A data collection method for optoelectronic-radar sensor networks based on equal clustering

By adopting a compressed data collection method based on equal clustering and a moving collector, the problems of uneven energy consumption and data redundancy in cluster heads in WSN are solved, achieving energy balance and efficient data collection.

CN115987867BActive Publication Date: 2024-11-19THE 54TH RESEARCH INSTITUTE OF CHINA ELECTRONICS TECHNOLOGY GROUP CORPORATION
View PDF 5 Cites 0 Cited by

Patent Information

Application Number
CN202211660745.1
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2022-12-23
Publication Date
2024-11-19
Estimated Expiration
2042-12-23

AI Technical Summary

Technical Problem

In existing WSN compressed data collection algorithms based on clustered topology, the compression encoding process cannot effectively utilize network routing information, resulting in uneven energy consumption at the cluster heads and data redundancy, especially in networks with low node density where energy consumption is excessive.

Method used

A compressed data collection method based on equal clustering and a moving collector is adopted. The observation matrix is ​​constructed by equal clustering algorithm, cluster heads and candidate cluster heads are selected, routes are constructed using local data forwarding tree, and the shortest path is planned by moving collector. The original readings are recovered by combining compressed sensing technology.

Benefits of technology

It reduces cluster head energy consumption, reduces the number of nodes accessed by the mobile collector, balances network energy consumption, reduces data redundancy, and improves data collection efficiency.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN115987867B_ABST
    Figure CN115987867B_ABST
Patent Text Reader

Abstract

The present invention discloses a method for collecting data of an optoelectronic radar sensor network based on equal clustering, and belongs to the field of network optimization. The present invention comprises: setting a Sink node and N ordinary sensor nodes; dividing the network into M clusters, and constructing an observation matrix; selecting a cluster head and an alternative cluster head; constructing a route between ordinary nodes involved in each row of the observation matrix; multiplying the node reading by the weight according to the route and adding the weighted sum of all the previous hops sent by the node to obtain a new result, and sending it to the next hop node; collecting the observation values ​​at all cluster heads according to the set path and transmitting them to the Sink; the Sink node restores the original reading of the sensor node. The present invention solves the problem that the compression encoding process in the WSN compression data collection algorithm based on clustering topology cannot effectively utilize the network routing information, reduces the energy consumption of the cluster head, and reduces the number of nodes that the mobile collector needs to visit.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention belongs to the field of network optimization design. Its value orientation is aimed at the specific border geographical environment characteristics and the complex and changeable terrain where the nodes are located, and proposes a compressed data collection method based on equal clustering and mobile collectors. Background Art

[0002] The wireless sensor network for border protection is data-centric and can collaboratively monitor, perceive and collect information from various environments or monitored objects in real time, and transmit the processed information to the base station. It is an effective means of border protection and monitoring.

[0003] In view of the complex terrain environment of the border, some nodes are distributed among the mountains and some nodes are distributed in the vast plains. Therefore, when collecting data from each node, the WSN (wireless sensor network) data collection algorithm based on the cluster topology can usually balance the energy consumption between nodes well and is more robust. In the commonly used cluster routing algorithm that does not use CS (compressed sensing), the cluster head needs to collect the raw data perceived by all nodes in the cluster, and then complete the data convergence to the Sink through multi-hop transmission or mobile data collection technology. This data collection method requires the cluster head to undertake more data transmission and reception tasks, consume more energy, and thus cause the imbalance of energy consumption between nodes, which is the so-called energy black hole problem. Secondly, the cluster head simply collects the data of the nodes in the cluster, and does not use the spatiotemporal correlation of the data of the nodes in the same cluster, resulting in data redundancy and increased network energy consumption. Therefore, some studies have combined compressed sensing technology with clustering algorithms to compress and transmit data. The most typical one is the HcsCluster algorithm. This algorithm stores an observation matrix at each cluster head, which is responsible for compressing and measuring the data of nodes in the cluster. Then, the nodes in the cluster do not process the data and send it to the cluster head. Finally, the cluster head compresses and encodes the received data and transmits it to the Sink. The Sink restores the original node data based on the received observations. From the above process, it can be found that HcsCluster does not compress the data transmitted within the cluster, so it still has problems of data redundancy and uneven energy consumption, and it does not consider the energy consumption problem in the process of observation aggregation. Although another compressed data collection algorithm, the MSTP algorithm, can use CS to compress all node data, it needs to construct matching routes based on the observation matrix, and then use the routes to complete the encoding process. For networks with relatively small node density, the compression encoding process will consume too much energy. Summary of the invention

[0004] In view of this, the present invention proposes a compressed data collection method based on equal clustering and mobile collectors. The present invention solves the problem that the compression encoding process in the WSN compressed data collection algorithm based on clustering topology cannot effectively utilize network routing information, reduces the energy consumption of cluster heads, and reduces the number of nodes that mobile collectors need to visit.

[0005] The technical solution adopted by the present invention is:

[0006] A method for collecting data from an optoelectronic-radar sensor network based on equal clustering comprises the following steps:

[0007] Step 1: Set up a Sink node responsible for collecting data and N ordinary sensor nodes;

[0008] Step 2: According to the location distribution of nodes, the network is divided into M clusters using an equal clustering algorithm, and an observation matrix is ​​constructed based on the clustering information;

[0009] Step 3: Use the distributed cluster head selection algorithm for M clusters to select cluster heads and candidate cluster heads and prepare for a new round of data collection;

[0010] Step 4: Based on the node location information and the cluster head determined in the previous step, the local data forwarding tree construction algorithm is used to construct the routing between the common nodes involved in each row of the observation matrix;

[0011] Step 5: Each node multiplies its own reading by the weight according to the constructed route, and then adds the weighted sum of all the previous hops to get a new result, and sends this result to the next hop node in the route;

[0012] Step 6: Use the mobile collector shortest path planning algorithm to plan the collector's moving path, and then collect the observations at all cluster heads according to the set path and transmit them to the Sink;

[0013] Step 7: The Sink node uses the collected observations and observation matrix Φ as well as the compressed sensing restoration algorithm to restore the original readings of the sensor node.

[0014] Furthermore, in step 1, it is assumed that the sensor distribution area is a narrow area with a side length of A*B, and the Sink node is located in the center of the area; all sensors in the target area are responsible for detecting the target data, and each ordinary node will perceive a reading x in each round of data collection. i , Sink node collects all node data, X=[x1,x2,...,x N ] T .

[0015] Furthermore, step 2 is divided into two stages:

[0016] The first stage: clustering based on the agglomerative hierarchical clustering method AGNES. When the network is in the initial state, the number of clusters in the network is greater than the required M. In each subsequent round, the two clusters with the highest similarity are merged, and the number of clusters is reduced by one until only M clusters remain, and this stage ends. In the wireless sensor network model, the similarity between clusters is defined as follows: as long as the shortest distance between two clusters is less than the maximum communication distance of the nodes, the two clusters are considered to be connected. Therefore, the following formula is selected as the definition of similarity:

[0017]

[0018] Among them, C i and C j denote the i-th and j-th clusters respectively, and x and z denote the nodes in the corresponding clusters;

[0019] The steps of the first stage of the AGNES-based equal clustering algorithm are as follows:

[0020] (1) Network initialization, each node forms a cluster C i , i=1,2,…,N;

[0021] (2) According to formula (1), the similarity between any two clusters is calculated, that is, C i ,C j The minimum distance between any two nodes in the cluster is used to construct the similarity matrix between clusters;

[0022] (3) Find the two clusters C with the highest similarity in the similarity matrix m and C n , merge the two clusters, namely C mn =C m ∪C n , and obtain a new cluster C mn , the number of clusters is reduced by 1, and the other clusters and C are recalculated mn The similarity between them is calculated and the similarity matrix is ​​updated;

[0023] (4) Repeat step (3) until the number of clusters in the network reaches M, ending the first stage of clustering process;

[0024] The second stage: adjust the number of nodes in each cluster; the specific method is:

[0025] Define a restriction level for each cluster, which is initially 0. When a cluster borrows a node from another cluster, the restriction level of the cluster increases by 1.

[0026] For every cluster C containing less than N / M nodes i, find all nodes that are within the communication range of the cluster and do not belong to the cluster, add these nodes to a set Q, and sort them from small to large according to the distance to the cluster;

[0027] If there is a node k in Q, and the cluster C to which the node k belongs k If the number of nodes is greater than N / M, then node k is moved from C k Remove and join cluster C i Otherwise, if there is a node k in Q, the cluster C to which it belongs k The restriction level is 0, then k is also changed from C k Move into C i , and change C i Otherwise, find the node k in Q to which cluster C belongs. k The restriction level is the smallest, and k is also reduced from C k Move into C i , and change C i the level of restriction;

[0028] Repeat the above steps until the number of nodes in all clusters is greater than N / M and the number of cluster nodes is basically the same.

[0029] Furthermore, the specific method of step 3 is:

[0030] The ratio of the remaining energy and the distance to the sink is defined as an indicator Ed. As shown in formula (2), the possibility of a node being elected as a cluster head increases as Ed increases:

[0031]

[0032] In order to find the two nodes with the largest Ed value in the cluster, we first define the data packet P sent by the node during the cluster head selection process. i As shown below:

[0033]

[0034] Then the selection process of cluster head and backup cluster head is completed through the following steps:

[0035] (1) Randomly select a node ClusterTmp in the cluster to serve as a temporary cluster head, and use the minimum spanning tree algorithm in graph theory to build a data forwarding tree with ClusterTmp as the root node;

[0036] (2) Each node in the cluster calculates its own Ed value according to formula (2), and sets the corresponding Ed of node 1 in data packet P to its own sequence number and the calculated Ed value, and sets the corresponding Ed of node 2 to -1 and 0 respectively;

[0037] (3) The leaf node of the data forwarding tree sends a data packet P to its next-hop node. The node that receives the data packet compares the two received Ed values ​​with its own stored Ed value, selects two nodes with the largest Ed, and fills them into the data packet P in sequence. When the middle non-leaf node receives all the data packets sent by the previous-hop node, it sends its updated data packet to the next-hop node.

[0038] (4) After receiving all the data packets P, the temporary cluster head node ClusterTmp obtains the final cluster head node and the temporary cluster head node and the corresponding Ed value. It notifies all nodes of the new cluster head and the information of the candidate cluster heads by broadcasting.

[0039] Furthermore, the specific method of step 4 is:

[0040] (1) Using the minimum spanning tree algorithm in graph theory, a data forwarding tree with the projection node as the root node is constructed to complete the construction of some routes. At this time, some isolated nodes cannot be added to the tree.

[0041] (2) For each isolated node, find the node closest to it that has been added to the forwarding tree, and use the Dijkstra algorithm to find the shortest path between the two.

[0042] Furthermore, the specific method of step 5 is:

[0043] Each node sends its own reading x i The non-zero elements Φ corresponding to the observation matrix Φ ij Multiply them to get Φ ij x i , add this value to the weighted sum data sent by all previous hop nodes of this node to get a new result, and then forward this result to the next node until the data converges to the cluster head to complete the collection of an observation value; the observation matrix has M rows, corresponding to M observation values.

[0044] Furthermore, the specific method of step 6 is:

[0045] (1) Use the simulated annealing algorithm to obtain the original moving trajectory S; the specific method is:

[0046] ① Initialization: Set the initial temperature T and the number of iterations required for each temperature T, and randomly generate an initial path S;

[0047] ② Randomly construct a new path S1;

[0048] ③ According to the evaluation function C, calculate the difference between the new solution and the original solution ΔT = C(S1)-C(S), where C is the summation function of the trajectory length;

[0049] ④If ΔT<0, then accept S1 as the current solution, otherwise Accept it as the current solution;

[0050] ⑤ Repeat steps 2 to 4 until the number of iterations is reached;

[0051] ⑥Reduce the temperature value T = 0.99*T, and end when T is less than 0.001;

[0052] (2) Use the sensor node communication model to further reduce the length of the trajectory S and obtain the final trajectory S*; the specific method is:

[0053] ⑦ Initialization: start and end are the first and last node numbers of the trajectory S, the new trajectory S*={start,end}, k=start+1;

[0054] ⑧ Connect start and k, and let the line segment between these two points be L;

[0055] ⑨ Determine the intersection point P of the line segment L to the communication radius circle of k. If the intersection point exists, set start = P, k = k + 1, and add start to S*; if the intersection point does not exist, determine whether L is less than the communication radius. If so, add start to S*, start remains unchanged, k = k + 1, otherwise start = k, k = k + 1, and start is added to S*;

[0056] If start = end, the algorithm ends, otherwise jump to step ⑧.

[0057] The present invention has the following beneficial effects:

[0058] 1. The present invention combines the clustering of the network with the construction of the observation matrix. The cluster head of each cluster collects the compressed data within the cluster and combines data aggregation with data compression to reduce the data forwarding amount of ordinary nodes in the cluster.

[0059] 2. In order to balance the energy consumption within the cluster, the present invention takes into account the node energy and node spatial location, and proposes a cluster head rotation and cluster head backup mechanism.

[0060] 3. For a large-scale sensor network, the observation values ​​of the cluster heads are aggregated to the base station, which consumes a lot of energy. To this end, the present invention proposes a mobile data collection strategy based on regional division according to the density of node distribution.

[0061] 4. The present invention solves the problem that the compression encoding process in the WSN compression data collection algorithm based on cluster topology cannot effectively utilize network routing information, reduces the energy consumption of the cluster head, and reduces the number of nodes that the mobile collector needs to visit. BRIEF DESCRIPTION OF THE DRAWINGS

[0062] Figure 1 Schematic diagram of the node clustering and observation matrix construction process. DETAILED DESCRIPTION

[0063] In order to make the objectives, technical solutions and advantages of the present invention more clearly understood, the present invention is further described in detail below with reference to the accompanying drawings.

[0064] A compressed data collection method based on equal clustering and mobile collectors comprises the following steps:

[0065] Step 1: Network initialization. Set up a Sink node (can be a fixed collector or a mobile collector) responsible for collecting data and N ordinary sensor nodes. Assume that the sensor distribution area is a narrow area with a side length of A*B, and the Sink node is located in the center of the area. All sensors in the target area are responsible for detecting the target data. Each ordinary node will perceive a reading x in each round of data collection. i , the Sink node needs to collect all node data, X = [x1, x2, ..., x N ] T Due to the spatiotemporal correlation of node data, the original data X is sparse under an appropriate sparse basis.

[0066] Step 2: According to the location distribution of nodes, the network is divided into M clusters using an equal clustering algorithm, and an observation matrix is ​​constructed based on the clustering information.

[0067] The first stage is clustering based on AGNES. When the network is in the initial state, the number of clusters in the network is greater than the required M. In each subsequent round, the two clusters with the highest similarity are merged, and the number of clusters is reduced by one until only M clusters remain and this stage ends. In the WSN model, the similarity between clusters is defined as long as the shortest distance between two clusters is less than the maximum communication distance of the nodes, the two clusters can be considered connected. Therefore, this algorithm selects the following formula as the definition of similarity:

[0068]

[0069] Among them C i and C j Represent the i-th and j-th clusters respectively, and x and z represent the nodes in the corresponding clusters. The main steps of the first stage of the algorithm based on the AGNES equal clustering algorithm are as follows:

[0070] (1) Network initialization, each node forms a cluster C i ,i=1,2,…,N

[0071] (2) According to formula (1), the similarity between any two clusters is calculated, that is, Ci ,C j The minimum distance between any two nodes in the cluster is used to construct the similarity matrix between clusters.

[0072] (3) Find the two clusters C with the highest similarity in the similarity matrix m and C n , merge the two clusters, namely C mn =C m ∪C n , and obtain a new cluster C mn , the number of clusters is reduced by 1, and the other clusters and C are recalculated mn The similarity between them is calculated and the similarity matrix is ​​updated.

[0073] (4) Repeat the third step until the number of clusters in the network reaches M, ending the first stage clustering process of this algorithm.

[0074] The second stage needs to adjust the number of nodes in each cluster. First, define a limit level for each cluster, which is 0 at the beginning. When a cluster borrows a node from another cluster, the limit level of the cluster increases by 1. In the first step, for each cluster C with a number of nodes less than N / M, i , find all nodes that are within the communication range of the cluster and do not belong to the cluster, add these nodes to a set Q, and sort them from small to large according to the distance to the cluster. The second step is, if there is a node k in Q, and the cluster C to which the node k belongs k If the number of nodes is greater than N / M, then node k is moved from C k Remove and join cluster C i Otherwise, if there is a node k in Q, the cluster C to which it belongs k The restriction level is 0, then k is also changed from C k Move into C i , and change C i Otherwise, find the node k in Q to which cluster C belongs. k The restriction level is the smallest, and k is also reduced from C k Move into C i , and change C i Repeat the above steps until the number of nodes in all clusters is greater than N / M and the number of cluster nodes is basically the same, and the whole algorithm ends.

[0075] Step 3: Use the distributed cluster head selection algorithm for the M clusters to select cluster heads and candidate cluster heads and prepare for a new round of data collection.

[0076] The ratio of the residual energy and the distance to the sink is defined as an indicator Ed. As shown in formula (2), the possibility of a node being elected as a cluster head increases with the increase of Ed.

[0077]

[0078] In order to find the two nodes with the largest Ed value in the cluster, we first need to define the data packet P sent by the node during the cluster head selection process. i As shown below:

[0079]

[0080] Then the selection process of cluster head and backup cluster head can be completed through the following steps:

[0081] (1) Randomly select a node ClusterTmp in the cluster to serve as the temporary cluster head, and use the minimum spanning tree algorithm in graph theory to build a data forwarding tree with ClusterTmp as the root node.

[0082] (2) Each node in the cluster calculates its own Ed value according to formula (2), and sets the corresponding Ed of node 1 in data packet P to its own sequence number and the calculated Ed value, and sets the corresponding Ed of node 2 to -1 and 0 respectively.

[0083] (3) The leaf node of the data forwarding tree sends a data packet P to its next-hop node. The node that receives the data packet compares the two received Ed values ​​with its own stored Ed value, selects two nodes with the largest Ed, and fills them into the data packet P in sequence. Similarly, when the middle non-leaf node receives the data packets sent by all the previous-hop nodes, it sends its updated data packet to the next-hop node.

[0084] (4) After receiving all the data packets P, the temporary cluster head node ClusterTmp can obtain the final cluster head node and the temporary cluster head node and the corresponding Ed value. It notifies all nodes of the information of the new cluster head and the candidate cluster heads by broadcasting.

[0085] Step 4: Based on the node location information and the cluster head determined in the previous step, the local data forwarding tree construction algorithm is used to construct the routing between the common nodes involved in each row of the observation matrix.

[0086] (1) Start with the minimum spanning tree algorithm in graph theory (such as Kruskal or Prims algorithm) to build a data forwarding tree with the projection node as the root node, completing the construction of part of the route. However, some isolated nodes cannot be added to the tree at this time.

[0087] (2) For each isolated node, find the node closest to it that has been added to the forwarding tree, and use the Dijkstra algorithm to find the shortest path between the two.

[0088] Step 5: Data collection. Each node multiplies its own reading by the weight according to the constructed routing, adds the weighted sum sent by all its upstream nodes to obtain a new result, and sends this result to the next node in the routing. That is, each node multiplies its own reading x i by the non-zero elements Φ corresponding to the observation matrix Φ ij to obtain Φ ij x i . Add this value to the weighted sum data sent by all the upstream nodes of this node to obtain a new result, and then forward this result to the next node until the data converges to the cluster head to complete the collection of an observation value. The observation matrix has M rows, corresponding to M observation values.

[0089] According to the aforementioned compressive sensing theory, Z = ΦX, where Z = [z1, z2,..., z N is a column vector of M*1 (M << N), that is, the M observation values collected by the Sink node, is the observation matrix constructed by clustering the network. Each z m represents the weighted sum of the non-zero coefficients in the m-th row of Φ and the readings of the corresponding sensor nodes. As Figure 1 shown, z1 is the weighted sum of the readings of nodes 1, 7, 8, 14, and 19, stored at the cluster head node (node 8). Correspondingly, the observation matrix Φ is shown in Equation (1). It has only four rows, and each row corresponds to a cluster in the network. For example, the first row has 5 non-zero elements, indicating that these nodes are a cluster and generate an observation value in one round of data collection.

[0090]

[0091] M clusters in the network generate M observation values in one round of data collection cycle. Then, by using a mobile data collector to visit all cluster heads, the observation values can be collected. The size of the parameter M is related to the sparsity of the original readings. There is a trade-off between the sparsity of each row of Φ (the number of non-zero elements in each row) and the number of rows M of Φ. To make the distribution of non-zero elements of Φ more uniform, the sparsity of each row of the matrix is the same, that is, the number of nodes in each cluster is greater than or equal to N / M.

[0092] Step 6: Use the shortest path planning algorithm for the mobile collector to plan the moving path of the collector, and then collect the observation values at all cluster heads according to the set path and transmit them to the Sink.

[0093] (1) Use the simulated annealing algorithm to obtain the original moving trajectory S;

[0094] ① Initialization: Set the initial temperature T and the number of iterations L required for each temperature T, and randomly generate an initial path S;

[0095] ② Randomly construct a new path S1;

[0096] ③ According to the evaluation function C, calculate the difference between the new solution and the original solution ΔT = C(S1)-C(S), where C is the summation function of the trajectory length;

[0097] ④If ΔT<0, then accept S1 as the current solution, otherwise Accept it as the current solution;

[0098] ⑤ Repeat steps 2 to 4 L times;

[0099] ⑥Reduce the temperature value T = 0.99*T, and end when T is less than 0.001;

[0100] (ii) Using the sensor node communication model to further reduce the length of the trajectory S, we obtain the final trajectory S*;

[0101] ⑦ Initialization: start and end are the first and last node numbers of the trajectory S, and the new trajectory S*={start,end}, k=node;

[0102] ⑧Connect start and k, and let the line segment between these two points be L.

[0103] ⑨ Determine the intersection point P of the line segment L to the communication radius circle of k. If the intersection point exists, set start = P, k = k+1, and add start to S*; if the intersection point does not exist, determine whether L is smaller than the communication radius. If so, add start to S*, start remains unchanged, k = k+1, otherwise start = k, k = k+1, and start is added to S*.

[0104] If start = end, the algorithm ends, otherwise jump to step ⑧.

[0105] Step 7: Data recovery. The sink node uses the collected observations and the observation matrix Φ and related compressed sensing restoration algorithms, such as OMP, to recover the original N sensor readings.

[0106] The above is a detailed introduction to the method, and the objectives, technical solutions and beneficial effects of the present invention are further described in detail. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A data collection method for optoelectronic-radar sensor networks based on equal clustering. It is characterized in that The steps include: Step 1: Set up a Sink node responsible for collecting data and N ordinary sensor nodes; Step 2: According to the location distribution of nodes, the network is divided into M clusters using the equal clustering algorithm, and the observation matrix Φ is constructed based on the clustering information; Step 3: Use the distributed cluster head selection algorithm for M clusters to select cluster heads and candidate cluster heads and prepare for a new round of data collection; Step 4: Based on the node location information and the cluster head determined in the previous step, the local data forwarding tree construction algorithm is used to construct the routing between the common nodes involved in each row of the observation matrix Φ; Step 5: Each node multiplies its own reading by the weight according to the constructed route, and then adds the weighted sum of all the previous hops to get a new result, and sends this result to the next hop node in the route; Step 6: Use the mobile collector shortest path planning algorithm to plan the collector's moving path, and then collect the observations at all cluster heads according to the set path and transmit them to the Sink; Step 7: The Sink node uses the collected observations and the observation matrix Φ as well as the compressed sensing restoration algorithm to restore the original readings of the sensor node; In step 1, assume that the sensor distribution area is a narrow area with a side length of A*B, and the Sink node is located in the center of the area; all sensors in the target area are responsible for detecting the target data, and each ordinary node will perceive a reading x in each round of data collection. i , Sink node collects all node data, X = [x 1 ,x 2 ,...,x N ] T ; Step 2 is divided into two stages: The first stage: clustering based on the agglomerative hierarchical clustering method AGNES. When the network is in the initial state, the number of clusters in the network is greater than the required M. In each subsequent round, the two clusters with the highest similarity are merged, and the number of clusters is reduced by one until only M clusters remain, and this stage ends. In the wireless sensor network model, the similarity between clusters is defined as follows: as long as the shortest distance between two clusters is less than the maximum communication distance of the nodes, the two clusters are considered to be connected. Therefore, the following formula is selected as the definition of similarity: Among them, C i and C j denote the i-th and j-th clusters respectively, and x and z denote the nodes in the corresponding clusters; The steps of the first stage of the AGNES-based equal clustering algorithm are as follows: (1) Network initialization, each node forms a cluster C i , i=1,2,…,N; (2) According to formula (1), the similarity between any two clusters is calculated, that is, C i ,C j The minimum distance between any two nodes in the cluster is used to construct the similarity matrix between clusters; (3) Find the two clusters C with the highest similarity in the similarity matrix m and C n , merge the two clusters, namely C mn =C m ∪C n , and obtain a new cluster C mn , the number of clusters is reduced by 1, and the other clusters and C are recalculated mn The similarity between them is calculated and the similarity matrix is ​​updated; (4) Repeat step (3) until the number of clusters in the network reaches M, ending the first stage of clustering process; The second stage: adjust the number of nodes in each cluster; the specific method is: Define a restriction level for each cluster, which is initially 0. When a cluster borrows a node from another cluster, the restriction level of the cluster increases by 1. For every cluster C containing less than N / M nodes i , find all nodes that are within the communication range of the cluster and do not belong to the cluster, add these nodes to a set Q, and sort them from small to large according to the distance to the cluster; If there is a node k in Q, and the cluster C to which the node k belongs k If the number of nodes is greater than N / M, then node k is moved from C k Remove and join cluster C i Otherwise, if there is a node k in Q, the cluster C to which it belongs k The restriction level is 0, then k is also changed from C k Move into C i , and change C i Otherwise, find the node k in Q to which cluster C belongs. k The restriction level is the smallest, and k is also reduced from C k Move into C i , and change C i the level of restriction; Repeat the above steps until the number of nodes in all clusters is greater than N / M and the number of cluster nodes is basically the same; The specific method of step 3 is: The ratio of the remaining energy and the distance to the sink is defined as an indicator Ed. As shown in formula (2), the possibility of a node being elected as a cluster head increases as Ed increases: In order to find the two nodes with the largest Ed value in the cluster, we first define the data packet P sent by the node during the cluster head selection process. i As shown below: Then the selection process of cluster head and backup cluster head is completed through the following steps: (1) Randomly select a node ClusterTmp in the cluster to serve as a temporary cluster head, and use the minimum spanning tree algorithm in graph theory to build a data forwarding tree with ClusterTmp as the root node; (2) Each node in the cluster calculates its own Ed value according to formula (2), and sets the corresponding Ed of node 1 in data packet P to its own sequence number and the calculated Ed value, and sets the corresponding Ed of node 2 to -1 and 0 respectively; (3) The leaf node of the data forwarding tree sends a data packet P to its next-hop node. The node that receives the data packet compares the two received Ed values ​​with its own stored Ed value, selects two nodes with the largest Ed, and fills them into the data packet P in sequence. When the middle non-leaf node receives all the data packets sent by the previous-hop node, it sends its updated data packet to the next-hop node. (4) After receiving all the data packets P, the temporary cluster head node ClusterTmp obtains the final cluster head node and the temporary cluster head node and the corresponding Ed value. It notifies all nodes of the new cluster head and the information of the candidate cluster heads by broadcasting.

2. The method for collecting data of an optoelectronic-radar sensor network based on equal clustering according to claim 1, It is characterized in that The specific method of step 4 is: (1) Using the minimum spanning tree algorithm in graph theory, a data forwarding tree with the projection node as the root node is constructed to complete the construction of some routes. At this time, some isolated nodes cannot be added to the tree. (2) For each isolated node, find the node closest to it that has been added to the forwarding tree, and use the Dijkstra algorithm to find the shortest path between the two.

3. The method for collecting data of an optoelectronic-radar sensor network based on equal clustering according to claim 2, It is characterized in that The specific method of step 5 is: Each node sends its own reading x i The non-zero elements Φ corresponding to the observation matrix Φ ij Multiply them to get Φ ij x i , add this value to the weighted sum data sent by all previous hop nodes of this node to get a new result, and then forward this result to the next node until the data converges to the cluster head to complete the collection of an observation value; the observation matrix Φ has M rows, corresponding to M observation values.

4. The method for collecting data of an optoelectronic-radar sensor network based on equal clustering according to claim 3, It is characterized in that The specific method of step 6 is: (1) Use the simulated annealing algorithm to obtain the original moving trajectory S; the specific method is: ① Initialization: Set the initial temperature T and the number of iterations required for each temperature T, and randomly generate an initial path S; ② Randomly construct a new path S 1 ; ③ According to the evaluation function C, calculate the difference between the new solution and the original solution ΔT = C (S 1 )-C(S), where C is the summation function of the trajectory length; ④If ΔT<0, then accept S 1 As the current solution, otherwise with probability Accept it as the current solution; ⑤ Repeat steps 2 to 4 until the number of iterations is reached; ⑥Reduce the temperature value T = 0.99*T, and end when T is less than 0.001; (2) The sensor node communication model is used to further reduce the length of the trajectory S and obtain the final trajectory S*. The specific method is as follows: ⑦ Initialization: start and end are the first and last node numbers of the trajectory S, the new trajectory S*={start,end}, k=start+1; ⑧ Connect start and k, and let the line segment between these two points be L; ⑨ Determine the intersection point P of the line segment L to the communication radius circle of k. If the intersection point exists, set start = P, k = k + 1, and add start to S*; if the intersection point does not exist, determine whether L is less than the communication radius. If so, add start to S*, start remains unchanged, k = k + 1, otherwise start = k, k = k + 1, and start is added to S*; If start = end, the algorithm ends, otherwise jump to step ⑧.

Citation Information

Patent Citations

  • Simulated annealing-based wireless sensor network (WSN) hierarchical routing method

    CN102711206A

  • Compressed sensing-based clustered data collecting method

    CN104703262A

  • Wireless sensor network-based non-uniform routing transmission method and device

    CN107182091A

  • Clustering and data sensing method applied to wireless sensor of photovoltaic power station

    CN112672302A

  • Data fusion method and system for distributed sensor network

    WO2022100191A1