CN101686262B - Multi-node collaboration based storage method for sensor network - Google Patents

Abstract

The invention provides a multi-node collaboration based storage method for a sensor network. The method comprises the following steps: (1) determining a node collaboration mode, which is to extract image characteristics and identification images on nodes of the sensor network, and determine a node set for data fusion; (2) determining an updating mode for storing and exhausting the nodes of the sensor network; (3) dynamically calculating the size of the most appropriate data of the nodes of the sensor network in each time of data updating or data transiting; (4) solving appropriate positions of Sink nodes by using a genetic algorithm; (5) constructing a hierarchical network topology structure to optimize a user search processing process by improving the TopDisc algorithm; and (6) optimizing a data searching process, which is to establish the data searching process driven by messages by establishing and managing multiple data transcriptions. The method is used for solving the problems of more events detected by a single node and unbalanced consumption of energy and storage among different nodes, and prolongs the life time of the network.

Description

A kind of storage method for sensor network based on multi-node collaboration

One, technical field

The present invention is directed to large-scale wireless multimedia sensor network; Designed a kind of storage method for sensor network based on multi-node collaboration; Adopt this method can solve more energy consumption that causes of incident and the too fast problem of storage consumption that individual node senses; Energy and storage consumption lack of uniformity problem between different nodes, and the life span that prolongs network.

Two, background technology

Wireless multimedia sensor network is on traditional wireless sensor networks (WSNs) basis, to have introduced a kind of novel sensor network that forms behind the multimedia messages sensing functions such as audio frequency, video, image; Its node generally also is equipped with multi-medium data collecting devices such as CMOS camera, mini microphone except the transducer that is equipped with simple environmental data collection function.Compare with the traditional WSNs that only has simple scalar (like temperature, humidity etc.) environmental data collection function, multimedia messagess such as the audio frequency that WMSNs ability sensitive information amount is abundant more, video, image can be realized fine granularity, the precisely monitoring of information; Energy consumption is higher aspect data acquisition, processing energy consumption; Complicated more during data processing, node often need be done operations such as image compression, coding and decoding, distributed video processing, data fusion to the data that collect; Application scenarios is more paid attention to the quality of data, require network can to the inquiry response fast as far as possible, for the user, network should possess higher throughput; Multimedia sensor network is pursued energy consumption minimized under the situation that qos requirement is met again, and is that cost is pursued the energy consumption minimum with the characteristic of sacrificing aspect a part of QoS often at traditional WSNs.The existence of these characteristics makes in WMSNs individual node seem more outstanding because of the unbalanced problem of energy, storage consumption between the too much and a plurality of nodes of data volume that produce, and node is easy to quit work because of depleted of energy or storage exhaust.In addition, be large scale network because we consider, the interstitial content of large scale network is many, coverage is big, and conflict collision, data redundancy, control expense, also increase thereupon of probability makes mistakes.

We consider to alleviate the reduction of these factors to network performance through strengthening internodal cooperation, before this, do not see that other are used for the route and the design Storage scheme in order to the task of accomplishing data collection, inquiry of this comparatively harsh operational environment.

Three, summary of the invention

The present invention seeks to: solve the more and data volume that produces of incident that individual node senses too much with the different nodes that bring thus between energy and storage consumption lack of uniformity problem, and the method that topology optimization and a plurality of data trnascription are managed when adopting net stored, response user inquiring in the design large scale multimedia sensor network.And the life span that prolongs network.

To achieve these goals; Technical scheme of the present invention is: based on the storage method for sensor network of multi-node collaboration; Especially massive wireless sensor storage means and based on inquiry processing method (the A Collaboration-basedStorage Management Scheme in Multimedia Sensor Networks of many copies; CBSM), step is following:

(1) confirm the node cooperation mode: the node at sensor network extracts characteristics of image and identification image, and communicates by letter with neighbor node with the mode of message-driven, to confirm to do the set of node of data fusion;

(2) confirm that storage exhausts the update mode of sensor network nodes: an important optimization to the conventional store distributed storage is the Data Update strategy that has added " seemingly-dead " node, designs this tactful purpose and is in energy consumption and preserve the quality of data this two and obtain a compromise;

(3) take a kind of load-balancing method: dynamically each Data Update of calculating sensor network node or the only size of data when shifting, equilibrium on the same to a certain extent node of this method between energy consumption and the storage consumption and the energy consumption between different nodes and storage consumption equilibrium;

(4) take the layout type of a plurality of Sink nodes:, use the correct position of genetic algorithm for solving Sink to the deployment general layout at random of node.The Sink node is the network key point;

(5) building network topology: adopt the mode of sub-clustering commonly used, through TopDisc (topological control algolithm) algorithm is improved, the network topology structure that makes up stratification is optimized the user inquiring processing procedure;

(6) optimize the data query process:, set up the data query process of message-driven through creating and managing a plurality of data trnascriptions;

The invention has the beneficial effects as follows: the method for employing can solve more energy consumption that causes of incident and the too fast problem of storage consumption that individual node senses, energy and storage consumption lack of uniformity problem between different nodes, and the life span that prolongs network.

Four, embodiment

Stage 1: confirm the node cooperation mode

The computing time of the local time that each node writes down respectively is poor, utilizes relative current time definition of time difference event identifier.For example: after incident E took place, node A, B had sensed this incident and have come to be recorded as respectively T according to separately local zone time _AEAnd T _BE, the moment that A communicates by letter with B is T _AAnd T _B, through more whether satisfying

Δ T _A(=T _A-T _AE)=Δ T _B(=T _B-T _BE), whether what judge that they monitor is same incident.

After the sign to incident; The node cooperation mode is following: after the node of the incident of sensing obtains view data; To be buffered in the RAM of this node earlier and, generate cooperation message in view of the above, then the query messages of the following form of broadcasting in communication range image feature extraction:

Msg

{

MsgType=QueryMsg; The definition of // type of message

SensorId; The node identification of this message of // transmission

EventTime; // Time To Event (TAE)

CurrentTime; // the current time (TA)

ImageDescription; // characteristics of image is described

ImageSize; The size of data of // description image

}

The purpose of sending this message is to contain related data to which neighbor node around this this node of incident inquiry, so that carry out data fusion.

After supposing that the A node has sent QueryMsg; After the B node is received this message; If be judged as same incident; Then B node ImageDescription attribute from the message that A sends gets access to the characteristics of image described in the A, and calculates the similarity (through calculating covariance) with the ImageDescription value of this node, reaches a given in advance threshold alpha as if this similarity ₁, then show among A and the B similarly about the described data height of the image of same incident, only need this moment one of them node to preserve this view data and get final product, we arrange to judge that ImageSize preserves big or small smaller, another node abandons the data in the buffer memory; If this similarity does not reach given in advance threshold alpha ₁, but reached α ₂Though this shows that A is highly not similar with the data among the B, and certain correlation is arranged, and can consider to do data fusion, B will return the message that an affirmation can be done data fusion to A this moment, and message format is:

Msg

{

MsgType=AgreMsg; The definition of // type of message

SensorId; The node identification of this message of // transmission

Simility; // the degree of correlation

}

When A sent QueryMsg, timer of initialization was opened a Messages-Waiting process, when the timer time is 0, just no longer waited for receiving message.Before this timing course stopped, the A node received that whenever the SensorId that an AgreMsg just will send this node joins in the set, and this set is the alternative collection (AggregationSet) that can do data fusion with A.By the time after the Messages-Waiting process finished, Aggregation Set was definite, this moment the A node in order to do data fusion, and data are passed to a highest node of similarity among the Aggregation Set;

After the A node has been selected the transmission destination node; Earlier image is adopted wavelet transformation during the transmission data; On many levels, data are decomposed, picture breakdown is become several different bags of priority, and the requirement of the picture quality bag that importance is low is abandoned according to using; Make image in the consumption of storage and energy, obtain reduction, adopt half reliable transmission means (semi-reliable transmission) that transfer of data is arrived destination node then.

Stage 2: confirm that storage exhausts the update mode of node

For node A, if, just being considered to storage less than preset threshold value, its residual memory space is about to exhaust, this moment, we adopted following steps to upgrade its memory space:

At first, from the storage of A, select the take up space image of maximum of institute, obtain description this image, and in communication range broadcast query message QueryImageMsg:

Msg

{

MsgType=QueryImageMsg; The definition of // type of message

SensorId; The node identification of this message of // transmission

ImageDescription; // characteristics of image is described

ImageSize; // image size (approximation)

}

Node B receives the relatively similarity of the ImageDescription of existing image on image and the node in message of this message, as if there being similar image, then returns AckMsg message to the A node, and structure is following:

Msg

{

MsgType=AckMsg; The definition of // type of message

SensorId; The node identification of this message of // transmission

}

Termination messages is waited for process after the A node is received first AckMsg, and describes the data deletion of this image in will storing, and adds " pointer " that points to the B node simultaneously and replaces original storage data, and storage update has been accomplished in this operation.

Equally; A timer is set after node A has broadcasted query messages QueryImageMsg opens the Messages-Waiting process; If this wait process finishes also not receive any AckMsg that returns; Show that then adjacent node does not have the similar data with this view data, in order to upgrade the memory space of oneself, the A node can only be transferred to this partial data the enough neighbor nodes of nearest remaining space or directly delete this part data.

Stage 3: designed a kind of load-balancing method

Process in the data migration can produce the phenomenon of " reverse "; That is: the A node gets into " seemingly-dead " state; For updated stored with transfer of data to the B node, two kinds of special circumstances appear sometimes: (1) but the B node of original operate as normal is the data " seemingly-dead " of ι because of having received size of data; (2) the A node is become " very dead " owing to send big or small data for ι by " seemingly-dead ".The root of the appearance of both of these case is that all data ι's is big or small excessive.

For avoiding less ι can not alleviate " seemingly-dead " state of A node, excessive ι can cause " reverse " phenomenon, and we dynamically control the size of ι, makes it to become a value that changes with node current remaining and store status.

Be respectively E if be about to take place A, B two node primary powers and the initial memory of data migration _A0, E _B0, V _A0, V _B0, current energy is E _A, E _B, V _A, V _B, then need the size of transferring data to get at every turn

The time; It is big as far as possible both can to have satisfied the space that makes the A node updates; " reverse " phenomenon can not take place again, has accomplished to a certain extent on the same node that equilibrium between the energy consumption and storage consumption and the energy consumption between different nodes and storage consumption are balanced.

Stage 4: the layout type that designs a plurality of Sink

The layout of Sink done distribute rationally, optimal target is to find the optimum position of arranging the Sink node, makes each sensor node and Sink node Weighted distance and minimum.This problem belongs to the nonlinear combination optimization problem; Be proved to be the NP-hard problem; Use enumerative technique can reach optimal solution in theory, but computing time, expense was excessive, and along with after the WSN work node failure being arranged constantly; This computational process needs constantly dynamically to carry out, and is difficult to put into practice.

This programme adopts genetic algorithm to find the solution the correct position of Sink, and this problem modeling is following:

Table 1 variable-definition

In the candidate collection M of Sink node, find K position that is fit to, total Weighted distance that the optimal solution of being made up of these positions is satisfied between sensor node and the Sink node reaches minimum, promptly asks for

$\min \underset{i=1}{\overset{N}{\mathrm{\Σ}}}\underset{j\∈M}{\mathrm{\Σ}}{q}_i{d}_{\mathrm{ij}}{T}_{\mathrm{ij}}{S}_j.$

The Master that separates of genetic algorithm is the parameter coding of asking problem a chromosome, utilizes the mode of iteration to select again, mating and variation computing exchange chromosomal information in the population, the final chromosome that meets optimization aim that generates.In the expression formula that we will find the solution, as the decision variable of parameter and possibly value be:

$S_j=\left\{\begin{array}{cc}1& \mathrm{if\; j\; is\; a}\sin k\\ 0& \mathrm{other}\end{array}\right.$

$T_{\mathrm{ij}}=\left\{\begin{array}{cc}1& \mathrm{if\; node\; i\; com\; micate\; with}\sin \mathrm{k\; j}\\ 0& \mathrm{other}\end{array}\right.$

Constraints is:

$\left\{\begin{array}{c}\underset{j}{\mathrm{\Σ}}{T}_{\mathrm{ij}}=1\\ \underset{j}{\mathrm{\Σ}}{S}_j=K\\ T_{\mathrm{ij}}-S_j\≤0\end{array}\right.$

Finding the solution each required key element of genetic algorithm is:

Chromosome coding: the two-dimensional coordinate with the grid overcrossing point is a coded object, adopts binary coding method.

Fitness function:

$F\left(x\right)=\left\{\begin{array}{cc}{C}_{\max }-f\left(x\right)& (f\left(x\right)<C_{\max })\\ 0& (f\left(x\right)\&GreaterEqual;C_{\max })\end{array}\right.$

Wherein, C _MaxBe a bigger constant, guarantee that the individual fitness F (x) of F (x) always is a nonnegative value, Be target function.

Selection operation: adopt the fitness ratio method to come the individuality in the population is selected.If group size is M, the fitness of individual i is F _i, the selected probability of then individual i does

Interlace operation: adopt the single-point cross method.

Mutation operation: adopt basic position variation method.

After having defined each required key element of genetic algorithm,, calculate the result, try to achieve the particular location of Sink node according to these results its input as genetic algorithm kit among the matlab.

Stage 5: building network topology

This programme adopts the mode of sub-clustering commonly used, and through the TopDisc algorithm is improved slightly, the network topology structure that makes up stratification is optimized the user inquiring processing procedure.The state of network is divided into: topology constructing stage and data transfer phase.Indeterminate at whole topology of networks of topology constructing stage, node all is in active state, the request data package that is used for Topology Discovery is carried out inundation propagate.At first, agent node sends the message of setting up topology to the Sink node; Then, by the broadcast that starts " discovery neighbor node " from the nearest sensor node of Sink node, each message is all carried a kind of state information; Neighbor node waits for a period of time after receiving message, and this time is inversely proportional to according to distance and the current dump energy of node between itself and the sending node, that is:

$\mathrm{\&Delta;T}=\frac{c_1}{d}+\frac{c_2}{E}$

Wherein, Δ T is the wait time delay among the TopDisc, c ₁, c ₂Be delay constant, d is for sending the distance of query messages, and E is the current energy of node.

Along with the Topology Discovery request message is propagated in network, sensor network is divided into a plurality of bunches (cluster), and each bunch all has a bunch of head.By bunch head be responsible for to Sink communication and in this bunch other nodes send Query Informations, can direct communication between bunch head or communicate through the Sink node, thus formed a hierarchical structure that can cover whole guarded region sensor network.

At data transfer phase, except transmitting normal task property data, also regularly around leader cluster node, carry out topology reconstruction, thereby realize rotating of bunch head.

Stage 6: optimize the data query process

S Sink node all received this query requests; And in communication range separately broadcasting this query requests; Receive that the leader cluster node of this query requests transmits this query requests in this bunch scope; Each receives that the node of this query requests judges whether to contain the needed data of user through the comparing data feature description; If wherein node i comprises the data of being inquired about, then the leader cluster node to its place bunch returns Query Result, and by this leader cluster node Query Result is returned to pairing Sink node; This node itself will write down these data and added one by the counter of inquiry times simultaneously, and whether the number of times that record data are inquired about is to provide the judgment data can be by the foundation of preferential deletion in order to exhaust at memory space.

In order to improve query responding time, when certain data on the node reached predetermined threshold value α by the enquiry frequency counting time, on the leader cluster node at its place bunch, create the copy of these data.This can cause the generation of two kinds of situation: the one, and bunch epicranium is rotated there are data in the back in same bunch a plurality of copies; The 2nd, possibly there is inconsistent situation before the copy of bunch head and the clean copy, for the copy in making bunch is consistent, and the number of control data copy; Each leader cluster node receive from bunch in during the data trnascription of other nodes; Bunch in broadcasting a notification message, notify other nodes that have this data trnascription to abandon these copies, the form of this message is:

Msg

{

MsgType=UpdateReplicaMsg; The definition of // type of message

DataCharacter; The feature description of // data query

ReplicaID; // copy sign is given by the node that generates initial data

}

In bunch other all receive nodes of this message if there is same data trnascription, but the copy sign is inconsistent, then abandons original copy, this operation can keep the renewal of copy.

When the situation of " seemingly-dead " appears in certain node, and the current state of this node is non-leader cluster node, and at first all data trnascriptions upgrade its memory space on this node through deleting.For " seemingly-dead " of leader cluster node, then its data are transferred to bunch head of adjacent clusters.

Claims (4)

1. storage method for sensor network based on multi-node collaboration is characterized in that step is following:

(1) confirm the node cooperation mode: the node at sensor network extracts characteristics of image and identification image, and communicates by letter with neighbor node with the mode of message-driven, to confirm to do the set of node of data fusion;

(2) confirm that storage exhausts the update mode of sensor network nodes: an important optimization to the conventional store distributed storage is the Data Update strategy that has added " seemingly-dead " node, in energy consumption with preserve the quality of data this two and obtain a compromise;

(3) take a kind of load-balancing method: dynamically each Data Update of calculating sensor network node or the only size of data when shifting make equilibrium and energy consumption between different nodes and storage consumption equilibrium between the energy consumption and storage consumption on the same to a certain extent node of this method;

(4) take the layout type of a plurality of Sink nodes:, use the correct position of genetic algorithm for solving Sink node to the deployment general layout at random of Sink node;

(5) building network topology: adopt the mode of sub-clustering commonly used, through the TopDisc algorithm is improved, the network topology structure that makes up stratification is optimized the user inquiring processing procedure;

(6) optimize the data query process:, set up the data query process of message-driven through creating and managing a plurality of data trnascriptions;

Be used to solve more energy consumption that causes of incident and the too fast problem of storage consumption that individual node senses, with energy between different nodes and storage consumption lack of uniformity problem, and the life span that prolongs network.

2. a kind of storage method for sensor network according to claim 1 based on multi-node collaboration; It is characterized in that: the network using message-driven, the image to sensing between adjacent node extracts characteristics of image; Exchange message is in order to accomplish data fusion and data compression; For the node of " seemingly-dead " state of entering, Dynamic Selection adjacent node transferring data, each data length that shifts obtains according to given formula.

3. a kind of storage method for sensor network according to claim 1 based on multi-node collaboration; It is characterized in that: network node is according to spreading at random; A plurality of Sink nodes adopt the genetic algorithm for solving deployed position; Net the data ghost high in stored and the query script in execution, and pass through the TopDisc algorithm sub-clustering management data copy of improved access frequency.

4. a kind of storage method for sensor network based on multi-node collaboration according to claim 1 is characterized in that the building network topology step:

Adopt the mode of sub-clustering commonly used, through the TopDisc algorithm is improved slightly, the network topology structure that makes up stratification is optimized the user inquiring processing procedure; The state of network is divided into: topology constructing stage and data transfer phase;

Topology constructing during the stage node all be in active state, the request data package that is used for Topology Discovery is carried out inundation propagates; At first, agent node sends the message of setting up topology to the Sink node; Then, by the broadcast that starts " discovery neighbor node " from the nearest sensor node of Sink node, each message is all carried a kind of state information; Neighbor node waits for a period of time after receiving message, and this time is inversely proportional to according to distance and the current dump energy of node between itself and the sending node, that is:

Wherein,

is the wait time delay among the TopDisc;

,

are delay constant; D is for sending the distance of query messages, and E is the current energy of node;

Along with the Topology Discovery request message is propagated in network, sensor network is divided into a plurality of bunches (cluster), and each bunch all has a bunch of head; By bunch head be responsible for to Sink communication and in this bunch other nodes send Query Informations, can direct communication between bunch head or communicate through the Sink node, thus formed a sensor network that can cover the hierarchical structure of whole guarded region;

At data transfer phase, except transmitting normal task property data, also regularly around leader cluster node, carry out topology reconstruction, thereby realize rotating of bunch head.