CN105636094A - Wireless sensor network early warning method and system based on clustering compressed sensing - Google Patents

Abstract

The invention discloses a wireless sensor network early warning method and system based on clustering compressed sensing. The method comprises the following steps: S1) calculating optimal cluster number and optimum cluster head space distribution by a gateway according to geographic position and an energy consumption model of each sensor node in a wireless sensor network; S2) obtaining monitoring data in a cluster, and converting the monitoring data into binary reading and a binary bit string sequence according to a threshold value; S3) carrying out dense random projection and sparse random projection thereon to obtain a compressed sensing sequence, and carrying out reconstruction on the compressed sensing sequence to obtain a reconstructed sequence; S4) carrying out judgment according to estimated value of the reconstructed sequence, and summarizing the number of effective neighbor nodes executing a majority vote method; and S5) if the number of the effective neighbor nodes is larger than a preset threshold value, determining that the node has an abnormal event. Real-time performance and fault tolerance of the monitoring method can be improved; influence of the fault nodes on abnormal event detection reliability is reduced; and network energy consumption in the data collection process can be reduced.

Description

Based on wireless sense network method for early warning and the system of sub-clustering compressed sensing

Technical field

The present invention relates to technology of Internet of things field, particularly relate to a kind of wireless sense network method for early warning based on sub-clustering compressed sensing and system.

Background technology

Wireless sensor network is considered as one of important component part of thing networking; wireless sense network, as the core technology of the emerging strategic industries thing networking of " perception China ", achieves practical application on a small scale in fields such as geologic hazard monitoring, environment protection, Industry Control, car networking, intelligence households. But, in event monitoring type wireless sense network application system, sensor node all stops in " hundred " this magnitude, and the time of Network Survivability is so short that to can't see advantage especially. The wireless sense network that we once imagined is all have thousands of nodes, and the meaning that wireless sense network exists is large-scale data collection, and environment is carried out continuous monitor in real time. Therefore, the extensive long-term wireless sense network event monitoring technology disposed faces renewal, higher requirement:

1) energy saving. Present most of sensor node such as CC2530 powers by 1 piece of similar cellular li-ion battery. The maximum technology barrier of wireless sense network application is that node has very limited margin of energy, and the life-span of node battery and network depends on the energy consumption of node. Between node, radio communication accounts for the 70% of total energy consumption, and the energy consumption of node perceived and calculating is less by contrast. When battery reserves technology is made slow progress, urgent needs develops more efficient data compression technique, farthest eliminate the redundancy between data, like this while ensureing significantly to fall low amount of transmitting data, still can keep the event perception performance of high precision, be most important to extending the network survivability cycle as far as possible.

2) real-time. Most of monitoring and measuring application inherently requires that wireless sense network reports exception or early warning in time. Real-time criterion is that network is to the time of response of event. In the resource-constrained large-scale wireless sensing net with energy starvation, will meeting the abnormal information of accurate rapid extraction from the dimension flow data of magnanimity, compared with traditional distributed database, its Real-time ensuring technology is more challenging.

3) reliability. Wireless sense network is formed by thousands of low side sensor node self-organization, and the power supply energy of node, calculating, storage capacity and communication capacity are all limited, and the various faults in neighbourhood noise, node and network easily cause monitoring data abnormal. Therefore, wireless sense network early warning technology to be had fault-tolerance, ensures lower rate of false alarm and fails to report rate.

In a word, wireless sense network Monitoring techniques must have three character: energy saving, real-time and reliability. Wherein real-time and reliability are two primary metric of monitoring performance. Owing to the portable power ability of node is limited, event monitoring performance and sensing net work life cycle is made to become two conflicting optimizing index, for the problems referred to above, the present invention provides a kind of wireless sense network monitoring method being applicable to extensive long-term deployment, the energy saving of complex optimum monitoring method, real-time and reliability preferably.

Existing WSN generally adopts direct mode to carry out event detecting method (referred to as DR method), and DR method that is one is when there being the monitoring value of node to exceed threshold value, and WSN will have abnormal event to occur to user report. The advantage time of response of the method fast (time of response is referred to as RT), postponing little, promptness is good; Its shortcoming is that energy expenditure is big and rate of false alarm is higher. Consider economic feasibility, WSN is formed by a large amount of low side sensor node self-organizations, there is limited hardware and software resource in node, generally also it is deployed in severe environment unmanned on duty, malfunctioning node is likely when occurring without event, its monitoring value also can exceed event threshold, thus has abnormal event to occur to user's wrong report.

Therefore, in WSN, fault-tolerant Monitoring techniques is subject to the attention of investigator always, the majority vote method (referred to as TV method) of the distributed Bayes proposed the earliest is a kind of fault-tolerant method detecting abnormal event, TV method make use of abnormal event in essence and occurs in and spatially have dependency, and malfunctioning node spatially not have dependency be random generation. Emulation result be displayed in node failure rate lower time, the accuracy rate of the abnormal event of the method detection is higher, and rate of false alarm is low, and fault-tolerance is better; But, the shortcoming of TV method is that the time of response is long, postpones big, and real-time is poor. Because this kind of method make use of node monitoring value, to have spatial coherence fault-tolerant to realize, although the reliability of improve, but the inevitable time of response extending the abnormal event of detection. Owing to sensing network generally adopts powered battery, existing TV method and a series of improving one's methods thereof all need neighbor node communication to cooperate, in DR method and TV method, node in event generation area also exists and repeats to report to the police in addition, this all can increase network communication energy consumption expense, reduces wireless sense network work life cycle.

At present, also there are a lot of technology barriers in extensive wireless sense network event monitoring of disposing, and is the important factor hindering sensor network to affix one's name on a large scale as there is a small amount of bottleneck node in network energy consumption and network. But sensor network also has the feature of self, the data as gathered have very strong spatio-temporal redundancies usually, utilize compression technology in net can reduce the data packet of transmission in network, dispose scale such that it is able to reduce network energy consumption and expand network. In tradition net, data compression technique does not have clear and definite compression method or needs a large amount of calculating or the communication resource, and they are all not too applicable to the resource-constrained feature of wireless sensor network. In recent years, along with the proposition that compressed sensing (CompressiveSensing, referred to as CS) is theoretical, collect for wireless sensor network data and warded off a new research road. Therefore, in conjunction with WSN relevant feature, excavate the data dependence between node by compression sensing method, eliminate the redundancy between data to greatest extent, while ensureing significantly to fall low amount of transmitting data, still can keep the event monitoring performance of high precision.

In the process of WSN event monitoring, if the interstitial content participating in single observed value too much can cause problem below: 1. single transmitting measured values cost is excessive causes the raising of whole network performance limited; 2. owing to wireless sensor network is network of easily makeing mistakes, therefore the jumping figure participating in the collection of single observed value is more many more easily makes mistakes. 1. bunch the significant challenge of clustering routing event monitoring has two aspects: in, how bunch head distributes; In bunch, transmission cost is optimum; 2. having how many bunches in network, event monitoring cost is optimum; Owing in tradition net, data compression method is relevant to route strategy, this just causes being difficult to set up unified energy consumption model in the process of clustering routing event monitoring, so the calculating of optimum bunch also becomes very difficult.

Summary of the invention

The technical problem to be solved in the present invention is for two class DR in prior art and TV method and adopts tree path policy to carry out the defect that in the transmitting procedure of data compression perception, energy expenditure is big, one is provided can effectively to reduce energy expenditure, and Detection accuracy height, the time of response fast wireless sense network method for early warning based on sub-clustering compressed sensing and system.

The technical solution adopted for the present invention to solve the technical problems is:

The present invention provides a kind of wireless sense network method for early warning based on sub-clustering compressed sensing, comprises the following steps:

S1, gateway according to the geographical position of each sensor node in radio sensing network and energy consumption model, calculate the member node in optimum bunch number, best bunch head spatial distribution, each bunch, each bunch set up shortest distance spanning tree between head and gateway;

S2, obtain in this bunch the monitoring data of i-th node in tSetup times detection alarm threshold value ��_e, and according to this threshold value, monitoring data is converted to scale-of-two readingObtain in this bunch i-th node to the monitoring data in p moment of the time slide window storage before t, and be converted to two value bit string sequences according to its normal range �� that fluctuates

S3, obtain this bunch bunch head to bunch in two value sequence X of sensor node monitoring value and bit string sequenceRespectively it is carried out dense accidental projection and sparse accidental projection, obtains compressed sensing sequenceWithAnd by sequenceWithIt is transferred to gateway to be reconstructed, obtains reproducing sequenceWith

S4, gateway judge according to the estimated value of in reproducing sequence i-th node, if estimated valueThen initiated event detects and adds up the effective neighbor node number performing majority vote methodIf the estimated value of the neighbor node j of i-th nodeThen effective neighbor node numberAdd 1;

If estimated valueThen by i-th nodeWith its neighbor node j'sStep-by-step carries out logic and operation, if operation result withIdentical and have non-zero value, then effective neighbor node numberAdd 1;

I-th node that initiated event is detected by S5, gateway performs majority vote method, if effectively neighbor node number is greater than the threshold value of setting, then judges that this node has abnormal event to occur.

Further, the step S1 of the present invention calculates optimum bunch number h_optCalculation formula be:

$h_{opt}=\sqrt{\frac{{\mathrm{\ϵ}}_{fs}\×N\×A\×S}{12E_{elec}}}$

Wherein, N represents the sensor node sum that wireless sense network is disposed, and A represents wireless sense network monitored area area, and S represents the sparse projection matrix �� of compressed sensing^rSparse degree, E_elecRepresent the energy sending or receiving a bit and consume, ��_fsRepresent transmission amplifying power.

Further, formula monitoring data being converted to scale-of-two reading in the step S2 of the present invention is:

Wherein,For in this bunch, i-th node obtains monitoring data in t, ��_eFor event detection alarm threshold value.

Further, formula monitoring data being converted into two value bit string sequences in the step S2 of the present invention is:

Wherein, i-th node to storing the monitoring value in nearest p moment at time slide window before t is�� is switching threshold,It is the differential conversion bit of i-th node in t.

Further, the step S3 of the present invention adopt dense accidental projection and sparse accidental projection to obtain compressed sensing sequenceWithCalculation formula be:

$\tilde{X}={\mathrm{\φ}}^{e}X$

${\tilde{Y}}_i^t={\mathrm{\φ}}^r{Y}_i^t$

Wherein, dense accidental projection matrix ��^eL column element production method be:

Wherein, each sensor node produces m random elementThe value of m is by perception data in networkSparse degree determine;

Sparse accidental projection matrix ��^rL column element production method be:

Wherein, each sensor node producesIndividual random element Value by perception data in networkSparse degree threshold value S=log (n)/(2n) determine.

Further, the reconstructing method in the step S3 of the present invention is specially:

Gateway is to sequenceEmploying orthogonal matching pursuit algorithm is reconstructed, and obtains sequence

To sequenceAdopt the Bayes based on local time's dependency to learn joint sparse reconstruct algorithm to be reconstructed, obtain sequence

Further, the threshold value arranged in the step S5 of the present invention is 0.5 Represent the neighbor node number in a jumping communication range of i-th node.

Further, the determination methods in the step S5 of the present invention is specially:

Its effective neighbor node number of i-th node calculate that initiated event is detected by gateway Represent the neighbor node number in a jumping communication range of i-th node, if havingThen illustrating has the neighbor node of more than half event will be detected, and abnormal event occurs to having to adjudicate this nodal test according to majority vote method gateway.

The present invention provides a kind of wireless sense network early warning system based on sub-clustering compressed sensing, comprising:

Data Computation Unit, for gateway according to the geographical position of each sensor node in radio sensing network and energy consumption model, calculate the member node in optimum bunch number, best bunch head spatial distribution, each bunch, each bunch set up shortest distance spanning tree between head and gateway;

Two value sequences conversion unit, for obtaining in this bunch the monitoring data of i-th node in tSetup times detection alarm threshold value ��_e, and according to this threshold value, monitoring data is converted to scale-of-two readingObtain in this bunch i-th node to the monitoring data in p moment of the time slide window storage before t, and be converted to two value bit string sequences according to its normal range �� that fluctuates

Reproducing sequence calculates unit, for obtain this bunch bunch head to bunch in two value sequence X of sensor node monitoring value and bit string sequenceRespectively it is carried out dense accidental projection and sparse accidental projection, obtains compressed sensing sequenceWithAnd by sequenceWithIt is transferred to gateway to be reconstructed, obtains reproducing sequenceWith

Effective neighbor node number calculates unit, judges according to the estimated value of in reproducing sequence i-th node for gateway, if estimated valueThen initiated event detects and adds up the effective neighbor node number performing majority vote methodIf the estimated value of the neighbor node j of i-th nodeThen effective neighbor node numberAdd 1;

If estimated valueThen by i-th nodeWith its neighbor node j'sStep-by-step carries out logic and operation, if operation result withIdentical and have non-zero value, then effective neighbor node numberAdd 1;

Abnormal deciding means, i-th node detected by initiated event for gateway performs majority vote method, if effectively neighbor node number is greater than the threshold value of setting, then judges that this node has abnormal event to occur.

The useful effect that the present invention produces is: the wireless sense network method for early warning based on sub-clustering compressed sensing of the present invention, utilizes compressed sensing to have the compression feature unrelated with route and sets up unified energy consumption model and calculate optimum bunch number and the best bunch head spatial distribution; Adopt the variation tendency of bit string monitoring node image data in time slide window that can node being detected, event is predicted, it is to increase the real-time of monitoring method and fault-tolerance, reduce malfunctioning node to the impact of the abnormal event reliability of detection;

Bunch in adopt dense accidental projection and sparse accidental projection to carry out the data gathering of compressed sensing respectively based on monitoring value and the bit string value of threshold value sensor node, it is possible to the effective network energy consumption reduced in data-gathering process;

The Bayes based on local time's dependency is adopted to learn joint sparse reconstruct algorithm at gateway, by this algorithm, the observed value of compressed sensing is carried out fast accurately reconstruct, then at gateway execution majority vote method and bit string matching method, the monitoring data after reconstruct is carried out event detection; Contrasting existing event monitoring scheme, experimental result shows that the present invention monitors that the accuracy rate height of event anomalies, time of response be fast and energy efficient.

Accompanying drawing explanation

Below in conjunction with drawings and Examples, the invention will be further described, in accompanying drawing:

Fig. 1 is the schema of the wireless sense network method for early warning in sub-clustering compressed sensing of the embodiment of the present invention;

Fig. 2 is the change velocity of diffusion V of the wireless sense network method for early warning in sub-clustering compressed sensing of the embodiment of the present invention_eTime three kinds of methods time of response RT scheme;

When Fig. 3 is the change switching threshold �� of the wireless sense network method for early warning in sub-clustering compressed sensing of the embodiment of the present invention, time of response RT of three kinds of methods schemes;

Fig. 4 is the change event detection alarm threshold value �� of the wireless sense network method for early warning in sub-clustering compressed sensing of the embodiment of the present invention_eTime three kinds of methods time of response RT scheme;

Fig. 5 is the change velocity of diffusion V of the wireless sense network method for early warning in sub-clustering compressed sensing of the embodiment of the present invention_eTime three kinds of methods net spent energy figure;

When Fig. 6 is the concept transfer fault probability in percent of the wireless sense network method for early warning in sub-clustering compressed sensing of the embodiment of the present invention, the event rate of false alarm FRR of two kinds of methods schemes;

Fig. 7 is the block diagram of the wireless sense network early warning system in sub-clustering compressed sensing of the embodiment of the present invention;

In figure, 701-Data Computation Unit, 702-bis-value sequence conversion unit, 703-reproducing sequence calculates unit, and the effective neighbor node number of 704-calculates unit, 705-abnormal deciding means.

Embodiment

In order to make the object of the present invention, technical scheme and advantage clearly understand, below in conjunction with drawings and Examples, the present invention is further elaborated. It is to be understood that specific embodiment described herein is only in order to explain the present invention, it is not intended to limit the present invention.

As shown in Figure 1, the wireless sense network method for early warning based on sub-clustering compressed sensing of the embodiment of the present invention, comprises the following steps:

S1, gateway according to the geographical position of each sensor node in radio sensing network and energy consumption model, calculate the member node in optimum bunch number, best bunch head spatial distribution, each bunch, each bunch set up shortest distance spanning tree between head and gateway;

S2, obtain in this bunch the monitoring data of i-th node in tSetup times detection alarm threshold value ��_e, and according to this threshold value, monitoring data is converted to scale-of-two readingObtain in this bunch i-th node to the monitoring data in p moment of the time slide window storage before t, and be converted to two value bit string sequences according to its normal range �� that fluctuates

S3, obtain this bunch bunch head to bunch in two value sequence X of sensor node monitoring value and bit string sequenceRespectively it is carried out dense accidental projection and sparse accidental projection, obtains compressed sensing sequenceWithAnd by sequenceWithIt is transferred to gateway to be reconstructed, obtains reproducing sequenceWith

S4, gateway judge according to the estimated value of in reproducing sequence i-th node, if estimated valueThen initiated event detects and adds up the effective neighbor node number performing majority vote methodIf the estimated value of the neighbor node j of i-th nodeThen effective neighbor node numberAdd 1;

If estimated valueThen by i-th nodeWith its neighbor node j'sStep-by-step carries out logic and operation, if operation result withIdentical and have non-zero value, then effective neighbor node numberAdd 1;

I-th node that initiated event is detected by S5, gateway performs majority vote method, if effectively neighbor node number is greater than the threshold value of setting, then judges that this node has abnormal event to occur.

In another embodiment of the present invention, the concrete steps of present method are:

Step one: gateway according to the geographical position of each sensor node in wireless sense network and energy consumption model calculate the member node in optimum bunch number, best bunch head spatial distribution, each bunch, each bunch set up shortest distance spanning tree between head and gateway;

Step 2: in this bunch, i-th node obtains monitoring data in tAnd according to event detection alarm threshold value ��_eBe converted to scale-of-two readingIn this bunch, perception data in same moment t of n node can represent to be two value sequencesWhereinRepresent normal,Represent that event occurs;

Step 3: in this bunch, i-th node is to time slide window storage p the moment monitoring data before t, and is converted to two value bit string sequences according to its fluctuation normal range ��WhereinRepresent that monitoring value is in normal fluctuation range ��,Represent that monitoring value has exceeded normal fluctuation range ��;

Step 4: this bunch bunch head to bunch in two value sequence X of sensor node monitoring value and bit string sequenceDense accidental projection and sparse accidental projection is adopted to obtain compressed sensing sequence respectivelyWith

Step 5: according to the route strategy transmission sequence of square distance shortest spanning tree between wireless sense network bunch headWithTo gateway;

Step 6: gateway is to sequenceOrthogonal matching pursuit algorithm (OMP algorithm) classical in compressed sensing theory is adopted to carry out accurate reproducing sequenceTo sequenceAdopt a kind of Bayes based on local time's dependency to learn joint sparse reconstruct algorithm and carry out accurate reproducing sequence

Step 7: gateway is according to the estimated value of i-th nodeThen initiated event detects and adds up the effective neighbor node number performing majority vote methodIf the estimated value of the neighbor node j of i-th nodeThen effective neighbor node numberAdd 1; IfBy i-th nodeWith its neighbor node j'sStep-by-step carries out logic and operation, if operation result withIdentical and have non-zero value, then effective neighbor node numberAdd 1;

Step 8: i-th node that initiated event is detected by gateway performs majority vote method, if effective neighbor node numberThen gateway is adjudicated this nodal test abnormal event is occurred to having. Gateway comprehensively then can determine the spatial dimension that abnormal event occurs in all node geographical position detecting that abnormal event occurs further.

The network model assuming the sub-clustering compressed sensing of the present invention is: gateway (BaseStation) and all sensor nodes are all static, gateway has enough electric energy and computing power, the geographical position of gateway each sensor node known, gateway jumps the neighbor node in communication range according to communication radius each node known one. Each physical quantity taken turns event monitoring process interior joint and periodically collect monitoring, meanwhile, the compressed sensing data of gateway periodic harvest and recovery the whole network, all nodes have all had the respective column element of dense accidental projection matrix and sparse projection matrix. Bunch interior nodes directly communicates with a bunch head, data are sent by the form of multihop routing to gateway between leader cluster node, in the process of event monitoring, ignore each node data and process the energy consumed, this is because perform several additive operations and multiplication operation based on the just simple of compressed sensing, the energy that node calculate consumes to be far smaller than the energy that data corresponding consumes.

In clustering routing scheme, network energy consumption is mainly by the impact of leader cluster node number in network and bunch head position distribution, and consider how single measurement value transmitting procedure obtains minimum communication energy consumption in based on the data gathering of compressed sensing, based on above three factors, gateway calculates optimum bunch number according to the geographical position of each sensor node and energy consumption model. Whole network is divided cluster and keeps each bunch of size equal as far as possible by gateway, and gateway selects the node being in bunch central position as a bunch head as far as possible. Gateway sends bunch head information to each leader cluster node, and the nearest leader cluster node of each sensor selection problem is as bunch head of oneself, and groove when each bunch of head be bunch interior nodes distribution, sets up shortest distance spanning tree for all bunches between head and gateway. Employing energy consumption model is as follows:

E_Tx(L, d)=L �� E_elec+L����_fs��d²(1)

E_Rx(L, d)=L �� E_elec(2)

Wherein, E_Tx(L, d) represents that by the data transmission distance of L bit be the energy that d consumes, E_Rx(L, d) represents that node receives the energy of L Bit data consumption, E_elecRepresent the energy sending or receiving a bit and consume, ��_fsRepresent transmission amplifying power.

Each wheel is called an event detection cycle, often take turns event monitoring terminate after each node send oneself residue energy to its bunch of head, each bunch of head selects the node that residue energy is maximum to take turns a bunch head as next, and each bunch of hair send new bunch of head information to new leader cluster node; Each sensor selection problem is from the new leader cluster node of oneself nearest bunch head as oneself; New leader cluster node is bunch interior nodes groove when distributing; New leader cluster node and gateway set up shortest distance spanning tree, ensure that to take turns network average energy consumption in event detection procedure minimum each like this.

Assume based on above these, the optimum bunch number h of network_optCalculation formula be:

$h_{opt}=\sqrt{\frac{{\mathrm{\ϵ}}_{fs}\×N\×A\×S}{12E_{elec}}}---\left(3\right)$

Wherein, N represents the sensor node sum that wireless sense network is disposed, and A represents wireless sense network monitored area area, and S represents the sparse projection matrix �� of compressed sensing^rSparse degree.

In bunch, the node obtains monitoring value in the moment and is converted to scale-of-two reading according to alarm threshold value, and in this bunch, data in the same moment of node can represent to be two value sequences, wherein represents normal, indicates abnormal event generation, and conversion method is as follows:

Each sensor node obtains the primary monitoring data of monitoring target, and according to corresponding alarm threshold value ��_eThe monitoring value in current detection cycle is converted into scale-of-two reading 0 or 1 (two values), and in this bunch, i-th node obtains monitoring value in tBe converted to scale-of-two readingIn this bunch, data in same moment t of n node can represent to be two value sequencesWhereinRepresent normal,Indicating that abnormal event occurs, conversion method is as follows:

Meanwhile, i-th node is to the monitoring value storing nearest p the moment before t at time slide windowAccording to its fluctuation normal range ��, monitoring value sequence is carried out difference computing and it is converted into two value bit string sequencesWhereinRepresent that monitoring value is in normal fluctuation range ��,Representing that monitoring value has exceeded normal fluctuation range ��, conversion method is as follows:

Wherein, �� is switching threshold, represents the normal fluctuation range of adjacent two monitoring moment node monitoring values,It is the differential conversion bit (two value) of i-th node in t. Difference scale-of-two sequence reflects the variation tendency of node monitoring value within for some time. In the monitoring of environmental of reality, when the monitoring value of normal node occurs without event in monitored area, change is little, and monitoring value is in normal fluctuation range, and scale-of-two sequence is full 0 sequence; When in monitored area, event occurs, on the dispersal direction of event area and event, the monitoring value of node there will be bigger change, exceedes the normal fluctuation range �� of monitoring value, and scale-of-two sequence comprises 1. The temperature (unit is Kelvin) that such as i-th node stores nearest 11 moment at time slide window is { 294,295,293.4,292.9,297.8,308.1,322.7,322.9,332.3,359.1,396.3}, when ��=5, corresponding difference scale-of-two sequenceAs the neighbor node j of node i, the temperature storing nearest 11 moment at time slide window is { 291.1,292.7,290.9,293.6,293.8,299.5,304.3,304.7,308,315. 7,325.9}, corresponding difference scale-of-two sequenceWork as ��_eDuring=380K, the monitoring value 396.3K of node i has exceeded event detection alarm value, and neighbor node j does not also exceed, butWithStep-by-step carries out logic and operation, namelyAndIn comprise 1 number be not 0, then difference scale-of-two sequence between node is mated mutually, it will be recognized that this neighbor node will detect event, and when majority voting, this neighbor node will be voted for, and accelerates the time of response fast RT that WSN detects event like this.

This bunch bunch of head to bunch in two value sequence X of sensor node monitoring value adopt dense accidental projection to carry out data gathering, obtain compressed sensing sequenceCalculation formula be:

Wherein,For dense accidental projection matrix ��^eRow k l column element, each node produces the random element of mThe value of m is determined by the sparse degree of sequence X in network, owing to perception sequence X itself is sparse, because occurring abnormal event to be sparse for the whole network monitored area, only a small amount of node can detect that event occurs, therefore, a bunch head adopts dense accidental projection to carry out data gathering, can ensure sequenceReconstruction accuracy in turn save communication energy consumption expense in event monitoring process.

This bunch bunch of head to bunch in the bit string sequence of sensor node monitoring valueAdopt sparse accidental projection to carry out data gathering, obtain compressed sensing sequenceCalculation formula be:

Wherein,For sparse accidental projection matrix ��^rRow k l column element, each node produceIndividual random element Value by sequenceSparse degree threshold valueDetermine, due toItself not being very sparse, therefore, a bunch head adopts sparse accidental projection to carry out data gathering, can ensure sequenceReconstruction accuracy in turn save communication energy consumption expense in event monitoring process.

In fact, node needs to participate in the data of its institute's perception of collecting that and if only if of compressed sensing observed value and corresponding calculation matrix element is nonzero value. For sequence X and sequenceSparse degree different, the reconstruction accuracy of sequence in turn saves transport communication energy consumption expense in data gathering to adopt dense and sparse accidental projection to ensure respectively.

Then, according to the route strategy transmission compressed sensing sequence of square distance shortest spanning tree between network each bunch of headWithTo gateway, gateway is to the sequence receivedOrthogonal matching pursuit algorithm (OMP algorithm) classical in compressed sensing theory is adopted to carry out accurate reproducing sequenceTo the sequence receivedAdopt a kind of Bayes based on local time's dependency to learn joint sparse reconstruct algorithm (LT-SBL algorithm) accurately to reconstruct.

In each time window, it is proposed to a kind of Bayes based on local time's dependency learns joint sparse signal model and is:

${\tilde{Y}}^t={\mathrm{DY}}^t+V---\left(8\right)$

Wherein,I is unit matrix, and V is unknown normal distribution noise.

If p (V) is normal distribution N (0, ��), then haveFor normal distribution �� (0, ��), p (Y^t|��_i,B_i) it is normal distribution �� (0, ��₀). Wherein:

If estimating parameter lambda, ��_i, B_i(1��i��p �� n), then can derive Y based on Bayes principle^tMAP estimation:

$Y^t={\mathrm{\Σ}}_0{D}^T{(\mathrm{\λ}I+{\mathrm{D\Σ}}_0{D}^T)}^{-1}{\tilde{Y}}^t---\left(10\right)$

Adopt Bayes's method to learn the various parameter of this model equally, these parameter characterizations allied signal sparse internal structured information, �� can be obtained through deriving_i, B and �� calculation formula be respectively:

${\mathrm{\γ}}_i\←\frac{1}{p\×n}{Y}^t{B}^{-1}{\left(Y^t\right)}^T---\left(11\right)$

$B\←\underset{i=1}{\overset{p\×n}{\Σ}}\frac{{\left(Y^t\right)}^T{Y}^t}{{\mathrm{\γ}}_i}---\left(12\right)$

$\mathrm{\λ}\←\frac{1}{p\×n}\left|\right|{\tilde{Y}}^t-{\mathrm{DY}}^t|{|}^2---\left(13\right)$

Gateway reproducing sequence sequence X and sequenceAfter, according to the estimated value of i-th nodeThen initiated event detects and adds up the effective neighbor node number performing majority vote methodIf the estimated value of the neighbor node j of i-th nodeThen effective neighbor node numberAdd 1; IfBy i-th nodeWith its neighbor node j'sStep-by-step carries out logic and operation, if operation result withIdentical and have non-zero value, then effective neighbor node numberAdding 1, this illustrates that node j is identical with the monitoring value variation tendency of node i, although the monitoring value of node j does not exceed alarm threshold value ��_e, but occurred exceeding normal fluctuation range, therefore think that node j will detect event.

If havingThen illustrating has the neighbor node of more than half event will be detected, and abnormal event occurs to having to adjudicate this nodal test according to majority vote method gateway. Gateway is comprehensively all detects the node geographical position that abnormal event occurs, then can determine the spatial dimension that abnormal event occurs further.

The present invention is directed to the dispersion pattern event of a class extensively existence in the true world and carry out monitoring index system such as landslide, hazardous gas spillage, noxious pollutant diffusion and fire etc. Surrounding environment along the diffusion of some specific direction, and can be had an impact by dispersion pattern event under the impact of outside environmental elements. On the dispersal direction of event, from event more close to region, the impact of event is more big. The impact of surrounding environment is all permanent, extensively dynamic change by this kind of event over time and space. In the actual environment, the abnormal event of dispersion pattern can be divided into two kinds: ascending-type event and down type event. The feature of ascending-type event is the normal change scope that event value is greater than attribute, and the feature of down type event is event value is less than the normal change scope of attribute. Emulation adopts ascending-type event as event model (the present invention is equally applicable to down type event), therefore when event occurs in monitored area, event to around regional diffusion process in, the sensor node being subject to events affecting has high monitoring value, and the node not being subject to events affecting then has low monitoring value.

Network model assumes that each node knows event detection alarm threshold value ��_eWith the proper distribution scope �� of monitoring value. Obviously, it is abnormal that fault and event all can make the monitoring value of sensor node occur, when representing node to the judgement of event by a binary bits, this kind of denominator of node fault and event can cause individual node cannot differentiate node fault and event, cause sensing network to the erroneous judgement of event, for improving the accuracy of sensing network event detection, sensor node needs the mode cooperated with its neighbor node event to be detected.

Experimental data adopts the wireless sense network laboratory sensor node collecting temperature data in California, USA university Berkeley branch school, and node i obtains monitoring valueAdopt diffusion event model as follows:

$A_i^t={\mathrm{\λ}}_e{(1+\frac{d\left(i\right)}{\sqrt{\left|D({\mathrm{\λ}}_e,t)\right|/\mathrm{\π}}})}^{-2}$

Wherein, d (i) represents node i and the distance of event generation area, D (��_e, t) presentation of events generation area, D (��_e, t) by velocity of diffusion V_eImpact.

In simulation process, basic parameter comprises interstitial content, Target monitoring area area, node communication radius and probability of node failure; Other parameters comprise time slide window yardstick W, event velocity of diffusion V_e, event detection alarm threshold value ��_e, and switching threshold �� can according to the suitable value of actual monitoring Object Selection.

Contrast experiment is carried out with classical DR method and TV method in order to assess the validity of the present invention's (referred to as CS-BMV method). Adopt time of response RT (in units of taking turns), event rate of false alarm FRR and net spent energy three kinds of indexs to weigh the energy saving of the present invention, real-time and reliability. Time of response is defined as event generation and event area interior nodes detects the time difference between this event first, illustrates detection method to the susceptibility of diffusion event, and the time of response is more short, and the speed of detection method detection event is more fast, and its performance is more good. Event rate of false alarm is defined as when occurring without event, the node number of event and the ratio of node total number detected, has reacted fault-tolerance and the reliability of detection method, and event rate of false alarm is more low, and detection method fault-tolerance is more good. The summation that net spent energy is defined as in event detection procedure all the sensors node consumed energy, net spent energy is more little, and detection method energy saving is more good.

As shown in Figure 2, at change event velocity of diffusion V_eTime CS-BMV method detection RT and the DR method of event, TV method comparison figure. As can be seen from Figure 3, RT and the DR of CS-BMV close to and much smaller than TV, experimental result demonstrates the validity of CS-BMV method. Because CS-BMV method adopts moving window matching mechanisms to predict whether neighbor node can detect event so that (its monitoring value does not also reach affair alarm value �� to have Similar trend neighbor node with this ground node_e) have an opportunity to participate in majority voting, shorten the time of response of local nodal test to abnormal event. Further, along with the increase of event velocity of diffusion, the time of response of three kinds of methods all can reduce. Therefore, when event occurs in monitored area, the CS-BMV method of the present invention compared with TV can detect event at faster speed, it is to increase the promptness of the abnormal event of sensing network report.

As shown in Figure 3, RT and the DR method of CS-BMV method detection event, the comparison figure of TV method when changing switching threshold ��. As can be seen from Figure 3, the RT of DR method is not by the impact changing switching threshold ��, and the abnormal event of DR method report is only by alarm threshold value ��_eImpact. CS-BMV method adopts moving window matching mechanisms to predict whether neighbor node can detect event, to time slide window store monitoring data carry out two values time determine by switching threshold ��, the RT of CS-BMV method with �� increase slightly increase with DR close to but much smaller than TV method.

As shown in Figure 4, at change event detection alarm threshold value ��_eTime CS-BMV method detection RT and the DR method of event, TV method comparison figure. As can be seen from Figure 4, the RT of three kinds of methods is with ��_eIncrease all can increase, but RT and the DR method of CS-BMV method closer to, have better real-time than TV method.

As shown in Figure 5, at change event velocity of diffusion V_eTime net spent energy and the DR method of CS-BMV method detection event, TV method compare figure. As can be seen from Figure 5, the net spent energy of CS-BMV method detection event is less than DR method and TV method. Because CS-BMV method adopts dense accidental projection and sparse accidental projection based on compressed sensing theory to carry out data gathering, effectively reduce the energy expenditure that the node in receipt collection process participates in communication, adopt orthogonal matching pursuit algorithm (OMP algorithm) and the Bayes based on local time's dependency to learn joint sparse reconstruct algorithm (LT-SBL algorithm) at gateway and respectively compression data have been carried out Exact recovery; Excavate the data dependence between node by compression sensing method, eliminate the redundancy between data to greatest extent, while ensureing significantly to fall low amount of transmitting data, still can keep the event monitoring performance of high precision, extend the network life cycle.

As shown in Figure 6, during probability of node failure per-cent the detection of CS-BMV method event rate of false alarm FRR and TV method compare figure, owing to DR method is without its rate of false alarm height of fault-tolerance, so Fig. 5 only compares the rate of false alarm FRR of CS-BMV method and TV method. As can be seen from Figure 5, along with probability of node failure per-cent increases, CS-BMV method detection event rate of false alarm FRR is almost 0. On the contrary, TV method only when probability of node failure is lower rate of false alarm FRR less, one when probability of node failure per-cent is more than 30%, and the rate of false alarm FRR of TV method increases clearly, so CS-BMV method has better fault-tolerance than other method.

CS-BMV method has better fault-tolerance and not only adopts majority vote method to eliminate malfunctioning node to the impact of event detection precision, have employed the variation tendency of bit string pattern match image data in time slide window further, can this node can not only be detected that abnormal event is predicted by bit string pattern match, and reduces malfunctioning node to the impact of the abnormal event reliability of detection. Assume that sensor node produces permanent fault, owing to permanent trouble duration is longer, sensor node influence time is longer, it is all zero string that permanent malfunctioning node produces bit string in time slide window, can not with the bit string pattern match of normal node, sensing network interior joint and neighbor node thereof occur that the spatial coherence that permanent fault affects is very little simultaneously simultaneously, and therefore present method is easy to get rid of permanent malfunctioning node to the impact on event detection precision; If sensor node produces transient fault, owing to the transient fault time length is short, the bit string of transient fault node also can not be identical with the Changing Pattern of the image data of normal node, the probability that both generation bit string patterns can be mated is extremely low, therefore, transient fault node can not participate in majority voting, it is possible to gets rid of instantaneity to the impact on event detection precision.

In a word, comparing with existing DR with MV, the present invention utilizes compressed sensing to have the compression feature unrelated with route and sets up unified energy consumption model, calculates optimum bunch number and best bunch head spatial distribution. Adopt the variation tendency of bit string pattern match image data in time slide window, can this node can not only be detected that abnormal event is predicted, and reduce malfunctioning node to the impact of the abnormal event reliability of detection, it is to increase the real-time of monitoring method and fault-tolerance. Bunch in adopt dense accidental projection and sparse accidental projection to carry out data gathering in compressed sensing mode respectively based on monitoring value and the bit string value of threshold value sensor node, can effectively reduce the network communication energy consumption in receipt collection process, adopt the Bayes based on local time's dependency to learn joint sparse reconstruct algorithm at gateway, by this algorithm, compression data are carried out fast accurately reconstruct. Then, at gateway execution majority vote method and bit string matching method, the monitoring data after reconstruct is carried out event detection. Contrasting existing event monitoring scheme, experimental result shows that the accuracy rate height of the abnormal event of early warning of the present invention, time of response be fast and energy efficient.

As shown in Figure 7, the wireless sense network early warning system based on sub-clustering compressed sensing of the embodiment of the present invention, for realize the embodiment of the present invention in the wireless sense network method for early warning of sub-clustering compressed sensing, comprising:

Data Computation Unit 701, for gateway according to the geographical position of each sensor node in radio sensing network and energy consumption model, calculate the member node in optimum bunch number, best bunch head spatial distribution, each bunch, each bunch set up shortest distance spanning tree between head and gateway;

Two value sequences conversion unit 702, for obtaining in this bunch the monitoring data of i-th node in tSetup times detection alarm threshold value ��_e, and according to this threshold value, monitoring data is converted to scale-of-two readingObtain in this bunch i-th node to the monitoring data in p moment of the time slide window storage before t, and be converted to two value bit string sequences according to its normal range �� that fluctuates

Reproducing sequence calculates unit 703, for obtain this bunch bunch head to bunch in two value sequence X of sensor node monitoring value and bit string sequenceRespectively it is carried out dense accidental projection and sparse accidental projection, obtains compressed sensing sequenceWithAnd by sequenceWithIt is transferred to gateway to be reconstructed, obtains reproducing sequenceWith

Effective neighbor node number calculates unit 704, judges according to the estimated value of in reproducing sequence i-th node for gateway, if estimated valueThen initiated event detects and adds up the effective neighbor node number performing majority vote methodIf the estimated value of the neighbor node j of i-th nodeThen effective neighbor node numberAdd 1;

If estimated valueThen by i-th nodeWith its neighbor node j'sStep-by-step carries out logic and operation, if operation result withIdentical and have non-zero value, then effective neighbor node numberAdd 1;

Abnormal deciding means 705, i-th node detected by initiated event for gateway performs majority vote method, if effectively neighbor node number is greater than the threshold value of setting, then judges that this node has abnormal event to occur.

Should be understood that, for those of ordinary skills, it is possible to improved according to the above description or convert, and all these improve and convert the protection domain that all should belong to claims of the present invention.

Claims (9)

1. the wireless sense network method for early warning based on sub-clustering compressed sensing, it is characterised in that, comprise the following steps:

S1, gateway according to the geographical position of each sensor node in radio sensing network and energy consumption model, calculate the member node in optimum bunch number, best bunch head spatial distribution, each bunch, each bunch set up shortest distance spanning tree between head and gateway;

S2, obtain in this bunch the monitoring data of i-th node in tSetup times detection alarm threshold value ��_e, and according to this threshold value, monitoring data is converted to scale-of-two readingObtain in this bunch i-th node to the monitoring data in p moment of the time slide window storage before t, and be converted to two value bit string sequences according to its normal range �� that fluctuates

S3, obtain this bunch bunch head to bunch in two value sequence X of sensor node monitoring value and bit string sequenceRespectively it is carried out dense accidental projection and sparse accidental projection, obtains compressed sensing sequenceWithAnd by sequenceWithIt is transferred to gateway to be reconstructed, obtains reproducing sequenceWith

S4, gateway judge according to the estimated value of in reproducing sequence i-th node, if estimated valueThen initiated event detects and adds up the effective neighbor node number performing majority vote methodIf the estimated value of the neighbor node j of i-th nodeThen effective neighbor node numberAdd 1;

If estimated valueThen by i-th nodeWith its neighbor node j'sStep-by-step carries out logic and operation, if operation result withIdentical and have non-zero value, then effective neighbor node numberAdd 1;

I-th node that initiated event is detected by S5, gateway performs majority vote method, if effectively neighbor node number is greater than the threshold value of setting, then judges that this node has abnormal event to occur.

2. the wireless sense network method for early warning based on sub-clustering compressed sensing according to claim 1, it is characterised in that, step S1 calculates optimum bunch number h_optCalculation formula be:

$h_{opt}=\sqrt{\frac{{\mathrm{\ϵ}}_{fs}\×N\×A\×S}{12E_{elec}}}$

Wherein, N represents the sensor node sum that wireless sense network is disposed, and A represents wireless sense network monitored area area, and S represents the sparse projection matrix �� of compressed sensing^rSparse degree, E_elecRepresent the energy sending or receiving a bit and consume, ��_fsRepresent transmission amplifying power.

3. the wireless sense network method for early warning based on sub-clustering compressed sensing according to claim 1, it is characterised in that, the formula that monitoring data is converted in step S2 scale-of-two reading is:

Wherein,For in this bunch, i-th node obtains monitoring data in t, ��_eFor event detection alarm threshold value.

4. the wireless sense network method for early warning based on sub-clustering compressed sensing according to claim 1, it is characterised in that, the formula that monitoring data is converted in step S2 two value bit string sequences is:

Wherein, i-th node to storing the monitoring value in nearest p moment at time slide window before t is�� is switching threshold,It is the differential conversion bit of i-th node in t.

5. the wireless sense network method for early warning based on sub-clustering compressed sensing according to claim 1, it is characterised in that, step S3 adopt dense accidental projection and sparse accidental projection obtain compressed sensing sequenceWithCalculation formula be:

$\tilde{X}={\mathrm{\φ}}^{e}X$

${\tilde{Y}}_i^t={\mathrm{\φ}}^r{Y}_i^t$

Wherein, dense accidental projection matrix ��^eL column element production method be:

Wherein, each sensor node produces m random elementThe value of m is by perception data in networkSparse degree determine;

Sparse accidental projection matrix ��^rL column element production method be:

Wherein, each sensor node producesIndividual random element Value by perception data in networkSparse degree threshold value S=log (n)/(2n) determine.

6. the wireless sense network method for early warning based on sub-clustering compressed sensing according to claim 1, it is characterised in that, the reconstructing method in step S3 is specially:

Gateway is to sequenceEmploying orthogonal matching pursuit algorithm is reconstructed, and obtains sequence

To sequenceAdopt the Bayes based on local time's dependency to learn joint sparse reconstruct algorithm to be reconstructed, obtain sequence

7. the wireless sense network method for early warning based on sub-clustering compressed sensing according to claim 1, it is characterised in that, the threshold value arranged in step S5 isRepresent the neighbor node number in a jumping communication range of i-th node.

8. the wireless sense network method for early warning based on sub-clustering compressed sensing according to claim 7, it is characterised in that, the determination methods in step S5 is specially:

Its effective neighbor node number of i-th node calculate that initiated event is detected by gatewayRepresent the neighbor node number in a jumping communication range of i-th node, if havingThen illustrating has the neighbor node of more than half event will be detected, and abnormal event occurs to having to adjudicate this nodal test according to majority vote method gateway.

9. the wireless sense network early warning system based on sub-clustering compressed sensing, it is characterised in that, comprising:

Data Computation Unit, for gateway according to the geographical position of each sensor node in radio sensing network and energy consumption model, calculate the member node in optimum bunch number, best bunch head spatial distribution, each bunch, each bunch set up shortest distance spanning tree between head and gateway;

Two value sequences conversion unit, for obtaining in this bunch the monitoring data of i-th node in tSetup times detection alarm threshold value ��_e, and according to this threshold value, monitoring data is converted to scale-of-two readingObtain in this bunch i-th node to the monitoring data in p moment of the time slide window storage before t, and be converted to two value bit string sequences according to its normal range �� that fluctuates

Reproducing sequence calculates unit, for obtain this bunch bunch head to bunch in two value sequence X of sensor node monitoring value and bit string sequenceRespectively it is carried out dense accidental projection and sparse accidental projection, obtains compressed sensing sequenceWithAnd by sequenceWithIt is transferred to gateway to be reconstructed, obtains reproducing sequenceWith

Effective neighbor node number calculates unit, judges according to the estimated value of in reproducing sequence i-th node for gateway, if estimated valueThen initiated event detects and adds up the effective neighbor node number performing majority vote methodIf the estimated value of the neighbor node j of i-th nodeThen effective neighbor node numberAdd 1;

If estimated valueThen by i-th nodeWith its neighbor node j'sStep-by-step carries out logic and operation, if operation result withIdentical and have non-zero value, then effective neighbor node numberAdd 1;

Abnormal deciding means, i-th node detected by initiated event for gateway performs majority vote method, if effectively neighbor node number is greater than the threshold value of setting, then judges that this node has abnormal event to occur.