CN110932805A

CN110932805A - Network topology structure dynamic self-adaptive compressed sensing data collection method

Info

Publication number: CN110932805A
Application number: CN201910352123.4A
Authority: CN
Inventors: 张平; 唐艳艳; 陈荣元
Original assignee: HUNAN UNIVERSITY OF COMMERCE
Current assignee: Hunan University of Technology
Priority date: 2019-04-19
Filing date: 2019-04-19
Publication date: 2020-03-27
Anticipated expiration: 2039-04-19
Also published as: CN110932805B

Abstract

The invention discloses a network topology structure dynamic self-adaptive compressed sensing data collection method, which comprises the following steps: (1) node deployment and initialization; (2) double random walk is realized by the combination of the same-layer random walk and the cross-layer random walk; (3) based on the double random walk, completing the processes of compressed sensing measurement and synchronous uploading of measurement results; (4) and the SINK node recovers the compressed sensing data. The invention gets rid of the dependence on global position coordinates; the directed random walk and the nonuniformity compensation of the routing node spatial distribution are realized simultaneously through the dual random walk; the nodes of the random walk path have good spatial distribution uniformity, so that the bottleneck problem caused by the convergence effect can be avoided, and the data recovery precision is higher; the method realizes the unification of the compressed sensing measurement and the measurement result transmission, gets rid of the dependence on a static transmission path, and has good dynamic self-adaption performance of the network topology.

Description

Network topology structure dynamic self-adaptive compressed sensing data collection method

Technical Field

The invention relates to the field of wireless sensor networks, in particular to a method for realizing data collection by combining a compressed sensing technology and a dynamic routing technology.

Background

A typical Wireless Sensor Network (WSN) generally includes one SINK node and a large number of wireless sensor nodes. The sensor nodes are usually powered by batteries, energy reserves are limited, and improvement of energy utilization efficiency is an important research content. Compressed sensing has been widely applied in wireless sensor networks to achieve efficient data acquisition. The method combines data acquisition and data compression, and exceeds the traditional Nyquist-Shannon boundary by excavating sparse structures in sensor data.

The topology of a wireless sensor network typically has certain dynamic characteristics. WSNs are typically deployed in unattended environments, even in harsh environments. The wireless network signal environment is usually unstable, and the problems of channel failure and the like are easy to occur. Sensor nodes are generally low cost and are powered by batteries with limited capacity, which can easily cause node failure for various reasons. These factors cause the topology of the wireless sensor network to exhibit dynamically changing characteristics. Therefore, it is necessary to design a scheme with dynamic adaptation capability of network topology. Random walk is a typical dynamic routing mechanism. In each hop of random walk, the current node randomly selects one adjacent node as a target node of the next hop. It has better network dynamic adaptability than the traditional static routing mechanism. Failure of a single node or channel in a random walk scheme has less impact on overall performance. If there are no such failed nodes or channels, the random walk process can still be done through other nodes or channels. The random walk mechanism is also beneficial to realizing local load balance, lightening the formation of network bottleneck and prolonging the service life of the network. In a compressed sensing data acquisition scheme based on random walk, the compressed sensing measurement is typically an accumulation of sensor data along a random walk path. The method is essentially a compressed sensing scheme based on sparse sampling, fully excavates the time-space correlation of sensor data, and greatly reduces the data redundancy.

Although, combining compressed sensing with random walks offers the possibility of achieving both efficiency and dynamic adaptability. However, there remains a challenge in how to effectively and rationally integrate these two technologies. The compressive sensing technology requires that all the compressive sensing measurement results are transmitted to the SINK node for recovery. Conventional random walks are directionless and can be routed to any node in the network. Therefore, the compressed sensing measurement along the conventional random walk path cannot guarantee that the measurement result directly reaches the SINK node.

There are two representative solutions to this challenge.

One solution is to solve this challenge by introducing an additional upload process. A representative protocol is the M-CSR protocol proposed in Ad Hoc Networks, 2017 by MinhT. The protocol is shown in fig. 1, and their protocol consists of two steps. The first step is to perform compressed sensing measurements along random walk paths. The second step is the compressed sensing measurement uploading process along the static routing tree. This solution has some disadvantages. First, although the compression measurement process is performed along the dynamic routing path, the measurement result is uploaded along the static routing path, which means that it is still a static solution in nature, and thus the problem of poor network dynamic adaptability still exists. Second, the total path length of each compressed sensing measurement consists of the length of these two steps. The lengthy multi-hop path increases the overall transmission overhead. Third, the compressed sensing measurement process is performed in an accumulative manner, and the corresponding compressed sensing result is an offset estimation of the original data. It will increase the measurement error with considerable probability.

Another solution is to introduce directed random walks, i.e. each random jump always selects a node closer to the SINK, typically including the CDC scheme on IEEE Transactions on Parallel and Distributed Systems 2015 and Huang J et al on IEEE Transactions on vehicle Technology 2019. As shown in fig. 2. However, there are two major problems with this solution: first, it assumes that each node has global location information for all nodes. This assumption is not always true, especially in large-scale wireless sensor networks. Secondly, it does not consider the convergence effect of the directed random walk, that is, the node density of the random walk path far away from the SINK location is much less than the node density of the random walk path near the SINK location, for example, in fig. 2, 3 directed random walk paths are included, and their end points are SINK. There is only one path node in the two solid circle regions far away from the SINK node, and there are 3 path nodes in the solid circle region near the SINK node. Obviously, the density of path nodes in the region near the SINK node is greater than that in the region far from the SINK node. On one hand, the convergence effect causes that the energy consumption near the SINK node is too fast, and a network bottleneck is easily formed, and on the other hand, the spatial distribution of the compressive sensing sampling is not uniform, so that the compressive sensing recovery performance is reduced.

An inventor team provides directed random walk on Computer communications in 2018, the problem of dependence of traditional directed random walk on global coordinates is solved, and the routing effect of the scheme is basically consistent with that of fig. 2. In order to realize the compensation of the compressed sensing sampling spatial distribution unevenness, the scheme requires that nodes on a part of random walk paths are randomly selected to not participate in the compressed sensing measurement. In particular, the closer to the SINK region, the greater the density of path nodes, and thus the less probability that they will participate in the compressed sensing measurement needs to be ensured. However, this scheme does not fundamentally solve the problem of convergence of directed random walks. The nodes of the random walk path are still in uneven spatial distribution, and the problem that the energy consumption near the SINK node is too fast and a network bottleneck is easily formed still exists.

Therefore, a new network dynamic adaptive distributed compressed sensing data collection scheme is needed to be designed, and the integration problem of the random walk scheme and the distributed compressed sensing scheme is solved. The random walk route in the scheme is directional and does not depend on global coordinates, so that the unification of random walk measurement and measurement result transmission can be ensured, and the dependence on additional static routes is avoided. The scheme also solves the problems of uneven compressed sensing sampling and non-distributed space nodes caused by convergence effect, so as to avoid network bottleneck and improve the performance of a compressed sensing measurement matrix, further prolong the life cycle of the network and improve the recovery performance of the compressed sensing.

Disclosure of Invention

The invention aims to solve the technical problem of providing a novel network topology structure dynamic self-adaptive compressed sensing data collection method aiming at the defects of the prior art.

The technical scheme of the invention is as follows:

a compressed sensing data collection method of a network topology structure dynamic self-adaption designs a compressed sensing data collection scheme by designing a double random walk method and based on the double random walk, and comprises the following steps:

(1) in the initialization stage, sensor node deployment is carried out, and a node level number layerID and two neighbor node sets are generated for each network node respectively, wherein the two neighbor node sets are respectively a neighbor node set S in the upper layer_uAnd same-layer neighbor node set S_c；

(2) Double random walk: in each measurement process, in the outermost two-layer nodes of the network, part of the nodes are randomly selected as dual random walk starting points and are based on the hierarchy number (layerID) of each node and two types of neighbor node sets (S)_uAnd S_c) Carrying out same-layer random walk and cross-layer random walk; when the same-layer random walk is carried out, the next hop node of the same-layer random walk comes from a same-layer neighbor node set S_cWhen cross-layer random walk is carried out, the next hop node of the cross-layer random walk comes from the neighbor node set S of the previous layer_u(ii) a Double random walk is realized by the combination of the same-layer random walk and the cross-layer random walk; each double random walk path is ended at the SINK node;

(3) a compressed sensing measurement process based on double random walks: a compressed sensing measurement process which is synchronous with the double random walk process; the compressed sensing measurement result is the weighted sum of the node data of the double random walk paths; transmitting the compressed sensing measurement result corresponding to each double random walk to the SINK node along the double random walk path;

(4) and the SINK node collects the compressed sensing measurement results from each double random walk path and adopts the existing compressed sensing data recovery technology to recover the data.

Further, the step (1) comprises the following specific steps:

(1.1) the SINK node sets the own level number in the ring network topology structure as 1 and broadcasts the own level number outwards;

(1.2) each node determines the level number of the node according to the message received for the first time, and the specific method is as follows: adding 1 to the layer number in the message as the layer number of the message, and adding the message sender to the neighbor node set S of the previous layer of the message sender_uThen, broadcasting the own layer number after proper time delay;

(1.3) each node determines two own neighbor node sets (namely S) according to the size of the hierarchy number in the received message_uAnd S_c) The contents of (1); the specific method comprises the following steps: if the current node has not set its own hierarchy number, then the sender who receives the message for the first time is added to its own upper layer neighbor node set S_uPerforming the following steps; if the current node has set its own level number and the level number in the message is smaller than its own level number, the message sender is added to its own neighbor node set S_uPerforming the following steps; if the current node has set its own hierarchy number and if the hierarchy number in the message is equal to its own hierarchy number, the message sender is added to its own set S of neighbor nodes in the same layer_cPerforming the following steps;

further, the step 2) comprises the following specific steps:

(2.1) setting a double random walk starting point: all the outermost 2-layer nodes independently and randomly determine whether to serve as random walk starting nodes or not; that is, the nodes of the outermost 2 layers independently generate a random number ρ which is uniformly distributed according to [0, 1], and then the random number ρ is compared with a certain predefined threshold value; the threshold value is equal to the reciprocal of the total node number of the two outermost layers; if the random number is larger than the threshold value, the current node is used as an initial node to initiate double random walk;

(2.2) respectively initializing an auxiliary decision variable CntCurr by each random walk starting node, wherein the initial values are respectively 0; in the subsequent double random walk process, all nodes on the same random walk path modify the value of CntCurr according to the predefined rule in the step 2.3, and send the modified CntCurr to the next hop node;

(2.3) the double random walk process is as follows:

(2.3.1) if the current node is the SINK node, ending the random walk; otherwise, receiving and extracting the value of CntCurr;

(2.3.2) if the CntCurr is smaller than k × layerID + b (where k and b are model parameters), performing peer random walk, namely randomly selecting one node from the peer neighbor node set as a next hop, and sending CntCurr ═ CntCurr +1 to the next hop node;

otherwise, performing cross-layer random walk, namely randomly selecting a node from the neighbor node set of the upper layer as a next hop, and sending CntCurr which is 0 to the next hop node;

(2.3.3) in the previous step, if the same-layer node set of the current node is empty when the same-layer random walk is carried out, modifying to initiate cross-layer random walk so as to prevent the random walk from being interrupted accidentally; similarly, if the cross-layer random walk node set of the current node is found to be an empty set when cross-layer random walk is performed, the same-layer random walk is modified to be initiated;

(2.3.4) repeating the steps 2.3.1, 2.3.2 and 2.3.3 until the SINK node is reached.

Further, the step 3) comprises the following specific steps:

initializing a respective path node number list ID list and a measurement coefficient list flag list of a corresponding node to be null values respectively at each double random walk starting point; initializing each double random walk starting point to 0 according to the compressed sensing measurement result DATA; starting from the starting point of the double random walk path, each node on the double random walk path cooperatively completes the compressed sensing measurement process based on the double random walk according to the following mode:

firstly, randomly selecting-1 or +1 as a measurement coefficient flag with equal probability at the current node on the double random walk path, and updating the compressed sensing measurement result as follows: DATA ═ Σ_{i∈ID list}data (i) x flag (i) indicating the ID and selected coefficient are added to ID list and flag list, respectivelyData collected by the ith node on the random walk path, flag (i), a measurement coefficient selected for the ith node on the random walk path;

then, the current node on the double random walk path sends the updated ID list, flag list and DATA to the next hop node of the double random walk together;

and finally, repeating the two steps until the current node becomes the SINK node.

Has the advantages that:

according to the scheme adopted by the invention, a double random walk is designed, and the directed random walk with uniformly distributed path nodes in space is realized through the combination of the same-layer random walk and the cross-layer random walk without global coordinate information. Furthermore, the invention designs a scheme for collecting data compressed and sensed based on dynamic routing completely based on the double random walk, realizes the unification of the two processes of the compressed and sensed measurement and the transmission of the measurement result of the random walk, improves the dynamic self-adaptive capacity of a network topological structure, and well solves the problems of uneven distribution of path node space caused by the convergence effect of the directed random walk, network bottleneck caused by the uneven distribution of the path node space, reduced recovery performance of the compressed and sensed data and the like. Compared with the existing scheme, the dynamic self-adaptability of the network is higher, the spatial distribution of the routing nodes is more uniform, the generation of network bottleneck can be avoided, and the performance of the compressed sensing measurement matrix is improved, so that the life cycle of the network is longer, and the compressed sensing recovery performance is better.

Drawings

FIG. 1 is a schematic diagram of a random walk compressed sensing scheme (M-CSR) with static routing

FIG. 2 is a diagram of a conventional directed random walk compressed sensing scheme

FIG. 3 shows the network hierarchy position of node a and its two neighbor node sets (S)_c(a) And S_u(a))

FIG. 4 is a diagram of dual random walk state transitions

FIG. 5 is a schematic diagram of a double random walk (partial)

FIG. 6 double random walk schematic (Global)

FIG. 7 evaluation results of network dynamic adaptivity

FIG. 8 probability of each node appearing on each random walk path

FIG. 9 mathematical expectation values for compressed sensing measurement coefficients for each node

FIG. 10 recovery Performance evaluation results (k sparse signals)

FIG. 11 energy consumption evaluation results (energy consumption Top10 node)

FIG. 12 average energy consumption evaluation results

The specific implementation mode is as follows:

the specific implementation process of the invention is as follows:

first, network initialization, hierarchical number generation and creation of two types of neighbor node sets

In the invention, the traditional scheme is adopted to carry out sensor node deployment. In the initialization stage, a node layer number layer ID and two neighbor node sets are respectively generated for each node, wherein the two neighbor node sets are respectively a neighbor node set S of the previous layer_uAnd same-layer neighbor node set S_c. The two neighbor node sets of each node respectively reflect the relative position relationship between the node and the neighbor nodes in the two types of node sets.

According to the level numbers of all the nodes and the relative position relation reflected by the two types of neighbor node sets, a ring level network topology can be obtained. The hierarchical network topology is centered around SINK, and each node is located in a certain level of the hierarchical structure. The smaller the tier number ID close to the SINK, the larger the tier number ID far from the SINK.

Node i' S upper neighbor node set S_uIs defined as S_u(i) (j | d (i, j) < R, layer (i) > layer (j) }, its peer-neighbor node set S_cIs defined as S_c(i) Layer (j) < R, layer (i) ═ layer (j) }. Where R is the wireless communication radius, d (i, j) is the euclidean distance of nodes i and j, and layer (i) represents the level number of node i in the ring level network topology.

The specific generation process of the hierarchy number and the two types of neighbor node sets is as follows:

(1.1) the SINK node sets the own level number as 1 and broadcasts the own level number outwards;

(1.3) each node determines two own neighbor node sets (namely S) according to the size of the hierarchy number in the received message_uAnd S_c) The contents of (1); the specific method comprises the following steps: if the current node has not set its own hierarchy number, then the sender who receives the message for the first time is added to its own upper layer neighbor node set S_uPerforming the following steps; if the current node has set its own level number and the level number in the message is smaller than its own level number, the message sender is added to its own neighbor node set S_uThe method comprises the following steps: if the current node has set its own hierarchy number and if the hierarchy number in the message is equal to its own hierarchy number, the message sender is added to its own set S of neighbor nodes in the same layer_cIn (1).

Fig. 3 is a specific example of a network hierarchy position for node a and its two types of neighbor node sets. As shown, node a is located at the ith level in the hierarchical network structure. Two sets of data related to the node a, namely a same-layer neighbor node set S, are stored in the node a_c(a) And the neighbor node set S of the previous layer_u(a) In that respect All nodes of the previous set are located in the ith layer in the network and all nodes of the previous set are located in the i-1 layer in the network.

After initialization is complete, each node will be assigned a level number and both generate two similar sets of neighbor nodes.

Two and two double random walk

The double random walk is performed based on the two types of neighbor node sets. The dual random walk includes two types of random jump processes, one is random walk jump on the same layer, and the other is cross-layer random walk jump. The former is based on the same-layer neighbor node set, and the latter is based on the previous-layer neighbor node set. The two types of jumps are respectively carried out with certain probability and are mutually crossed. The two are combined with each other to form a complete double random walk.

The dual random walk state transition probability matrix is defined according to the following expression.

More specifically, the state jump of the double random walk is performed in two steps. First, a node decides whether to jump to its set of nodes in the same layer or to its set of neighbor nodes in an upper layer with a certain probability. Then, in the corresponding neighbor set, the node in the set is jumped to randomly with the same probability. Assuming that the current node is a, its corresponding random walk state transition diagram is shown in fig. 4. First, the nodes a jump to the set S with a certain probability respectively_c(a) Or set S_u(a) In that respect Then, the node jumps to set S_c(a) Or set S_u(a) The probability of a particular node within a set is determined by the size of the set itself. In the double random walk state transition diagram of FIG. 4, a → S_c(a) The state jump of → b realizes the random walk jump of the same layer, a → S_u(a) The state jump of → b implements a random walk jump across layers.

Fig. 5 is a schematic diagram of a specific dual random walk local structure. The figure includes a two-hop random walk path a- > b- > c. When the random walk starts, the current node is the a node located at the ith layer. First, the node a jumps to the set S with a certain probability_c(a) Or set S_u(a) I.e. to decide whether a layer random walk or a cross-layer random walk is to be performed. Then, skipping is carried out in the internal nodes of the corresponding set according to equal probability, and therefore same-layer random walk or cross-layer random walk is achieved specifically. In this case, the current node a finally selects S_c(a) Node b in the set as its next oneAnd (6) jumping. and after the jumping process of a- > b is completed, changing the current node on the random walk path into a node b. The next hop node selection process for node b is substantially similar to the process described above. In this case, the final selection of node b is at S_u(a) And taking the node c in the set as the next hop of the node c, thereby completing the hop of b- > c. In the random walk path a- > b- > c of the present case, the jump from a- > b is a same-layer jump (same-layer random walk), and the jump from b- > c is a cross-layer jump (cross-layer random walk).

Fig. 6 is a more specific example of a double random walk. Three dual random walk paths are included in the figure, where the dotted arrows are same-layer random walks and the solid arrows are cross-layer random walks.

Different ways exist in the idea of the double random walk in the specific implementation, and a specific double random walk implementation scheme is listed as follows:

(2.3) the double random walk process is as follows:

The k multiplied by the layerID + b in the invention is mainly used for controlling the number of nodes of the random walk path of the same layer on different layers in the hierarchical network, wherein k and b are model parameters, and the layerID is the hierarchical number of the current node. The larger the hierarchy number is, the larger the number of nodes of the random walk path at the same layer is, so that the compensation of the spatial non-uniform distribution problem caused by the convergence effect is realized. The linear compensation model is adopted here, and can be adjusted to other reasonable models to realize the spatial non-uniformity compensation. In the scheme listed in the invention, the quantity of random walks of the same layer is controlled by using a CntCurr variable to assist decision, so that the compensation of the non-uniform distribution problem is realized.

Third, compressed sensing measurement based on double random walks

In the invention, a compressed sensing measurement process is implemented along the double random walk paths and is synchronously carried out with the double random walk process; the compressed sensing measurement result is the weighted sum of the node data of the double random walk paths, and the weight coefficient is the content in the subsequent flag list; and transmitting the compressed sensing measurement result corresponding to each double random walk to the SINK node along the double random walk path.

firstly, randomly selecting-1 or +1 as a measurement coefficient flag with equal probability at the current node on the double random walk path, and updating the compressed sensing measurement result as follows: DATA ═ Σ_{i∈ID list}data (i) multiplied by flag (i) is added to the tail of the ID list and the flag list, wherein the ID of the data (i) represents the data collected by the ith node on the random walk path, and the selected coefficient is added to the tail of the ID list and the tail of the flag list;

Compared with the traditional scheme, the flag has two values of-1 and 1, and if a flag list is transmitted, extra transmission overhead needs to be added. Further, in order to reduce the transmission overhead of the flag List, the invention can also be optimized as follows, namely, the value of the flag List is directly determined by the ID List. To this end, the present invention may add the following conventions: the full value of the first element in each ID list is positive 1, and then-1 and positive 1 alternate in sequence. Since the random walk has good random characteristics, the random performance of the compressed sensing measurement matrix is not affected by the engagement mechanism.

In the invention, the double random walk compressed sensing process has the characteristic of sparse sampling. And the nodes on the double random walk paths participate in the compressed sensing measurement process, and the corresponding compressed sensing measurement coefficient is the corresponding flag value. Nodes which are not on the double random walk paths do not participate in the random walk compressive sensing measurement process, so that the corresponding compressive sensing measurement coefficient is zero. In the invention, the compressed sensing measurement coefficient comprises three values of { -1, 0, 1}, and compensation of 1 element in the measurement matrix is realized by introducing-1 into the measurement matrix, so that energy balance compensation of the whole measurement matrix is realized, and finally acquisition of an unbiased compressed sensing measurement result is realized.

The compressed sensing measurement matrix of the scheme is a Bernoulli measurement matrix in essence. The Beruli measurement matrix is widely used in the field of compressed sensing, the compressed sensing measurement result is unbiased estimation of an original result, and the performance similar to a Gaussian measurement matrix can be realized.

Fourthly, compressed sensing data recovery process

And finally, the SINK node collects the compressed sensing measurement results from each dual random walk path, and performs data recovery by adopting the traditional compressed sensing data recovery technology. In the compressed sensing data recovery stage, the existing mainstream compressed sensing data recovery technology can be adopted for data recovery. Such as OMP (organic Matching pursuit) and SAMP (sparse Adaptive Matching pursuit), among others.

Fifth, performance evaluation

The evaluation content comprises three parts, wherein one part is to evaluate the dynamic adaptivity of the scheme; secondly, evaluating the uniformity sampling performance; thirdly, evaluating the recovery performance; and fourthly, comparing the energy consumption.

The evaluation process uses a network model of random uniform deployment, i.e., nodes are assumed to randomly uniformly deploy 40 units of scale over the area. The SINK nodes are all located in the center of the network, and the communication radiuses of all the nodes are 5 unit scales. The basic scheme for comparison is the M-CSR scheme (hereinafter referred to as the conventional scheme) most relevant to the present invention, which is newly published on the ad hoc networks.

(1) Dynamic adaptive evaluation

The network dynamics in the invention means that some nodes fail due to various reasons in the network operation process. The deployment environment of the wireless sensor network is complex, and various factors can cause the node to fail. Network dynamics may cause some or even all of the pressure sensing measurements to fail to be transmitted to the receiver, thereby affecting the performance of the scheme.

The dynamic adaptability evaluation method of this experiment is as follows. In the network operation process, some nodes in the network are randomly selected as fault nodes, and the influence of the fault of the nodes on the performance of the whole scheme is evaluated. The node failure rate varies from 0.02 to 0.22. The experiment was repeated 500 times on both protocols. The final evaluation results are the average of these experiments. The network size varies from 500 to 1100 with a step size of 300.

In the conventional scheme, the random walk compressed sensing measurement result is uploaded based on a static routing tree. The construction of such a routing tree is costly and therefore we assume that the static routing tree is constructed only at the initialization stage. The static routing tree is not reconstructed in the subsequent network operation process. This means that if a failed node causes a failure in data upload when a packet is transmitted through this node.

In the present invention, both the compressed sensing measurement and the measurement uploading process are based on dual random walks. Specifically, both processes are implemented by randomly traversing two types of neighbor node sets. In this experiment, we assume that these two types of neighbor node sets and ring-type hierarchical topology are constructed only in the network initialization phase. It will no longer be re-established during subsequent network operations. This means that the failed node will not be able to participate in the subsequent double random walks. Double random walks by any failed node are considered invalid random walks.

Fig. 7 is an evaluation result of network dynamics adaptability, in which the horizontal axis is a failure node ratio and the vertical axis is a ratio at which a compression measurement result cannot be transmitted to the SINK. As can be seen from fig. 7, the node failure has a certain effect on both the conventional scheme and the scheme of the present invention. However, the network dynamic adaptability of the scheme of the invention is obviously superior to that of the traditional scheme.

The total number of nodes has a certain influence on the dynamic adaptability of the two schemes. As can be seen from fig. 7, the larger the number of nodes, the better the performance of the scheme under the same node failure rate. The more nodes, the greater the deployment density of the nodes, and the smaller the influence of partial node faults on network connectivity.

(2) Uniformity sampling performance evaluation

In order to evaluate the sampling performance of the double random walk uniformity, 5000 times of double random walk compressed sensing measurement experiments are carried out. Based on these experimental results, on the one hand, we count the probability that each node appears on each random walk path. On the other hand, the mathematical expectation value of the compressed sensing measurement coefficient of each node is counted. Fig. 8 and 9 are test results of these two tests, respectively. The horizontal axis is the node number, and the vertical axis of fig. 8 is the probability of each node appearing on each path. The vertical axis of fig. 9 is the mathematical expectation of the compressed perceptual measurement coefficients of the node. As can be seen from fig. 8, the probability that each node appears on a certain random walk path satisfies a better random uniform characteristic, so that the dual random walk process better achieves uniform sampling. As can be seen from FIG. 9, the mathematical expectation of the compressed sensing measurement coefficient of each node is located near 0, and the compressed sensing measurement coefficient has a good random uniform characteristic, so that the error introduced in the compressed sensing measurement stage is greatly reduced.

(3) Recovery performance evaluation

The recovery performance evaluation is performed based on the k-sparse signal. The k-sparse signal is a standard signal widely used for performance testing of compressed sensing schemes. The horizontal axis is the number of compressed sensing measurements and the vertical axis is the successful recovery ratio. Successful recovery is defined as a recovery error of less than 10^-6. The experiment was repeated 500 times at different parameter settings.

Fig. 10 is an evaluation result based on two types of k sparse signals (k 5 and k 10). In two different types of sparse signals, the successful recovery rate of the scheme is obviously superior to that of the traditional scheme. The measurement result of the conventional scheme is a bias estimate for the raw data. Therefore, the conventional scheme has higher probability of introducing compressed sensing measurement errors, thereby reducing the probability of realizing lossless recovery of the k-sparse signal. In this scheme, the random walk compressed sensing measurement is an unbiased estimate of the raw data. Therefore, the scheme of the invention can realize better recovery performance.

(4) Energy consumption assessment

The energy consumption evaluation comprises a maximum energy consumption value evaluation and an average energy consumption value evaluation. In a wireless sensor network, the maximum energy consuming node is typically located near the SINK. Failure of these nodes (e.g., energy exhaustion) will cause data to fail to be transmitted to the SINK through them in multiple hops, thereby forming a network bottleneck. Therefore, the energy consumption change trend of the maximum energy consumption node can be used for evaluating the life cycle of the network when the corresponding scheme is used. The average energy consumption is the average energy consumption of all nodes and is used for evaluating the overall energy consumption of the scheme.

Where fig. 11 is the energy consumption evaluation result of the maximum energy consumption node (the first 10 nodes), and fig. 12 is the average energy consumption evaluation result. The horizontal axis is the number of random walk compressive sensing experiments, each including a complete set of compressive sensing measurement processes. The vertical axis is the corresponding energy consumption. The energy consumption per unit represents the energy required for a single-hop transmission of 1-bit data. The vertical axis uses logarithmic coordinates. According to fig. 11 and 12, the maximum energy consumption and the average energy consumption of the proposed scheme are much better than the conventional scheme, at least by an order of magnitude.

To provide more specific scheme performance comparison results, we analyzed it through a more specific scenario. Assume that each node is equipped with 50000 units of energy, i.e. the dashed line in fig. 11 parallel to the horizontal axis. The first failed node occurs in the conventional scheme after only about 10 experiments, and the energy of all 10 nodes is exhausted after about 40 experiments. In contrast, the first energy-depleted node occurs with the present invention after about 100 experiments. After about 120 experiments, only one node still failed due to energy depletion.

It is noted that in the conventional scheme, the data transmission process of the compressed sensing result of random walk is based on a static path tree. The data transmission process of the invention is also based on random walk and dynamic route. A single node failure has less impact on network connectivity than conventional schemes. Thus, in practical applications, the effects of the present invention may be slightly better than the experimental test results presented herein.

Claims

1. A network topology dynamic self-adaptive compressed sensing data collection method is characterized by comprising the following steps:

(2) Double random walk: in each measurement process, in the outermost two-layer nodes of the network, part of the nodes are randomly selected as dual random walk starting points and are based on the hierarchy number (layerID) of each node and two types of neighbor node sets (S)_uAnd S_c) Carrying out same-layer random walk and cross-layer random walk; when the same-layer random walk is carried out, the next hop node of the same-layer random walk comes from a same-layer neighbor node set S_cWhen cross-layer random walk is carried out, the next hop node of the cross-layer random walk comes from the neighbor node set S of the previous layer_u(ii) a Double random walk is realized through the serial combination of the same-layer random walk and the cross-layer random walk; each double random walk path is ended at the SINK node;

(3) a compressed sensing measurement process based on double random walks: a compressed sensing measurement process which is synchronous with the double random walk process; the compressed sensing measurement result is the weighted sum of the node data of the double random walk paths, and the weight coefficient is determined by a measurement coefficient list (flag list); transmitting the compressed sensing measurement result corresponding to each double random walk to the SINK node along the double random walk path;

2. The method for collecting compressed sensing data with dynamically adaptive network topology according to claim 1, wherein the step (1) comprises the following steps:

(1.1) the SINK node sets the layer number of the SINK node in a ring network topology structure as 1 and broadcasts the layer number of the SINK node outwards;

(1.2) each node determines the level number of the node according to the message received for the first time, and the specific method is as follows: adding 1 to the layer number in the message as the layer number of the message, and adding the message sender to the neighbor node set S of the previous layer of the message sender_uIn the method, the own layer number is broadcasted after proper time delay;

(1.3) each node determines two own neighbor node sets (namely S) according to the size of the hierarchy number in the received message_uAnd S_c) The contents of (1); the specific method comprises the following steps: if the current node has not set its own hierarchy number, then the sender who receives the message for the first time is added to its own upper layer neighbor node set S_uPerforming the following steps; if the current node has set its own level number and the level number in the message is smaller than its own level number, the message sender is added to its own neighbor node set S_uPerforming the following steps; if the current node has set its own hierarchy number and if the hierarchy number in the message is equal to its own hierarchy number, the message sender is added to its own set S of neighbor nodes in the same layer_cIn (1).

3. The method for collecting compressed sensing data with dynamically adaptive network topology according to claim 1, wherein the step (2) comprises the following steps:

(2.1) setting a double random walk starting point: all the outermost 2-layer nodes independently and randomly determine whether to serve as random walk starting nodes or not; that is, the nodes of the outermost 2 layers independently generate a random number ρ which is uniformly distributed according to [0, 1], and then the random number ρ is compared with a certain predefined threshold value; the threshold value is equal to the reciprocal of the total number of the nodes of the two outermost layers; if the random number rho is larger than the threshold value, the current node is used as an initial node to initiate double random walk;

(2.3) the double random walk process is as follows:

(2.3.1) if the current node is the SINK node, ending the double random walk; otherwise, receiving and extracting the value of CntCurr;

(2.3.3) in the previous step, if the same-layer node set of the current node is empty when the same-layer random walk is performed, modifying to perform cross-layer random walk to prevent the random walk from being interrupted accidentally; similarly, if the cross-layer random walk node set of the current node is found to be an empty set when cross-layer random walk is performed, the same-layer random walk is modified to be initiated;

4. The method for collecting compressed sensing data with dynamically adaptive network topology according to claim 1, wherein the step (3) is specifically as follows: initializing a respective path node number list (IDlist) and a measurement coefficient list (flag list) of a corresponding node to be null values respectively for each double random walk starting point; initializing each double random walk starting point to 0 according to the compressed sensing measurement result DATA; starting from the starting point of the double random walk path, each node on the double random walk path cooperatively completes the compressed sensing measurement process based on the double random walk according to the following mode:

firstly, randomly selecting-1 or +1 as a measurement coefficient flag with equal probability at the current node on the double random walk path, and updating the compressed sensingThe measurement results are: DATA ═ Σ_{i∈ID list}data (i) multiplied by flag (i), and adding the ID of the self and the selected coefficient into the ID list and the flag list respectively, wherein the data (i) represents the data collected by the ith node on the random walk path, and the flag (i) is a measurement coefficient selected by the ith node on the random walk path;