CN106341842B

CN106341842B - Method and device for transmitting wireless sensor network data

Info

Publication number: CN106341842B
Application number: CN201610712279.5A
Authority: CN
Inventors: 尹长川; 梁瀚樱; 邓乔木; 刘丹谱
Original assignee: Beijing University of Posts and Telecommunications
Current assignee: Beijing University of Posts and Telecommunications
Priority date: 2016-08-23
Filing date: 2016-08-23
Publication date: 2020-05-01
Anticipated expiration: 2036-08-23
Also published as: CN106341842A

Abstract

The invention discloses a method and a device for transmitting data of a wireless sensor network, which are applied to the wireless sensor network with a cluster structure. According to the scheme, the data transmission quantity in the network can be effectively reduced on the premise of meeting application requirements, the node energy consumption is reduced, and the network life cycle is prolonged.

Description

Method and device for transmitting wireless sensor network data

Technical Field

The present invention relates to the field of data transmission technologies, and in particular, to a data transmission method and apparatus for a wireless sensor network.

Background

At present, a Wireless Sensor Network (WSN) has been widely applied in the fields of smart home, agricultural planting, environmental monitoring, chemical production, and the like.

A Wireless Sensor Network (WSN) is an ad hoc network formed by a plurality of sensor nodes distributed discretely in space through wireless communication, as shown in fig. 1, which is a schematic structural diagram of a conventional wireless sensor network system. The task management node is connected with the sink node through a communication network, and the sink node is in wireless communication with the sensor node in the target monitoring area.

The sensor nodes cooperate with each other to sense, collect, process and transmit various target information in the coverage area of the network, and provide network connection and data support service for upper layer application and user decision.

In practical applications, in order to ensure the quality of service, the WSN application often monitors a destination area through node intensive deployment. Therefore, data collected by the node needs to be forwarded to reach the destination node through a plurality of intermediate nodes, and the intermediate nodes need to select the most appropriate next hop (next hop) node among the neighbor nodes.

At present, a cluster routing protocol is widely adopted in the WSN to implement data transmission. Enabling sensors according to sensors in routing table

The clustering routing protocol divides the sensor nodes in the network into a plurality of clusters (cluster) by taking data as a center. Each cluster consists of a cluster head node (cluster head) and a plurality of cluster nodes (cluster members), and the cluster head node is not only responsible for the collection and aggregation processing of data in the cluster, but also responsible for the routing forwarding of data between clusters. Typical clustering routing algorithms include LEACH (Low Energy Adaptive clustering hierarchy), GAF (geographic Adaptive fidelity), Cougar protocol, etc., as shown in FIG. 2, which is a schematic diagram of a clustering routing structure.

With the integration and miniaturization of electronic devices, sensor nodes mostly adopt a battery power supply mode with limited energy, so that a path from a source sensor node for sending data to a sink node needs to be reasonably selected by data to reduce network energy consumption.

Considering that the sensor nodes are powered by batteries, continuous supply of node energy is difficult to realize, the storage capacity and the computing capacity of each sensor node are limited, and many application scenes need to meet application requirements through dense distribution of the nodes.

Research has indicated that the energy required for a sensor node to transmit 1 bit of data for 100 meters can perform approximately 3000 calculation instructions. Data collected by all nodes in intensive distribution often have correlation, so that redundancy of some related data is caused, and certain redundancy also exists in data collected by the same node according to time periodicity, so that the data transmission energy is high, and waste of network resources is caused.

Therefore, a technique is needed to solve the problem of high redundancy and data transmission energy consumption during data transmission by using a clustering routing structure.

Disclosure of Invention

Embodiments of the present invention provide a data transmission method and apparatus for a wireless sensor network, which can reduce data redundancy and reduce energy consumed by data transmission, thereby prolonging a life cycle of the network.

In order to achieve the above object, an embodiment of the present invention discloses a method for transmitting data of a wireless sensor network, which is applied to an intra-cluster node in a cluster structure, where the cluster structure further includes a cluster head node, and the method includes the following steps:

s101) determining a prediction model parameter according to historical data, initializing a shift frequency k and a shift threshold j, and transmitting the prediction model parameter to a cluster head node, so that the cluster head node calculates prediction data according to the prediction model parameter and time and stores the prediction data into a data cache corresponding to the node; the prediction model is obtained by curve fitting according to historical data in advance;

s102) data acquisition, wherein the acquired current actual data is compared with current prediction data obtained according to prediction model parameters;

s103) if the difference epsilon' between the current actual data and the predicted data is less than or equal to the error threshold epsilon, caching the actual data, adding 1 to the offset threshold j, and returning to the step S102);

s104) if the difference epsilon' between the current actual data and the predicted data is larger than the error threshold epsilon, caching the actual data, adding 1 to the offset times k, halving the offset threshold j, and transmitting the actual data to the cluster head node; comparing the offset times k with an offset threshold value j, and returning to the step S101 if k is more than or equal to j); if k < j, return to step S102).

Preferably, in the method, the prediction model is a unary linear prediction model, y 'is α t + β, t represents a sampling time point, and y' is a predicted value at a corresponding time;

nodes in a cluster are at the same interval t₀Successive N samples are stored in a buffer queue and parameters are calculated using the data in the time series (α).

The embodiment of the present invention further provides a device for transmitting data of a wireless sensor network, which is applied to an intra-cluster node in a cluster structure, where the cluster structure further includes a cluster head node, where the device includes:

the prediction model parameter transmission module is used for determining a prediction model parameter according to historical data, initializing the offset times k and the offset threshold value j, and transmitting the prediction model parameter to the cluster head node, so that the cluster head node calculates the prediction data according to the prediction model parameter and time and stores the prediction data into a data cache corresponding to the node; the prediction model is obtained by curve fitting according to historical data in advance;

the data acquisition module is used for acquiring data and comparing the acquired current actual data with current prediction data obtained according to the prediction model parameters;

the first cache module is used for caching the actual data under the condition that the difference epsilon' between the current actual data and the predicted data is less than or equal to the error threshold epsilon, adding 1 to the offset threshold j and triggering the data acquisition module;

the second cache module is used for caching the actual data under the condition that the difference epsilon' between the current actual data and the predicted data is larger than the error threshold epsilon, adding 1 to the offset times k, halving the offset threshold j, and transmitting the actual data to the cluster head node; comparing the offset times k with an offset threshold value j, and triggering the prediction model parameter transmission module if k is more than or equal to j; and if k is less than j, triggering the data acquisition module.

The embodiment of the invention also provides a method for receiving the data of the wireless sensor network, which is applied to the cluster head nodes in the cluster structure, the cluster structure also comprises the nodes in the cluster, and the method comprises the following steps:

s201) receiving prediction model parameters sent by nodes in a cluster;

s202) judging whether actual data sent by the nodes in the cluster are received currently;

s203) if yes, storing the data into a data cache corresponding to the node in the cluster;

s204) if not, calculating the prediction data according to the prediction model parameters and time, and storing the prediction data as the data in the cluster into a data cache corresponding to the node in the cluster.

Preferably, the method further comprises a polymerization compression treatment, and the method comprises the following steps:

B1) receiving data uploaded by each node in a cluster, and caching the data of each node in the cluster in an independent cache space;

B2) storing the data uploaded by each node in the cluster in each uploading period as a data matrix X in the cluster, wherein X is [ X ═ X₁,X₂,…,X_n]N is the number of nodes in the cluster, X_i＝[x_i1,x_i2,…,x_in]Data collected for the ith cluster node according to a certain time interval sequence;

B3) performing joint sparse processing on the data matrix X in the clusters, and decomposing the node data in each cluster into a common part and an independent part, namely decomposing X to obtain a common part z_cAnd independent part z thereof_nThe decomposition result is set as Z ═ Z_c,z₁,z₂,…,z_n]；

B4) Respectively carrying out compressed sensing processing on each component data in the decomposition result set to obtain a final compressed aggregation result set R ═ R_c,R₁,R₂,…,R_n]And measurement matrix set Φ ═ Φ [ { Φ_c1,Φ_c2},{Φ₁₁,Φ₁₂},…,{Φ_n1,Φ_n2}]；

B5) And sending the final compression aggregation result set and the measurement matrix set to a sink node or an application terminal along the inter-cluster route.

Preferably, in the method, the independent cache space includes a data cache space and a prediction model parameter cache space;

if the received data is the original sensing data, directly storing the data into a data cache space of the corresponding node; if the data is the prediction model parameter data, updating the prediction model parameter cache of the corresponding node, and calculating the prediction data according to the updated prediction model parameter and time and storing the prediction data into the data cache space of the corresponding node.

Preferably, the method wherein the joint sparseness processing is performed by applying a joint sparseness processing to each intra-cluster node data X_iRespectively using sparse random momentsThe array is sparsely transformed.

Preferably, the method, wherein the sparse random matrix is a bernoulli/mach random matrix, which is defined as follows:

in the above formula, phi_ijRepresenting an element at any position in the Bernoulli/Mach random matrix, wherein p is the value probability of the element, namely the probability p shown by the formula in the element in the matrix is randomly valued in { -1,0, +1}, wherein a parameter s controls the sparsity of the matrix, the larger the value of s is, the smaller the nonzero element in the matrix is, and the higher the sparsity is.

Preferably, the method, wherein the compressed sensing process, comprises the following steps:

A1) according to the number n of nodes in the cluster, the data length m after sensing compression and the sparse control parameter s, calculating the parameter a as m/2s,

Generating an initial null measurement matrix Φ₁、Φ₂；

A2) Generating two random 1 xn dimensional matrices T₁＝[t₁₁,t₁₂,…,t_1n]、T₂＝[t₂₁,t₂₂,…,t_2n]Wherein t is_1nAnd t_2nAre all in [1, b ]]A positive integer of the medium random value, and a matrix T₁、T₂Stored as a row of data in the measurement matrix phi₁、Φ₂In (1).

A3) Finding T₁Adding corresponding position elements in the original signal according to the positions of the elements with the median value i; finding T₂Adding corresponding position elements in the original signal according to the positions of the elements with the median value i; subtracting the second calculated value from the first calculated value, and storing the result in the corresponding compressed data R_iPerforming the following steps;

A4) repeating step a2) and step A3) a times;

A5) repeat steps A3) and a4) with b ═ m-a × b, to obtain the final pressureData of condensation polymerization results R_iAnd a measurement matrix (phi)₁、Φ₂)。

The embodiment of the present invention further provides a receiving device for wireless sensor network data, which is applied to a cluster head node in a cluster structure, wherein the cluster structure further includes an intra-cluster node, and the data receiving device includes:

the receiving module is used for receiving the prediction model parameters sent by the nodes in the cluster;

the judging module is used for judging whether actual data sent by the nodes in the cluster are received at present;

the first storage module is used for storing the actual data into the data cache corresponding to the node in the cluster when the judgment result is yes;

and the second storage module is used for calculating the prediction data according to the prediction model parameters and time and storing the prediction data serving as the intra-cluster data into the data cache corresponding to the intra-cluster node when the judgment result is negative.

Therefore, the scheme of the invention provides an energy-efficient data transmission mechanism based on the existing common clustering routing structure aiming at the condition that the node resources are limited in a plurality of wireless sensor network application scenes. According to the mechanism, aiming at the characteristic that the energy of the nodes is limited, the member nodes in the cluster carry out prediction approximation on the node data by using a prediction model, and the node data transmission is inhibited, so that the energy consumption is reduced. And on the cluster head node, the data in the cluster is subjected to joint compression processing through an improved compression sensing method, so that the data transmission quantity of the cluster head is compressed. According to the scheme, the data transmission quantity in the network can be effectively reduced on the premise of meeting application requirements, so that the energy consumption of the nodes is reduced, and the life cycle of the network is prolonged.

Drawings

In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the embodiments or the prior art will be briefly described below, it is obvious that the drawings in the following description are only some embodiments of the present invention, and for those skilled in the art, other drawings can be obtained according to the drawings without creative efforts.

FIG. 1 is a schematic structural diagram of a conventional wireless sensor network application system;

FIG. 2 is a schematic diagram of a clustering routing structure of a conventional wireless sensor network;

FIG. 3 is a schematic diagram of a compressed sensing measurement process according to an embodiment of the present invention;

fig. 4 is a flow chart of data acquisition and transmission of a node in a cluster according to embodiment 1 of the present invention;

fig. 5 is a schematic diagram of data transmission and model adjustment of nodes in a cluster according to embodiment 1 of the present invention;

fig. 6 is a schematic structural diagram of a data transmission apparatus according to an embodiment of the present invention;

fig. 7 is a data receiving flow chart of a cluster head node according to embodiment 2 of the present invention;

fig. 8 is a schematic structural diagram of a data receiving device according to an embodiment of the present invention.

Detailed Description

The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.

The invention relates to a wireless sensor network based on a clustering routing structure and an improvement of a data transmission process by a data compression sensing theory.

The clustering routing protocol divides the nodes into cluster head nodes and member nodes in the clusters on the basis of plane routing, and determines respective division of labor. The cluster head is responsible for communicating with the members in the cluster, collecting member information and data and being responsible for data forwarding; and the member in the cluster acquires target information in the monitoring area and sends the target information to the cluster head. A periodic cluster rotation mechanism is typically employed to balance energy consumption. The clustering strategy divides the network into different logic sub-areas, has better expandability and can further schedule and process data in the cluster.

The energy of the nodes is mainly consumed in data receiving and transmitting, so the embodiment of the invention realizes the high-efficiency utilization of the energy by compressing the data in the network. The compressed sensing theory is a signal compression and high-precision recovery technology, and is primarily used for processing digital signals in a discrete domain. The theory of compressed sensing states that: if the signal X is ═ X₁,x₂,…,x_n]At a certain transformation base psi ═ psi₁,ψ₂,…,ψ_n]The k-order sparse representation can be performed below, that is, X ═ Ψ S is satisfied, the number of elements in the vector S is much smaller than X, and if the number of elements in S is k, the energy of the original signal X is concentrated on the k nonzero elements, which is referred to as k-order sparse representation of the signal X. The compressed sensing process is shown in fig. 3. When the signal X satisfies the above condition, the passing signal vector Y ═ Y₁,y₂,…,y_m]I.e. recovering X, where Y ═ Φ X ═ Φ Ψ S, the length of the signal vector Y is much smaller than the original signal X, i.e. Y is the result of compressing X, and is called Φ ∈ R^m×nAn observation matrix of X.

In recovering the signal, the convex optimization problem can be solved by:

and obtaining an optimal solution S, and then realizing accurate recovery of the original signal X by using X ═ Ψ S. Solving the above equation requires traversing all non-zero value combinations in S, so that the above equation 0-norm can be converted into 1-norm approximate solution to obtain the following model:

considering error and noise factors, the above equation can be further converted:

most signals are not sparse in the strict sense, but can be sparsely represented in the transform domain by a suitable transform basis Ψ. A gaussian random matrix is often used as a measurement matrix; common sparse transform methods include fourier transform and wavelet transform. The recovery algorithms for the measurement signals include an Orthogonal Matching Pursuit (OMP) algorithm, a Basis Pursuit (BP) algorithm, an adaptive matching pursuit algorithm, and the like, and have advantages and disadvantages.

The invention provides an energy-efficient data aggregation transmission strategy based on a clustering routing structure wireless sensor network and aiming at different division characteristics of cluster head nodes and member nodes in a cluster. In the data transmission strategy, the cluster members and the cluster head nodes can respectively carry out aggregation compression processing on the transmission data according to application requirements, so that the transmission of redundant data in the network is reduced, the energy is saved, and the normal working time of the network is prolonged. It is particularly pointed out that the scheme of the invention mainly aims at the data transmission processing process in the cluster, and does not relate to the route establishment and the data transmission between the clusters. The specific content can be divided into an intra-cluster node processing mechanism and a cluster head node processing mechanism according to the type of the node. The following examples are provided for the purpose of illustrating the present invention.

Example 1

The embodiment provides a cluster node data transmission method of a wireless sensor network based on a clustering structure, as shown in fig. 4:

Specifically, in a common wireless sensor network with a cluster routing structure, an intra-cluster node serving as a network tip is generally only responsible for acquiring target information data in a monitored area and sending the target information data to a relay node of the network tip, and the intra-cluster node is only responsible for itself and does not need to communicate with adjacent intra-cluster nodes in the same area. It can be known through analyzing common environmental objects such as temperature, humidity and the like that such environmental data usually presents periodic change characteristics and presents a linear change rule within a certain time period, that is, the environmental data collected by the nodes has a linear relationship with time. The environmental data collected by the nodes in the cluster in a period of time is linearly related to the time, so that the time can be regarded as an independent variable, the corresponding data is regarded as a dependent variable, and a prediction function is constructed by utilizing a linear regression model to simulate the relation of dynamic changes of the two.

Defining a time sequence as a time data pair set which is sampled by nodes in a cluster according to the time sequence and has a certain time interval, and recording the time data pair set as S { (t)₁,y₁),(t₂,y₂),…,(t_n,y_n)}. Wherein (t)_n,y_n) Is shown at time t_iThe data collected at the moment is y_iAnd n is the sequence length. And establishing a dynamically adjusted piecewise linear prediction model by utilizing the computing and storing capacity of the nodes to carry out approximate estimation on the data acquisition data.

The initialization stage comprises the establishment process of a network clustering routing structure, the connection relation between the nodes in the cluster and the cluster head nodes is established in the initialization stage, and control information including but not limited to an error threshold epsilon, an offset threshold j, a cache length, a collection target, a collection period, a data uploading period and the like is received.

And then the intra-cluster nodes enter a data acquisition and transmission stage, which can be detailed into steps of constructing a prediction model, transmitting data and adjusting the model.

Predictive model building

Considering that the computing and storing capacity of the nodes is limited, establishing a unary linear prediction model at the nodes in the cluster:

y′＝αt+β

where t denotes a sampling time point, y' denotes a predicted value at the corresponding time, and (α) denotes a parameter.

According to the cache length, the nodes are at the same interval t₀The N consecutive samples are stored in a buffer sequence, and a parameter is calculated using data in the buffer sequence (α), and the intra-cluster node sends the parameter to the cluster head node, which then collects data that should be distributed around the prediction function along the time axis.

Data transfer and model adaptation

Referring to fig. 5, in each data acquisition and transmission process, the node calculates a difference value epsilon 'between actually acquired data y and a prediction estimation value y', compares the difference value epsilon 'with a preset error threshold epsilon, when the acquired data changes slowly, namely epsilon' < epsilon, the intra-cluster node does not transmit the data, and the cluster head calculates approximate data of the time point according to the existing prediction model parameters; when the sensing data changes violently, namely epsilon' > epsilon, the data is sent to the cluster head in real time.

The node can monitor the data change condition of the node while transmitting data, and if the prediction model deviates from the real sampling for a long time, the explanation parameter is not applicable and needs to be adjusted. The number of times the error e' exceeds the threshold value e is compared to an offset threshold to determine whether the parameter needs to be adjusted. The offset threshold is not a fixed value, and when each comparison is carried out, if the predicted value meets the requirement, the offset threshold is added with 1, otherwise, the offset threshold is halved.

To prevent the generation of "false dead nodes", at a time period T (T ═ T)₀Xn), the node in the cluster needs to forcibly send data to the cluster head node once, in order to make the new parameter calculated each time satisfy the current state, the time sequence needs to be updated in real time, so in the node software design, a buffer space is maintained to store the time sequence, as the data acquisition progresses, the model parameter at the current time is assumed to be (α), the buffer sequence is Sc, the model offset frequency is k,the offset threshold is j, and the data transmission and model adjustment method comprises the following steps:

step 1:

node acquisition data y_nAccording to the time point t_nComputing prediction data y with parameters (α)_n'and prediction error ε'

if(ε′<ε)

Will y_nAdding the Sc length into the tail of the queue, and removing the head of the queue if the Sc length is full;

j＝j+1；

go to step 1

if(ε′>ε)

k＝k+1；

j＝j/2；

the node sends sensing data;

if(k≥j)

turning to the step 2;

step 2:

k＝0；

and j is 1, (of course, the offset threshold j can also be initialized to any constant which is not 0) a new parameter is calculated by using the data in the Sc according to the least square method (α), and the new parameter is sent to the cluster head node, and the process goes to step 1.

By the method of the scheme, the nodes in the cluster can perform prediction simulation on the acquired data without transmitting all the acquired data, so that the data transmission amount is effectively reduced, the energy consumption is reduced, and the service life of the nodes is prolonged.

Next, referring to fig. 7, a data transmission apparatus of a wireless sensor network according to the present invention is described, where the data transmission apparatus is applied to an intra-cluster node in a cluster structure, and the cluster structure further includes a cluster head node. The data transmission device comprises a prediction model parameter transmission module 101, a data acquisition module 102, a first cache module 103 and a second cache module 104.

The prediction model parameter transmission module 101 is configured to determine a prediction model parameter according to historical data, initialize an offset number k and an offset threshold j, and transmit the prediction model parameter to a cluster head node, so that the cluster head node calculates prediction data according to the prediction model parameter and time and stores the prediction data in a data cache corresponding to the node; the prediction model is obtained by curve fitting according to historical data in advance;

the data acquisition module 102 is configured to acquire data and compare the acquired current actual data with current predicted data obtained according to the prediction model parameters;

the first cache module 103 is configured to cache the actual data when a difference epsilon' between the current actual data and the predicted data is less than or equal to an error threshold epsilon, add 1 to an offset threshold j, and trigger the data acquisition module;

the second cache module 104 is configured to cache the actual data when a difference epsilon' between the current actual data and the predicted data is greater than an error threshold epsilon, add 1 to the offset number k, reduce a threshold j by half, and transmit the actual data to the cluster head node; comparing the offset times k with an offset threshold value j, and triggering the prediction model parameter transmission module if k is more than or equal to j; and if k is less than j, triggering the data acquisition module.

Example 2

The present embodiment provides a method for receiving data of a cluster head node of a wireless sensor network based on a clustering structure, as shown in fig. 6, for a method for transmitting data of a node in a cluster provided in embodiment 1:

s201) receiving prediction model parameters sent by nodes in a cluster;

Specifically, in a wireless sensor network based on a cluster structure, each cluster is equivalent to an independent network node area, and a cluster head node is used as a manager and a coordinator of the area, so that the data forwarding amount is more than that of member nodes in the cluster, and the energy consumption is larger.

In order to reduce the load pressure of the cluster head nodes and reduce the energy consumption brought by data transmission, the technical scheme of the invention further performs aggregation compression processing on the data at the cluster head nodes. As shown in fig. 6, the method further includes step S205) of performing aggregation compression processing on the data stored in each preset period in the cache data.

In the initialization stage, the cluster head node needs to complete the establishment of the inter-cluster route, establish connection with the intra-cluster node and allocate a time slot, and meanwhile, receive control information of the application or the aggregation node, including but not limited to an upload period, destination data, an application tolerance threshold, and the like. The required completion work at this stage is closely related to the specific application scenario. After completing the node initialization, the cluster head node will start waiting to receive the data in the cluster.

As can be seen from embodiment 1, the intra-cluster nodes perform predictive approximation on the acquired data through the predictive model, thereby reducing the data transmission amount. The cluster head node needs to provide a cache space for each node in the cluster to store the model parameter data and the sensing data sent by the node respectively, and meanwhile, the cluster head needs to keep clock synchronization with the nodes in the cluster.

The cluster head node receives data of each node in the cluster, and if the received data is original sensing data, the data is directly stored in a data cache of a corresponding node; if the data is prediction model parameter data, the cluster head node firstly updates the parameter cache corresponding to the node in the cluster, and then calculates the prediction data according to the new parameters and time and stores the prediction data into the data cache corresponding to the node. At each uploading time point, the cluster head node needs to send the data cache of all the nodes to the sink node, and meanwhile, the cache is cleared.

The embodiment of the invention provides a new compression processing method for carrying out combined compression and aggregation on the data in the cluster aiming at the problem of overlarge data volume sent by the cluster head, thereby reducing the data volume sent by the cluster head node and saving energy consumption.

Improved data compression processing method

Compressed sensing can ensure a high probability of reconstructing an original signal under the condition of reducing the number of samples, and can be used for compressing data in a wireless sensor network. However, because the resources of the sensor nodes are limited, the multiplication operation of the measurement matrix and the original data is required to be processed for carrying out compressed sensing, and the requirements on the storage and calculation capacity of the sensor nodes are high. The solution of the invention proposes a suitable improvement in respect of this problem.

At present, a gaussian random measurement matrix is widely applied to a compressed sensing theory, but a random number generator is needed to generate m × n random numbers meeting requirements in the implementation process. The process of generating the measurement matrix and multiplying the matrix by the original data is high in complexity and is not suitable for the sensor node with limited resources.

According to the scheme, the main information of compressible data is acquired by adopting a sparse random matrix to make up the defects, and the Bernoulli/Mach random matrix is used as a common sparse matrix and is defined as follows:

in the above formula, phi_ijRepresenting an element at any position in the Bernoulli/Mach random matrix, wherein p is the value probability of the element, namely the probability p shown by the formula in the element in the matrix is randomly valued in { -1,0, +1}, wherein a parameter s controls the sparsity of the matrix, the larger the value of s is, the smaller the nonzero element in the matrix is, and the higher the sparsity is. If the original data and the compressed data are n and m in length, respectively, then the method still needs to generate and store an mxn sparse matrix.

The scheme of the invention divides the sparse random matrix according to the positive and negative of elements, then generates random numbers in a blocking mode to represent the coordinates of non-zero elements in the measurement matrix, and combines the coordinates into the improved sparse random measurement matrix phi₁、Φ₂Two matrices, phi₁、Φ₂Instead of storing the actual measurement matrix element dataThe position of its non-zero element is stored. Meanwhile, matrix multiplication is reduced to addition operation, the compressed sensing measurement process of the original signal is completed while the measurement matrix is generated, the complexity of data processing is reduced, and the real-time performance of the wireless sensor network system is improved.

Assuming that the data length after the compressed sensing processing is performed on the data X with the length n is m, the improved compressed sensing processing procedure is as follows:

(1) ignoring the normalization operator, taking the positive element and the negative element of each column to be +/-1, dividing each column into a groups, satisfying that a is m/2s, and calculating

Initial phi₁、Φ₂Are all empty matrices.

(2) Generating two random 1 xn dimensional matrices T₁＝[t₁₁,t₁₂,…,t_1n]、T₂＝[t₂₁,t₂₂,…,t_2n]，T₁And T₂Representing the position of the element in the measurement matrix. Wherein t is_1nAnd t_2nAre all in [1, b ]]The positive integer of the medium random value needs to make t_1nAnd t_2nThe same value is not taken, and each column of each block matrix is ensured to have only one positive element and one negative element. Then, the matrix T is divided₁、T₂Stored as a row of data in the matrix phi₁、Φ₂Among them.

(3) The raw data X is then subjected to a cyclic sensing process. In each process, first, T is found₁The median value is the position of the i element, the values of the corresponding position elements in the original signal X are added, and then the T is found out₂The median is the position of the i element, the values of the corresponding position elements in the original signal X are added, and finally the two results are subtracted and multiplied

Obtaining the result r_iAnd the result r is_iAnd storing the final compressed data R. i is initialized to 1 and accumulated after each cycle by 1, i equals b for the last processing.

(4) Repeating the process 2 and the process 3 for a times;

(5) finally, making b be m-a x b, making operation according to processes (3) and (4) and storing the result in R_iIn (1).

Through the above process, the data R obtained after compressing the original data X and the improved measurement matrix (Φ) corresponding to the data R are obtained₁、Φ₂)。

According to the process, the nodes only need a uniform random number generator and a storage space of 2(a +1) x n, and the space complexity is reduced; the measurement process of the original signal is also reduced from matrix multiplication to addition, and the time complexity is reduced; the generation of the measurement matrix is combined with the process of data perception measurement, and the efficiency is higher.

Cluster head processing flow

The distributed compressed sensing theory considers that for a plurality of signals or data, a joint sparse model can be constructed by utilizing the time domain sparsity and the spatial domain correlation of the signals or data, and joint coding and joint decoding recovery are carried out, so that the distributed compressed sensing theory has a better energy consumption characteristic than independent signal processing and can better recover the original signals.

Because each node in the cluster has continuity in geographic position, the target data collected by each node can be considered to have a certain correlation. The scheme of the invention decomposes the data of each node in the cluster into a common part and a unique independent part contained in all the nodes. Assuming that the data of the ith sensor node in the cluster cached by the cluster head in each uploading period is X_i＝[x_i1,x_i2,…,x_in]，X_i＝[x_i1,x_i2,…,x_in]For data collected by the ith cluster node sequentially according to a certain time interval, then all data cached by the cluster head in the time period is X ═ X₁,X₂,…,X_n]And n is the number of nodes in the cluster. Since not every intra-cluster node will send sensing data to the cluster head every sampling period, this reduces intra-cluster communication overhead accordingly.

The data processing of the cluster head node in combination with the data aggregation process of the nodes in the cluster is specifically described as follows:

(1) the cluster head nodes always keep working state and prepare to receive the data of the nodes in the cluster, and corresponding data cache space is distributed for each node in the cluster. The data sent by the nodes in the cluster are divided into two types, namely prediction model parameters and real sensing data.

(2) When the cluster head receives data, if the data is parameter data, updating the prediction model parameter cache of the corresponding node, and calculating the prediction data of the corresponding moment and storing the prediction data into the data cache; and if the sensing data is the sensing data, directly storing the sensing data into the data cache of the corresponding node.

(3) Every time the uploading period T passes, the cluster head generates the cache data into a signal vector X according to the node id in the cluster_i＝[x_1i,x_2i,…,x_ni]^TIf the number of nodes in the cluster is n, the data matrix stored by the cluster head is X ═ X₁,X₂,…,X_n]。

(4) And performing joint sparse processing on the data matrix X stored by the cluster head by using the distributed compressed sensing theory and the common sparse transformation method. The joint sparsification is performed in order to minimize the total sparsity of the transformed data, so that the optimization condition needs to be satisfied:

min(||z_c||₀+||z₁||₀+||z₂||₀+…+||z_n||₀)i＝1,2,…n

according to the above conditions to X_iThe data after sparse transform is a_i＝[a_1i,a_2i,…,a_ni]^T，a_i＝z_c+z_nWherein z is_cAs a common part of all data, z_nIs a_iA unique independent part.

(5) The common and independent portions of each signal may be jointly represented as Z ═ Z_c,z₁,z₂,…,z_n]And determining the length of the data after the perception compression according to the sparsity of each component. The data compression processing method is used for compressing the sparse data to obtain the final compressed aggregated data of R ═ R_c,R₁,R₂,…,R_n]And an improved measurement matrix phi [ { phi ] corresponding thereto_c1,Φ_c2},{Φ₁₁,Φ₁₂},…,{Φ_n1,Φ_n2}]。

And the cluster head node sends the compressed aggregation data and the measurement matrix data to the sink node in each uploading period T, and then the sink node recovers the data through a compressed sensing recovery algorithm.

Example 3

The embodiment takes agricultural internet of things application as an example. In a large farmland application scene, the growth environment state of crops needs to be mastered to realize more scientific crop cultivation. Data such as temperature, humidity, illumination and the like are collected through wireless sensor nodes to provide data support for scientific fertilization and efficient irrigation, and green agriculture is realized.

In the application scenes, the precision requirement on environmental parameter data such as temperature and humidity cannot be too high, and certain error tolerance can be realized. Because the sensor nodes placed in the farmland adopt a battery or solar power supply mode, and the cost of the nodes is not too high, the energy and the calculation storage capacity of the nodes are limited. By adopting the scheme of the invention, on the premise of meeting the application requirements, the energy consumption caused by data transmission can be reduced, the working time of each node in the farmland can be prolonged, and the production cost can be saved.

The user determines a target object to be monitored through the terminal and acquires object data in a target farmland collected by the sensor nodes, wherein the target object comprises but is not limited to temperature, humidity, illumination intensity, concentration of a certain substance in soil and the like. The specific application process is as follows:

① the sensor node in the target area completes the construction process of the cluster structure and completes the establishment of the inter-cluster route, and the sensor node reports its position information, ID, etc. to the sink node

② each cluster head node constructs a cache space for all nodes in the cluster.

③ the user sends control information to the sink node, including but not limited to the collection target, collection time interval, upload period, data error threshold, cluster head node rotation period, node time sequence length, etc.

④ the sink node sends control information to each sensor node and feeds back node information to the user terminal.

⑤ sensor nodes start data collection, each node constructs prediction model parameters according to the time sequence length by using the initial collected data, and sends the parameters to the cluster head node.

⑥ the cluster head node keeps receiving the data in the cluster, and when each uploading cycle arrives, the cluster head node carries out the aggregation compression processing to the data in the cluster according to the cluster head processing mechanism, and sends the data to the aggregation node along the inter-cluster route.

⑦ the sink node can send the data to the user end immediately after receiving the data, and the data can be restored by the user end, or can send the data to the user end after the sink node decompresses and restores the data.

⑧ every time a cluster rotation period is passed, the wireless sensor network in the target area needs to re-elect cluster heads, so that the energy is uniformly consumed.

⑨ the user can adjust the control parameters such as error threshold according to the specific requirement.

Next, referring to fig. 8, a data transmission apparatus of a wireless sensor network according to the present invention is described, where the data reception apparatus is applied to a cluster head node in a cluster structure, and the cluster structure further includes an intra-cluster node. The data transmission device comprises a receiving module 201, a judging module 202, a first storage module 203 and a second storage module 204.

The receiving module 201 is configured to receive a prediction model parameter sent by a node in a cluster;

the determining module 202 is configured to determine whether actual data sent by a node in a cluster is currently received;

the first storage module 203 is configured to, when the determination result is yes, store the actual data in the data cache corresponding to the node in the cluster;

and the second storage module 204 is configured to, when the determination result is negative, calculate prediction data according to the prediction model parameter and time, and store the prediction data as intra-cluster data in a data cache corresponding to the intra-cluster node.

It can be seen from the above embodiments that, in the solution of the present invention, for the case of limited node resources in the application of the wireless sensor network, based on the existing cluster routing structure, an energy efficient data transmission method is provided. And aiming at member nodes in the cluster, the node data is predicted and approximated by using a prediction model, and the transmission of the node data is inhibited, so that the energy consumption is reduced. And aiming at the cluster head nodes, the data in the cluster are subjected to joint compression processing by an improved compression sensing method, so that the data transmission quantity of the cluster heads is reduced. According to the scheme, the data transmission quantity in the network can be effectively reduced on the premise of meeting application requirements, so that the energy consumption of the nodes is reduced, and the life cycle of the network is prolonged.

It is noted that, herein, relational terms such as first and second, and the like may be used solely to distinguish one entity or action from another entity or action without necessarily requiring or implying any actual such relationship or order between such entities or actions. Also, the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but may include other elements not expressly listed or inherent to such process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising an … …" does not exclude the presence of other identical elements in a process, method, article, or apparatus that comprises the element.

All the embodiments in the present specification are described in a related manner, and the same and similar parts among the embodiments may be referred to each other, and each embodiment focuses on the differences from the other embodiments. In particular, for the system embodiment, since it is substantially similar to the method embodiment, the description is simple, and for the relevant points, reference may be made to the partial description of the method embodiment.

The above description is only for the preferred embodiment of the present invention, and is not intended to limit the scope of the present invention. Any modification, equivalent replacement, or improvement made within the spirit and principle of the present invention shall fall within the protection scope of the present invention.

Claims

1. A method for transmitting data of a wireless sensor network is applied to intra-cluster nodes in a cluster structure, the cluster structure further comprises cluster head nodes, and the method is characterized by comprising the following steps:

s101) determining a prediction model parameter according to historical data, initializing a shift frequency k and a shift threshold j, and transmitting the prediction model parameter to a cluster head node, so that the cluster head node calculates prediction data according to the prediction model parameter and time and stores the prediction data into a data cache corresponding to a node in the cluster; the prediction model is obtained by curve fitting according to historical data in advance;

2. The method of claim 1, wherein the predictive model is a unary linear predictive model, y' α t + β

Wherein t represents a sampling time point, and y' is a predicted value at a corresponding moment;

nodes in a cluster are at the same interval t₀Storing consecutive N samples into a buffer queueIn the method (8), a parameter is calculated using data in a time series according to a least square method (α).

3. The utility model provides a transmission device of wireless sensor network data, is applied to the interior node of cluster in the cluster structure, still include cluster head node in the cluster structure, its characterized in that includes:

the prediction model parameter transmission module is used for determining a prediction model parameter according to historical data, initializing the offset times k and the offset threshold value j, and transmitting the prediction model parameter to the cluster head node, so that the cluster head node calculates the prediction data according to the prediction model parameter and time and stores the prediction data into a data cache corresponding to the node in the cluster; the prediction model is obtained by curve fitting according to historical data in advance;

4. A method for receiving wireless sensor network data is applied to cluster head nodes in a cluster structure, the cluster structure further comprises cluster internal nodes, and the method is characterized by comprising the following steps:

s201) receiving prediction model parameters sent by nodes in a cluster;

s203) if yes, storing the actual data into a data cache corresponding to the node in the cluster;

s204) if not, calculating prediction data according to the prediction model parameters and time, and storing the prediction data serving as intra-cluster data into a data cache corresponding to the intra-cluster node;

the method also comprises an aggregation compression treatment, comprising the following steps:

B5) And sending the final aggregation compression result set and the measurement matrix set to a sink node or an application terminal along the inter-cluster route.

5. The method of claim 4, wherein the independent cache spaces comprise a data cache space and a prediction model parameter cache space;

6. The method of claim 4, wherein the joint sparseness processing is performed on each intra-cluster node data X_iAnd respectively carrying out sparse transformation by adopting sparse random matrixes.

7. The method of claim 6, wherein the sparse random matrix is a Bernoulli/Mach random matrix defined as follows:

8. The method according to claim 4, wherein the compressed sensing process comprises the steps of:

Generating an initial null measurement matrix Φ₁、Φ₂；

A2) Generating two random 1 xn dimensional matrices T₁＝[t₁₁,t₁₂,…,t_1n]、T₂＝[t₂₁,t₂₂,…,t_2n]Wherein t is_1nAnd t_2nAre all in [1, b ]]Middle random valuePositive integer of (1), will matrix T₁、T₂Stored as a row of data in the measurement matrix phi₁、Φ₂Performing the following steps;

A4) repeating step a2) and step A3) a times;

A5) repeating steps A3) and a4) with b being m-a × b, to obtain final compressed polymerization result data R_iAnd a measurement matrix (phi)₁、Φ₂)。

9. The utility model provides a receiving arrangement of wireless sensor network data, is applied to the cluster head node in the cluster structure, still include node in the cluster structure, its characterized in that includes:

the second storage module is used for calculating the prediction data according to the prediction model parameters and time and storing the prediction data serving as the intra-cluster data into the data cache corresponding to the intra-cluster node when the judgment result is negative;

the device further comprises: the aggregation compression processing module is specifically used for: