}}_{b2}& \&CenterDot;\&CenterDot;\&CenterDot;& {\mathrm{fp}}_{\mathrm{bN}}\\ {\mathrm{fe}}_{\mathrm{<figure-callout\; id="1"\; label="b"\; filenames="A200810236291E00131.png"\; state="\{\{state\}\}">b</figure-callout>}1}& {\mathrm{fe}}_{b2}& \&CenterDot;\&CenterDot;\&CenterDot;& {\mathrm{fe}}_{\mathrm{bN}}\\ {\mathrm{ft}}_{\mathrm{<figure-callout\; id="1"\; label="b"\; filenames="A200810236291E00131.png"\; state="\{\{state\}\}">b</figure-callout>}1}& {\mathrm{ft}}_{b2}& \&CenterDot;\&CenterDot;\&CenterDot;& {\mathrm{ft}}_{\mathrm{bN}}\end{array}\right]}_{3b\&times;N};$

E.) finish when the described database of step D makes up, then start artificial immune system, enter step F; Do not finish when the described database of step D makes up, then return steps A, continue the collection of data;

F.) artificial immune system carries out self check, then enters step G when artificial immune system is intact, otherwise, the report error message;

G.) artificial immune system starts after the fault detect that normal input pattern and known fault input pattern obtain unique characteristic vector through normalized and characterize input pattern as sample in the described database of extraction step D, promptly all characteristic vectors all in [0,1] ^NThe form space in, N representation space dimension;

H.) generate detector, comprise the steps:

I.) with the characteristic vector process fuzzy c mean cluster of the normal input pattern of the described sign of step G, obtaining the coordinate of cluster from the body center is SM _i, radius is SR _iThe coordinate of initial setting detector centre is DM _j, radius is DR _j, then the detector of initial setting and the nearest distance of cluster between body are: $d_{\mathrm{ij}}=\underset{i=1,j=1}{\overset{i=P,j=Q}{\min }}\left|\right|{\mathrm{DM}}_j-{\mathrm{SM}}_i\left|\right|,$ SM _iRepresent the coordinate of i cluster, SR from the body center _iRepresent i cluster from the body radius, the span that i is is 1 to P, and P represents that cluster generates from body classification number; DM _jThe coordinate of representing j detector centre, DR _jRepresent j detector radius, the span of j is 1 to Q, and Q represents the quantity of detector; The initial setting detector is passed through Adaptive Genetic: $F{({\mathrm{SR}}_i,d_{\mathrm{ij}})}_{\max }=\frac{d_{\mathrm{ij}}-{\mathrm{SR}}_i}{{\mathrm{SR}}_i}$ Obtain optimum detector;

Ii.) optimum detector that generates is joined cluster in body;

Iii.) finish when Q optimum detector generates, then enter step J; Do not finish when Q optimum detector generates, then return step I;

I.) characteristic vector with the described sign known fault of step G input pattern obtains memory antibody through the fuzzy c mean cluster;

J.) input fault is carried out after the normalization as antigen, the described detector of step H is an antibody, is the center with antigen, and antibody is divided into different distributed areas, according to the affinity Aff=(1-d between antibody and antigen _EC) ³Size carry out immunity and calculate, the antagonist processing of cloning, make a variation, d _ECBe the Euclidean distance between antigen and antibody;

1. as 0≤Aff＜R1, then antibody is not done any change, and wherein R1 is the minimum affinity Aff of antigen and antibody _Min

2. as R1≤Aff＜R2, then generate new antibodies to substitute original antibody:

$B_{n+1}=\frac{\underset{j=1}{\overset{M_B}{\mathrm{\&Sigma;}}}{B}_{\mathrm{nj}}+\mathrm{Ag}}{M_B+1},$ B wherein _NjBe current antibody, B _N+1Be new antibodies, M _BAntibody number in regional for this reason, Ag is an antigen, $R2=\frac{R3}{2},$ R3 is the mean value of affinity between antigen and all antibody $R3=\frac{\underset{j}{\overset{Q}{\mathrm{\&Sigma;}}}{\mathrm{Aff}}_j}{Q};$

3. as R2≤Aff＜R ₃, then antibody is carried out as lower variation:

B _N+1=B _n+ μ (Ag-B _n), wherein, B _nBe current antibody, B _N+1Be the antibody after the variation, Ag is an antigen, and μ is the coefficient of variation, and span is 0≤μ≤1, and R3 is the mean value of affinity between antigen and all antibody $R3=\frac{\underset{j=1}{\overset{Q}{\mathrm{\&Sigma;}}}{\mathrm{Aff}}_j}{Q},$ Q is the antibody number;

4. as Aff 〉=R ₃, then antigen is mated with normal input pattern and known fault pattern respectively:

When antigen and normal input pattern coupling, then antigen is from bulk-mode, returns step J new antigen is carried out diagnosis;

When antigen and known fault pattern matching, then antigen is the known fault pattern, and detector remains unchanged, and returns step J new antigen is carried out diagnosis;

When antigen and normal input pattern and known fault pattern all do not match, then this antigen is unknown failure, and a random site is cloned a feature antibody among region R 1≤Aff＜R2, and at regional Aff 〉=R ₃Middle antibody of picked at random is deleted it, enters step K;

K.) the feature antibody that step J is generated marks and adds memory antibody, returns step J new antigen is carried out diagnosis.

Beneficial effect: do not need priori when the present invention carries out failure diagnosis, be not subjected to the restriction of fault mode quantity, show good adaptive, can be widely used in the wireless sensor network under the complex conditions such as node random distribution, operational environment be unpredictable.

Description of drawings

Fig. 1: flow chart of the present invention;

Fig. 2: detector product process figure of the present invention;

Fig. 3: wireless sensor network schematic diagram.

Embodiment

As shown in Figure 1.A kind of wireless sensor network fault diagnosis method based on artificial immune system comprises the steps:

A.) startup wireless sensor network, initialization sensor node, employing TinyOS programme to sensor node, send acquisition;

B.) sensor node is carried out acquisition and the data and the consumption information of sensor node own of the monitoring target of gathering is fed back to leader cluster node;

C.) leader cluster node becomes packet with the data fusion of the described monitoring target of step B, and leader cluster node also sends to computer with packet and the consumption information of the described sensor node of step B own through the base station;

D.) computer reads the size of monitoring target data value in the described packet of C, the packet size information that constitutes with the own consumption information of monitoring target data message, sensor node that reads and monitoring target data fusion makes up database, database comprises that two kinds of patterns are normal input pattern and fault input pattern, and the structure of database is as follows:

The sensor node number of wireless sensor network is N, the pattern count of normal input pattern is a, the pattern count of known fault input pattern is b, packet size during normal input pattern is that the data of re, monitoring target are rt for rp, sensor node energy consumption, packet during known fault input pattern size is that the data of fe, monitoring target are ft for fp, sensor node energy consumption, and then normal input pattern and known fault input pattern are expressed as respectively:

${\mathrm{IN}}_a={\left[\begin{array}{cccc}{\mathrm{rp}}_{11}& {\mathrm{rp}}_{12}& \&CenterDot;\&CenterDot;\&CenterDot;& {\mathrm{rp}}_{1N}\\ {\mathrm{re}}_{11}& {\mathrm{re}}_{12}& \&CenterDot;\&CenterDot;\&CenterDot;& {\mathrm{re}}_{1N}\\ {\mathrm{rt}}_{11}& {\mathrm{rt}}_{12}& \&CenterDot;\&CenterDot;\&CenterDot;& {\mathrm{rt}}_{1N}\\ \&CenterDot;& \&CenterDot;& \&CenterDot;& \&CenterDot;\\ \&CenterDot;& \&CenterDot;& \&CenterDot;& \&CenterDot;\\ \&CenterDot;& \&CenterDot;& \&CenterDot;& \&CenterDot;\\ {\mathrm{<figure-callout\; id="1"\; label="rp\; a"\; filenames="A200810236291E00131.png"\; state="\{\{state\}\}">rp</figure-callout>}}_a& {\mathrm{rp}}_{a2}& \&CenterDot;\&CenterDot;\&CenterDot;& {\mathrm{rp}}_{\mathrm{aN}}\\ {\mathrm{re}}_{a1}& {\mathrm{re}}_{a2}& \&CenterDot;\&CenterDot;\&CenterDot;& {\mathrm{re}}_{\mathrm{aN}}\\ {\mathrm{rt}}_{a1}& {\mathrm{rt}}_{a2}& \&CenterDot;\&CenterDot;\&CenterDot;& {\mathrm{rt}}_{\mathrm{aN}}\end{array}\right]}_{3a\&times;N},$ ${\mathrm{IN}}_b={\left[\begin{array}{cccc}{\mathrm{fp}}_{11}& {\mathrm{fp}}_{12}& \&CenterDot;\&CenterDot;\&CenterDot;& {\mathrm{fp}}_{1N}\\ {\mathrm{fe}}_{11}& {\mathrm{fe}}_{12}& \&CenterDot;\&CenterDot;\&CenterDot;& {\mathrm{fe}}_{1N}\\ {\mathrm{ft}}_{11}& {\mathrm{ft}}_{12}& \&CenterDot;\&CenterDot;\&CenterDot;& {\mathrm{ft}}_{1N}\\ \&CenterDot;& \&CenterDot;& \&CenterDot;& \&CenterDot;\\ \&CenterDot;& \&CenterDot;& \&CenterDot;& \&CenterDot;\\ \&CenterDot;& \&CenterDot;& \&CenterDot;& \&CenterDot;\\ {\mathrm{<figure-callout\; id="1"\; label="fp\; b"\; filenames="A200810236291E00131.png"\; state="\{\{state\}\}">fp</figure-callout>}}_b& {\mathrm{fp}}_{b2}& \&CenterDot;\&CenterDot;\&CenterDot;& {\mathrm{fp}}_{\mathrm{bN}}\\ {\mathrm{fe}}_{\mathrm{<figure-callout\; id="1"\; label="b"\; filenames="A200810236291E00131.png"\; state="\{\{state\}\}">b</figure-callout>}1}& {\mathrm{fe}}_{b2}& \&CenterDot;\&CenterDot;\&CenterDot;& {\mathrm{fe}}_{\mathrm{bN}}\\ {\mathrm{ft}}_{\mathrm{<figure-callout\; id="1"\; label="b"\; filenames="A200810236291E00131.png"\; state="\{\{state\}\}">b</figure-callout>}1}& {\mathrm{ft}}_{b2}& \&CenterDot;\&CenterDot;\&CenterDot;& {\mathrm{ft}}_{\mathrm{bN}}\end{array}\right]}_{3b\&times;N};$

E.) finish when the described database of step D makes up, then start artificial immune system, enter step F; Do not finish when the described database of step D makes up, then return steps A, continue the collection of data;

F.) artificial immune system carries out self check, then enters step G when artificial immune system is intact; Otherwise, the report error message;

G.) artificial immune system starts after the fault detect that normal input pattern and known fault input pattern obtain unique characteristic vector through normalized and characterize input pattern as sample in the described database of extraction step D, promptly all characteristic vectors all in [0,1] ^NThe form space in, N representation space dimension;

H.) as shown in Figure 2, generate detector, comprise the steps:

I.) with the characteristic vector process fuzzy c mean cluster of the normal input pattern of the described sign of step G, the coordinate of acquisition cluster from the body center is ${\mathrm{SM}}_i=\frac{\underset{v=1}{\overset{a}{\mathrm{\&Sigma;}}}{\left(u_{\mathrm{iv}}\right)}^m{\mathrm{IN}}_v}{\underset{v=1}{\overset{a}{\mathrm{\&Sigma;}}}{\left(u_{\mathrm{iv}}\right)}^m},$ u _IvRepresent that i normal input pattern belongs to the degree of membership at v class center, and satisfy following constraints: $\underset{i=1}{\overset{P}{\mathrm{\&Sigma;}}}{u}_{\mathrm{iv}}=1,$ 0≤u _Iv≤ 1, IN _vBe normal input pattern, 1≤v≤a, m ∈ [1, ∞) be Weighted Index, radius is SR _iThe coordinate of initial setting detector centre is DM _j, radius is DR _j, then initial setting device and the nearest Euclidean distance of cluster between body are $d_{\mathrm{ij}}=\underset{i=1,j=1}{\overset{i=P,j=Q}{\min }}\left|\right|{\mathrm{DM}}_j-{\mathrm{SM}}_i\left|\right|,$ SM _iRepresent the coordinate of i cluster, SR from the body center _iRepresent i cluster from the body radius, the span that i is is 1 to P, and P represents that cluster generates cluster from body classification number; DM _jThe coordinate of representing j detector centre, DR _jRepresent j detector radius, the span of j is 1 to Q, and Q represents the quantity of detector; The initial setting detector is passed through Adaptive Genetic: $F{({\mathrm{SR}}_i,d_{\mathrm{ij}})}_{\max }=\frac{d_{\mathrm{ij}}-{\mathrm{SR}}_i}{{\mathrm{SR}}_i}$ Obtain optimum detector;

Ii.) optimum detector that generates is joined cluster in body;

Iii.) finish when Q optimum detector generates, then enter step J, the detector that obtain this moment has maximum radius R _i, and with from the minimum d of the distance of body _IjminDo not finish when Q optimum detector generates, then return step I;

I.) characteristic vector with the described sign known fault of step G input pattern obtains memory antibody through the fuzzy c mean cluster;

J.) input fault is carried out normalization and obtain fault vectors, as antigen, the described detector of step H is an antibody with fault vectors, is the center with antigen, and antibody is divided into different distributed areas, according to the affinity Aff=(1-d between antibody and antigen _EC) ³Size carry out immunity and calculate, the antagonist processing of cloning, make a variation, d _ECBe the Euclidean distance between antigen and antibody;

1. as 0≤Aff＜R1, then antibody is not done any change, and wherein R1 is the minimum affinity Aff of antigen and antibody _Min

2. as R1≤Aff＜R2, then generate new antibodies to substitute original antibody:

$B_{n+1}=\frac{\underset{j=1}{\overset{M_B}{\mathrm{\&Sigma;}}}{B}_{\mathrm{nj}}+\mathrm{Ag}}{M_B+1},$ B wherein _NjBe current antibody, B _N+1Be new antibodies, M _BAntibody number in regional for this reason, Ag is an antigen, $R2=\frac{R3}{2},$ R3 is the average of affinity between antigen and all antibody $R3=\frac{\underset{j}{\overset{Q}{\mathrm{\&Sigma;}}}{\mathrm{Aff}}_j}{Q},$ Q is the antibody number;

3. as R2≤Aff＜R ₃, then antibody is carried out as lower variation:

B _N+1=B _n+ μ (Ag-B _n), wherein, B _nBe current antibody, B _N+1Be the antibody after the variation, Ag is an antigen, and μ is the coefficient of variation, and span is 0≤μ≤1, and R3 is the average of affinity between antigen and all antibody $R3=\frac{\underset{j=1}{\overset{Q}{\mathrm{\&Sigma;}}}{\mathrm{Aff}}_j}{Q},$ Q is the antibody number;

4. as Aff 〉=R ₃, then antigen is mated with normal input pattern and known fault pattern respectively:

When antigen and normal input pattern coupling, then antigen is from bulk-mode, returns step J new antigen is carried out diagnosis;

When antigen and known fault pattern matching, then antigen is the known fault pattern, and detector remains unchanged, and returns step J new antigen is carried out diagnosis;

When antigen and normal input pattern and known fault pattern all do not match, then this antigen is unknown failure, and a random site is cloned a feature antibody among region R 1≤Aff＜R2, and at regional Aff 〉=R ₃Middle antibody of picked at random is deleted it, enters step K;

K.) the feature antibody that step J is generated marks and adds memory antibody, returns step J new antigen is carried out diagnosis.

Fig. 3 is a wireless sensor network system structured flowchart of the present invention, comprises monitoring target, sensor node, leader cluster node, base station and computer.Wireless sensor platform adopts the CUTE radio sensing network platform of smart material and structure science and technology of aviation key lab of Nanjing Aero-Space University development, and its radio frequency is 2.4GHZ.Sensor node is used to gather the data of monitoring target and sends to leader cluster node, and leader cluster node carries out information fusion with the information that receives and delivers to the base station.Each leader cluster node is managed several sensing nodes, utilizes TinyOS that sensor node is programmed, and sends acquisition, and node is received the collection of beginning object data after the order, and data are received through bunch head and base station is sent to computer with the form of wrapping after handling.Each node adopts No. 5 powered battery of four joints, obtains the situation of change of energy consumption by the node power consumption enquiry module.Whether characterize the work of wireless sensor network state with the size of monitoring target, packet and energy consumption as characteristic vector normal, above-mentioned three physical quantitys that obtain under normal condition and known fault conditions as fault diagnosis model, are stored in respectively in bulk-mode database and known fault pattern database and are used for failure diagnosis.

Comprise immune object, immune detection, immunity calculating and database four major parts.By hardware interface and software program, obtain three characteristic vectors that characterize the wireless sensor network state: the size of the packet of node energy consumption, transmission and monitoring target value, with these characteristic vector compositional model data as immune object; By to preliminary treatment such as immune object normalization, yojan clusters, utilize self-adapted genetic algorithm to generate detector, optimize the detector distribution space; With the input fault is antigen, and detector is an antibody, and according to the evolutionary learning mechanism of artificial immunity, the antagonism proterotype is carried out learning and memory; According to the division of antibody regions, antigen to be carried out immunity calculate, antibody produces clonal expansion and variation, according to clone, variation situation and the known fault information of antibody, obtains the area distribution of antigen correspondence on state space, realizes the judgement of input fault type; Newly-generated antibody is added in the memory antibody database, when have identical or similar fault to occur next time, can produce secondary immunity response fast, thereby shorten Diagnostic Time greatly.

Claims (1)

1. the wireless sensor network fault diagnosis method based on artificial immune system is characterized in that this method comprises the steps:

A.) start wireless sensor network, initialization sensor node, adopt TinyOS to send acquisition;

B.) sensor node is carried out acquisition and the data and the consumption information of sensor node own of the monitoring target of gathering is fed back to leader cluster node;

C.) leader cluster node becomes packet with the data fusion of the described monitoring target of step B, and leader cluster node also sends to computer with packet and the consumption information of the described sensor node of step B own through the base station;

D.) adopt the size of monitoring target data value in the described packet of computer read step C and the packet size information that constitutes with the own consumption information of monitoring target data message, sensor node that reads and monitoring target data fusion makes up database, database comprises that two kinds of patterns are normal input pattern and fault input pattern, and the structure of database is as follows:

The sensor node number of wireless sensor network is N, the pattern count of normal input pattern is a, the pattern count of known fault input pattern is b, packet size during normal input pattern is that the data of re, monitoring target are rt for rp, sensor node energy consumption, packet during known fault input pattern size is that the data of fe, monitoring target are ft for fp, sensor node energy consumption, and then normal input pattern and known fault input pattern are expressed as respectively:

${\mathrm{IN}}_a={\left[\begin{array}{cccc}{\mathrm{rp}}_{11}& {\mathrm{rp}}_{12}& \&CenterDot;\&CenterDot;\&CenterDot;& {\mathrm{rp}}_{1N}\\ {\mathrm{re}}_{11}& {\mathrm{re}}_{12}& \&CenterDot;\&CenterDot;\&CenterDot;& {\mathrm{re}}_{1N}\\ {\mathrm{rt}}_{11}& {\mathrm{rt}}_{12}& \&CenterDot;\&CenterDot;\&CenterDot;& {\mathrm{rt}}_{1N}\\ \&CenterDot;& \&CenterDot;& \&CenterDot;& \&CenterDot;\\ \&CenterDot;& \&CenterDot;& \&CenterDot;& \&CenterDot;\\ \&CenterDot;& \&CenterDot;& \&CenterDot;& \&CenterDot;\\ {\mathrm{rp}}_{a1}& {\mathrm{rp}}_{a2}& \&CenterDot;\&CenterDot;\&CenterDot;& {\mathrm{rp}}_{\mathrm{aN}}\\ {\mathrm{re}}_{a1}& {\mathrm{re}}_{a2}& \&CenterDot;\&CenterDot;\&CenterDot;& {\mathrm{re}}_{\mathrm{aN}}\\ {\mathrm{rt}}_{a1}& {\mathrm{rt}}_{a2}& \&CenterDot;\&CenterDot;\&CenterDot;& {\mathrm{rt}}_{\mathrm{aN}}\end{array}\right]}_{3a\&times;N},$ ${\mathrm{IN}}_b={\left[\begin{array}{cccc}{\mathrm{fp}}_{11}& {\mathrm{fp}}_{12}& \&CenterDot;\&CenterDot;\&CenterDot;& {\mathrm{fp}}_{1N}\\ {\mathrm{fe}}_{11}& {\mathrm{fe}}_{12}& \&CenterDot;\&CenterDot;\&CenterDot;& {\mathrm{fe}}_{1N}\\ {\mathrm{ft}}_{11}& {\mathrm{ft}}_{12}& \&CenterDot;\&CenterDot;\&CenterDot;& {\mathrm{ft}}_{1N}\\ \&CenterDot;& \&CenterDot;& \&CenterDot;& \&CenterDot;\\ \&CenterDot;& \&CenterDot;& \&CenterDot;& \&CenterDot;\\ \&CenterDot;& \&CenterDot;& \&CenterDot;& \&CenterDot;\\ {\mathrm{fp}}_{b1}& {\mathrm{fp}}_{b2}& \&CenterDot;\&CenterDot;\&CenterDot;& {\mathrm{fp}}_{\mathrm{bN}}\\ {\mathrm{fe}}_{b1}& {\mathrm{fe}}_{b2}& \&CenterDot;\&CenterDot;\&CenterDot;& {\mathrm{fe}}_{\mathrm{bN}}\\ {\mathrm{ft}}_{b1}& {\mathrm{ft}}_{b2}& \&CenterDot;\&CenterDot;\&CenterDot;& {\mathrm{ft}}_{\mathrm{bN}}\end{array}\right]}_{3b\&times;N};$

E.) finish when the described database of step D makes up, then start artificial immune system, enter step F; Do not finish when the described database of step D makes up, then return steps A, continue the collection of data;

F.) artificial immune system carries out self check, then enters step G when artificial immune system is intact; When artificial immune system is not intact, then report error message;

G.) artificial immune system starts after the fault detect that normal input pattern and known fault input pattern obtain unique characteristic vector through normalized and characterize input pattern as sample in the described database of extraction step D, promptly all characteristic vectors all in [0,1] ^NThe form space in, N representation space dimension;

H.) generate detector, comprise the steps:

I.) characteristic vector with the normal input pattern of the described sign of step G is SM through the coordinate of fuzzy c mean cluster acquisition cluster from the body center _i, radius is SR _iThe coordinate of initial setting detector centre is DM _j, radius is DR _j, then the detector of initial setting and the nearest distance of cluster between body are: $d_{\mathrm{ij}}=\underset{i=1,j=1}{\overset{i=P,j=Q}{\min }}\left|\right|{\mathrm{DM}}_j-{\mathrm{SM}}_i\left|\right|,$ SM _iRepresent the coordinate of i cluster, SR from the body center _iRepresent i cluster from the body radius, the span that i is is 1 to P, and P represents that cluster generates from body classification number; DM _jThe coordinate of representing j detector centre, DR _jRepresent j detector radius, the span of j is 1 to Q, and Q represents the quantity of detector; The initial setting detector is passed through Adaptive Genetic: $F{({\mathrm{SR}}_i,d_{\mathrm{ij}})}_{\max }=\frac{d_{\mathrm{ij}}-{\mathrm{SR}}_i}{{\mathrm{SR}}_i}$ Obtain optimum detector;

Ii.) optimum detector that generates is joined cluster in body;

Iii.) finish when Q optimum detector generates, then enter step J; Do not finish when Q optimum detector generates, then return step I;

I.) characteristic vector with the described sign known fault of step G input pattern obtains memory antibody through the fuzzy c mean cluster;

J.) input fault is carried out after the normalization as antigen, the described detector of step H is an antibody, is the center with antigen, and antibody is divided into different distributed areas, according to the affinity Aff=(1-d between antibody and antigen _EC) ³Size carry out immunity and calculate, the antagonist processing of cloning, make a variation, d _ECBe the Euclidean distance between antigen and antibody;

1. as 0≤Aff＜R1, then antibody is not done any change, and wherein R1 is the minimum affinity Aff of antigen and antibody _Min

2. as R1≤Aff＜R2, then generate new antibodies to substitute original antibody:

$B_{n+1}=\frac{\underset{j=1}{\overset{M_B}{\mathrm{\&Sigma;}}}{B}_{\mathrm{nj}}+\mathrm{Ag}}{M_B+1},$ B wherein _NjBe current antibody, B _N+1Be new antibodies, M _BAntibody number in regional for this reason, Ag is an antigen, $R2=\frac{R3}{2},$ R3 is the mean value of affinity between antigen and all antibody $R3=\frac{\underset{j=1}{\overset{Q}{\mathrm{\&Sigma;}}}{\mathrm{Aff}}_j}{Q};$

3. as R2≤Aff＜R ₃, then antibody is carried out as lower variation:

B _N+1=B _n+ μ (Ag-B _n), wherein, B _nBe current antibody, B _N+1Be the antibody after the variation, Ag is an antigen, and μ is the coefficient of variation, and span is 0≤μ≤1, and R3 is the mean value of affinity between antigen and all antibody $R3=\frac{\underset{j=1}{\overset{Q}{\mathrm{\&Sigma;}}}{\mathrm{Aff}}_j}{Q};$ Q is the antibody number;

4. as Aff 〉=R ₃, then antigen is mated with normal input pattern and known fault pattern respectively:

When antigen and normal input pattern coupling, then antigen is from bulk-mode, returns step J new antigen is carried out diagnosis;

When antigen and known fault pattern matching, then antigen is the known fault pattern, and detector remains unchanged, and returns step J new antigen is carried out diagnosis;

When antigen and normal input pattern and known fault pattern all do not match, then this antigen is unknown failure, and a random site is cloned a feature antibody among region R 1≤Aff＜R2, and at regional Aff 〉=R ₃Middle antibody of picked at random is deleted it, enters step K;

K.) the feature antibody that step J is generated marks and adds memory antibody, returns step J new antigen is carried out diagnosis.

Images