CN115426408A - Partition cooperative sensing method for large-scale Internet of things - Google Patents

Abstract

The present invention proposes a partition cooperative sensing method for the large-scale Internet of Things, including the following steps: S1. Divide the sensing area of the node; S2. The state of the node is divided into working and sleeping, and the state is converted by the probability of awakening the node; S3. Each node has a request storage table, which stores and transmits service request information for the perceived request information; S4. Each node has a preference location, and modifies the optimal location according to the sense request information and the request storage table; S5. After the optimal orientation sensing is completed, the node will calculate the selection operator for each orientation, and select the orientation represented by the largest selection operator for sensing; S6. After the sensing operation is completed, the node will wake up other nodes in the sensing orientation and transmit Request information; S7. After the transmission operation is completed, each node will calculate the capability value of the node and modify the wake-up probability, and filter out some inefficient nodes through the convolution method. The method can improve the perception accuracy of the network to the service request and reduce the energy consumption of the network.

Description

A Partition Cooperative Perception Method for Large-Scale Internet of Things

【Technical field】

The invention relates to the technical field of the Internet of Things, in particular to a partition cooperative sensing method oriented to the large-scale Internet of Things.

【Background technique】

At present, researchers have adopted hardware and software node optimization methods to deal with the problems of low node sensing accuracy and high energy consumption in the IoT sensing method. For the hardware, the researchers innovated the battery structure of the node and the sensing method, which can effectively reduce energy consumption and accuracy, but the cost is high and it is difficult to use on a large scale. In terms of software, researchers are mainly optimizing the node scheduling mechanism and improving the node optimization method, or combining self-organizing networks to innovate the sensing method, which can have better performance in terms of sensing accuracy or network energy consumption. But it is difficult to achieve a balance between the two. In recent years, machine learning and deep learning methods have been continuously applied to the Internet of Things, and there has been good progress in some scheduling algorithms and feature extraction, but there is no good application in the underlying sensing technology of the Internet of Things.

Aiming at the heterogeneous distribution of sensing nodes in the deployment scenario of the manufacturing Internet of Things, the dynamic change characteristics of uncertain events and the dynamic change characteristics of sensing node availability, the node scheduling optimization mechanism for reliable sensing of uncertain events and the efficient search for fast convergence are studied. An optimal method is necessary. Aiming at the node perception problem of the large-scale Internet of Things, a partition cooperative sensing method for the large-scale Internet of Things is proposed.

【Content of invention】

The purpose of the present invention is to solve the problems in the prior art, and propose a large-scale Internet of Things-oriented partition cooperative sensing method, which can improve the accuracy of network perception of service requests and reduce network energy consumption.

In order to achieve the above object, the present invention proposes a partition cooperative sensing method for large-scale Internet of Things, including the following steps:

S1. Divide the perception area of the node, each area represents a perception orientation, and the perception of each orientation is independent of each other;

S2. The state of the node is divided into two states of working and sleeping, and the mutual conversion of the two states is controlled by the wake-up probability of the node;

S3. Each node has a request storage table to store the perceived request information and transmit the service request information;

S4. Each node has a favorite location, and the favorite location is modified according to the perceived request information and the request storage table;

S5. After the perception of the preferred orientation is completed, the node will calculate the selection operator of each orientation, and select the orientation represented by the largest selection operator for perception;

S6. After a round of sensing operation is over, the node will wake up other nodes in the sensing position and transmit request information;

S7. After the transmission operation is completed, each node will calculate the node's ability value, modify the wake-up probability through the node's ability value, and screen out some inefficient nodes.

Preferably, the attributes of the nodes are as follows:

(Nid, N_X, N_Y, Rc, Rs, PreOri, Pwake, Per, Sleep, RSF, Wtime, Em)

Among them, Nid represents the node number; N_X and N_Y represent the geographical coordinates of the node; Rc and Rs represent the communication radius and perception radius respectively; PreOri represents the number of the preferred orientation; Pwake represents the wake-up probability of the node; Per and Sleep represent the perception and sleep flags respectively , and satisfy the following relationship:

RSF represents the perception marks of eight orientations; Wtime represents the waiting time of eight search areas, and Em is the energy of nodes.

Preferably, in step S1, each divided region has the same size.

Preferably, in step S3, in each round of sensing operation, node N _i will perceive or receive a service request, and node N _i will record the request information in the request storage table RST _i ; the recorded information is expressed as a set in the node, Its elements are as follows:

(Rid,R_X,R_Y,Ori,Rw,t)

Among them, Rid is the number of the service request; R_X and R_Y represent the geographic coordinates of the service request; Ori represents the orientation number of the perceived Rid; Rw represents the way to obtain the service request; t represents the time the request is stored in the node;

Preferably, in step S4, during the sensing process, the node Ni perceives the service request with preference orientation, the specific process is as follows: first update the request coordinates in RST _i according to the following formula; The location of the node is marked as 1; the HNN network is used to determine the preferred orientation of the node; by converting the lattice into a vector noi by row, the HNN network is used to simulate the lattice of noi and the eight search orientations divided in S1, and use the probability p Modify the result of the HNN network with a small probability, and the result obtained is the preferred orientation of the next round of node N _i ; where the probability p is a probability set separately to prevent the model from overfitting;

Where R_X and R_Y represent the request coordinates in RST _i , N_X and N_Y represent the coordinates of node N _i , and Rs _i represents the perception radius of node N _i .

Preferably, in step S5, the value of the selection operator of the node area is determined according to the search frequency and waiting time of each node, and the formula is as follows:

表示节点N_i的搜索区域k的等待时间。Among them, S(i,k) represents the value of the selection operator of the kth search area of node N _i , where Rnum represents the number of requests stored in node N _i ; Freq(i,k) represents the request in area k Total number; PFreq(i,k) indicates the total number of requests whose area is k and Rw is 1; RSF(i,k) indicates the perception flag of area k in node N _i ;

Indicates the waiting time of node N _i 's search area k.

As a preference, the ability value Ab of the node is calculated using the value of the selection operator S according to the following formula:

表示节点N_i中八个搜索方位中最少的等待时间；计算完选择算子之后，开启最大的选择算子所代表的方位进行感知。这样在喜好方位的基础上增加了一个感知方位，不仅能够让搜索面积持续的更新，而且避免模型陷入局部搜索。Among them, M is the control factor that controls the degree of difference between Ab, T represents the time from the beginning of the model to the current perception, S(i,m) represents the value of the selection operator of the m-th search area of node N _i ; Freq(i ,m) represents the total number of requests in area m in node N _i ;

Indicates the minimum waiting time among the eight search orientations in node N _i ; after calculating the selection operator, open the orientation represented by the largest selection operator for perception. In this way, a perceptual orientation is added on the basis of the preferred orientation, which not only allows the search area to be continuously updated, but also prevents the model from falling into local search.

Preferably, in step S7, the wake-up probability of the node is calculated by the following formula, and the wake-up probability is controlled between [a, b]:

Among them, Ab _i represents the ability value of node N _i , and Ab _min and Ab _max represent the smallest ability value and the largest ability value among all nodes;

Then use the convolution method to screen inefficient nodes, obtain the Ab of all nodes, and combine them into a 12×12 lattice, called feature map C1; then proceed as follows:

进行卷积，得到名为C2的10×10大小的特征映射，该特征映射得到每个3×3区域的平均能力值；Step a: Use a 3×3 convolution kernel

Perform convolution to obtain a feature map of size 10×10 named C2, which obtains the average ability value of each 3×3 region;

Step b: The 2×2 normative mode pooling operation is performed on C2, and the obtained result is used to represent the horizontal efficiency of the 2×2 region, and the obtained feature map is called C3;

Step c: Obtain the feature map C3 through convolution and mode pooling. By comparing the data in C3, it can be determined that the feature value with a smaller feature value is an area with weaker perception;

Step d: According to the corresponding relationship in step c, select the nodes with lower ability values in C1, and form their ability values into lattice C4;

Step e: Divide each C4 into 4 feature maps with a specification of 2×2, and each feature map contains the ability values of 4 nodes; perform convolution operation on each feature map; convolution operation is to remove its own ability value The sum of the ability values of the remaining three nodes;

Step f: By comparing the capability value in C5, select the node number with the smallest capability value to C6; C6 contains 4 results selected from each C4;

Step g: The node number recorded in C6 forms the final output result C7, and according to the node number in C7, C1 sets the Pwake of the recorded node number to 0.

Beneficial effects of the present invention: the present invention divides the self-organizing areas of sensing nodes in the Internet of Things, introduces the idea of orientation, and improves the perception accuracy of the network for service requests; on the basis of preferred orientation, the selection operator increases A perception orientation improves the flexibility of the perception area, not only updates the search area, but also avoids local search. By controlling the node state through adaptive wake-up probability, the number of nodes working in the model can be effectively controlled. Through the convolution screening method, the number of nodes can be further reduced to reduce the energy consumption of the network under the premise of ensuring the accuracy of perception.

1. In the present invention, the sensing area of the sensing node is divided into eight parts, and each part is represented as a sensing orientation. This division method refers to the cell sensitivity mechanism of human vision, which makes the perception of nodes more directional. Using the HNN neural network to determine the preferred orientation of the sensing node can make the determination of the orientation more accurate. Preference location allows nodes to quickly lock areas with frequent service requests, capture a large number of service requests in a short period of time, and improve the perception rate of the network.

2. The present invention designs eight search operators for the eight search directions divided by sensing nodes. By calculating the search operator, the present invention adds a perception orientation based on the preference orientation, which not only increases the perception area, allows the search area of the perception node to be continuously updated, but also prevents the perception node from falling into a local search. At the same time, the design of dual sensing orientations makes the perception of nodes more flexible. Each sensing orientation has its own functions, which improves the fault tolerance rate of the sensing network and the sensing rate of the network.

3. The present invention designs a dynamic node switching mode for a large-scale sensing network. Firstly, an adaptive wake-up probability is designed for each node, which can stabilize the number of nodes in the network, so that the number can change with the frequency of service requests. At the same time, the ability value is designed for each node. Using the convolution method to extract the perception ability of the region through the ability value of the node, it can skillfully screen out the nodes that are not helpful to the next round of perception in the region with low perception ability, so as to reduce the network perception ability while ensuring the network perception ability. The number of working nodes further reduces network energy consumption.

The features and advantages of the present invention will be described in detail by way of examples with reference to the accompanying drawings.

【Description of drawings】

Fig. 1 is a flow chart of a partition cooperative sensing method oriented to the large-scale Internet of Things of the present invention;

Fig. 2 is the design drawing of node;

Fig. 3 is the schematic diagram that adopts HNN network to determine the preference position of node;

4 is a schematic diagram of a direction-sensitive sensing mode;

Figure 5 is a schematic diagram of energy consumption comparison of each algorithm;

Figure 6 is a schematic diagram of the perception rate comparison of each algorithm;

Figure 7 is a schematic diagram of the comparison of the number of node wake-ups for each algorithm;

Fig. 8 is a schematic diagram of C3 obtained by performing the mode pooling operation of the 2×2 specification on C2;

Fig. 9 is a schematic diagram of feature map C3 obtained through convolution and mode pooling;

Fig. 10 is a schematic diagram of convolution operation for each feature map of C4.

【detailed description】

The present invention proposes a partition cooperative perception method for the large-scale Internet of Things, and it can be applied to the self-organizing Internet of Things. Referring to the algorithm flow in Figure 1, it includes the following steps: 1) initializing the node state; 2) modifying the node 3) Sensing request information; 4) Waking up surrounding nodes and sharing request information; 5) Modifying the awakening probability of nodes; 6) Modifying the state of nodes; 7) Screening for awakened nodes.

Algorithm design of the present invention, comprises the steps:

S1. Divide the perception area of the node, each area represents a perception orientation, and the perception of each orientation is independent of each other;

S2. The state of the node is divided into two states of working and sleeping, and the mutual conversion of the two states is controlled by the wake-up probability of the node;

S3. Each node has a request storage table to store the perceived request information and transmit the service request information;

S4. Each node has a favorite location, and the favorite location is modified according to the perceived request information and the request storage table;

S5. After the perception of the preferred orientation is completed, the node will calculate the selection operator of each orientation, and select the orientation represented by the largest selection operator for perception;

S6. After a round of sensing operation is over, the node will wake up other nodes in the sensing position and transmit request information;

S7. After the transmission operation is completed, each node will calculate the node's ability value, modify the wake-up probability through the node's ability value, and screen out some inefficient nodes.

The realization method of technical scheme of the present invention is as follows:

1. Node design

Divide the sensing area of each node into eight areas, and each area has the same size. The perception of each region is independent of each other, and two regions are opened for perception in each round of perception. The design of the nodes is shown in Figure 2.

Each node sets a preferred location, adjusts the preferred location according to the request perceived by the node and the transmission of other nodes, and wakes up the nodes with the preferred location within the communication range, and transmits the stored request information. The attributes of the nodes are as follows:

(Nid, N_X, N_Y, Rc, Rs, PreOri, Pwake, Per, Sleep, RSF, Wtime, Em)

Among them, Nid represents the node number; N_X and N_Y represent the geographical coordinates of the node; Rc and Rs represent the communication radius and perception radius respectively; PreOri represents the number of the preferred orientation; Pwake represents the wake-up probability of the node; Per and Sleep represent the perception and sleep flags respectively , and satisfy the following relationship:

RSF represents the perception marks of eight orientations; Wtime represents the waiting time of eight search areas; Em is the energy of nodes.

For each node, the present invention designs a request save table (request save table, RST) to store and transmit service request information. In each round, N _i will perceive or receive a service request; the node will record the request information in RST _i . In order to reflect the current perception situation, the stored information will only be memorized for 3 rounds of perception time. Recorded information is represented in a node as a collection. Its elements are as follows:

(Rid,R_X,R_Y,Ori,Rw,t)

Among them, Rid is the number of the service request; R_X and R_Y represent the geographical coordinates of the service request; Ori represents the orientation number of the perceived Rid; Rw represents the way to obtain the service request; t represents the time the request is stored in the node.

2. The preferred orientation of the node

Design a favorite orientation for each node. During the sensing process, Ni prioritizes sensing service requests with favored orientation.

As shown in Figure 3, the Hopfieldneural network (HNN) is used to determine the preference orientation of nodes.

First update the request coordinates in RST _i according to the following formula; then mark the corresponding position as 1 according to the coordinates in the lattice;

Where R_X and R_Y represent the request coordinates in RST _i , N_X and N_Y represent the coordinates of node N _i , and Rs _i represents the perception radius of node N _i .

By converting the lattice into a vector noi row by row, the noi is simulated with the lattice v1-v8 of the 8 search regions using HNN. In order to avoid over-fitting, the result of HNN is corrected using the probability p, where the probability p is a probability set separately to prevent the model from over-fitting. The result obtained is the PreOri for the next round of N _i .

3. Regional weight of nodes

In order to prevent partial search caused by too concentrated preferences, the value of the selection operator of the node area is designed according to the search frequency and waiting time of each node. The formula is shown in the following figure:

表示节点N_i的搜索区域k的等待时间。S(i, k) represents the value of the selection operator of the kth search area of node N _i , where Rnum represents the number of requests stored in node N _i ; Freq(i, k) represents the total number of requests in area k; PFreq(i,k) indicates the total number of requests whose area is k and Rw is 1; RSF(i,k) indicates the perception flag of area k in node N _i ;

Indicates the waiting time of node N _i 's search area k.

4. Orientation-sensitive perception mode of nodes

The present invention proposes an orientation-sensitive sensing mode, the core idea of which is to continuously use preferred orientations and selection operators, and selected orientation-aware service requests.

Referring to Fig. 4, the region of the preferred direction and the region with the highest value of the selection operator in N _A are indicated as the arrow region and the shaded region. When node N _A is in working state, it turns on the arrow area and the shaded area in turn to perceive service requests R1 and R2. Node NA then receives RST _B and _RSTc from node _NB and node _NC . Node N _A adds information to RST _A according to the received and perceived information. Then, nodes N and _A respectively determine their own preferred orientation and calculate the selection operator of the search area. After sensing, the node N _A will send the RST _A to the reachable nodes _ND and N _E within the two search areas.

5. Node state switching based on convolution method

According to the following formula, the node's ability value Ab is calculated using the value of the selection operator S.

表示节点N_i中八个搜索方位中最少的等待时间。where M is a control factor that controls the degree of difference between Abs. T represents the time when the model perceives the current time from the beginning, S(i,m) represents the value of the selection operator of the mth search area of node N _i ; Freq(i,m) represents the request in area m of node N _i total;

Indicates the minimum waiting time among the eight search azimuths in node N _i .

The wake-up probability of a node is calculated by the following formula, and the probability is controlled between [a,b]:

Among them, Ab _i represents the ability value of node N _i , and Ab _min and Ab _max represent the smallest ability value and the largest ability value among all nodes.

Then use the convolution method to screen inefficient nodes, obtain the Ab of all nodes, and combine them into a 12×12 lattice, called feature map C1. Then proceed as follows:

进行卷积，得到名为C2的10×10大小的特征映射，该特征映射得到每个3×3区域的平均能力值。Step a: Use a 3×3 convolution kernel

Convolution is performed to obtain a feature map of size 10×10 named C2, which obtains the average capability value of each 3×3 region.

Step b: The mode pooling operation of the 2×2 norm is performed on C2. The obtained results are used to express the average efficiency of the 2×2 area. The resulting feature map is C3, as shown in Figure 8. This operation can better distinguish different regions by eigenvalues.

Step c: Obtain the feature map C3 through convolution and mode pooling. The corresponding relationship is shown in Figure 9. The feature value with smaller feature value is the area with weaker perception ability.

Step d: According to the corresponding relationship in step c, select the nodes with lower ability values in C1, and form their ability values into lattice C4.

Step e: Divide each C4 into 4 feature maps with a dimension of 2×2. Each feature map contains capability values of 4 nodes. A convolution operation is performed on each feature map. The convolution operation is the sum of the ability values of the remaining three nodes after removing their own ability values, as shown in Figure 10.

Step f: By comparing the ability value in C5, select the node with the smallest ability value to C6. C6 contains 4 results selected from each C4.

Step g: The node Nid recorded in C6 forms the final output result C7. According to Nid in C7, the model sets the recorded Nid's Pwake to 0.

6. Experimental results

In the experiment, a 400*400(M) IoT area is set up, and 144 sensor nodes are evenly distributed in the area. The performance of the nodes in the experiment is consistent. The communication range is 32M, the perception range is 16M, and the number of the initial favorite position is a random integer from 1 to 8. The perception range of a node is divided into 8 parts equally.

The attributes requested in the experiment are as follows:

(Rid,R_X,R_Y,Rq,Rt)

Where Rid is the serial number of the service request. R_X and R_Y represent the geographic coordinates of the service request. Rq and Rt represent the requested coverage radius and perception flag, respectively.

The number of independent experiments of each example is 30 times, and the algorithm adopted in the present invention is recorded as the OSPA algorithm, and the OSPA algorithm of the invention is compared with DSPA, NIA_BA, CAOP, and ISOS algorithms.

6.1 Energy Consumption Issues

For each algorithm, only 30% of the nodes are initially set to be awakened, and the number of requests is set to 1000, 1500, 2000, 2500 respectively. The results are shown in Figure 5, corresponding to the graphs a, b, c, and d in Figure 5 . It can be seen from the figure that the energy consumption of the OSPA algorithm is obviously better than that of the other four algorithms, which is mainly due to the reasonable division of the sensing range of the nodes in the present invention, which reduces the energy consumption of the sending and receiving modules. Adaptive wake-up probability and convolution method effectively reduce the number of working nodes. When the number of service requests increases, the gap between OSPA and other algorithms continues to widen. This shows that the OSPA algorithm is more competitive in discovering large-scale service requests.

6.2 Comparison experiment of perception rate

For each algorithm, only 30% of the nodes are initially set to be awakened, and the number of requests is set to 1000, 1500, 2000, 2500. The results are shown in Figure 6, corresponding to Figure 6: a, b, c, and d. The experimental results show that the perception rate of OSPA is the fastest, and can quickly increase to close to 1. This is because location-preferring fast search services request dense areas to improve awareness. Secondly, the selection operator increases the perception area, making the perception more global and improving the fault tolerance rate. Explain that in the environment of large-scale service requests, OSPA will show its advantages.

6.3 Number of node wakeups

For each algorithm, the initial setting is only 30% of the nodes are awakened, the number of requests is set to 1000, 1500, 2000, 2500, the results are shown in Figure 7, corresponding to Figure 7: a, b, c, d. It can be seen from the figure that the OSPA algorithm can effectively control the number of working nodes, which benefits from the implementation of adaptive wake-up probability. Second, the convolution method modifies the useless nodes in the working nodes to sleep state, further reducing the number of working nodes. It shows that in the large-scale Internet of Things scenario, the OSPA algorithm can effectively control the number of nodes.

The above-mentioned embodiment is an illustration of the present invention, not a limitation of the present invention, and any solution after a simple transformation of the present invention belongs to the protection scope of the present invention.

Claims (8)

1. A partition cooperative sensing method for large-scale Internet of Things, characterized in that: comprising the following steps: S1. Divide the perception area of the node, each area represents a perception orientation, and the perception of each orientation is independent of each other; S2. The state of the node is divided into two states of working and sleeping, and the mutual conversion of the two states is controlled by the wake-up probability of the node; S3. Each node has a request storage table to store the perceived request information and transmit the service request information; S4. Each node has a favorite location, and the favorite location is modified according to the perceived request information and the request storage table; S5. After the perception of the preferred orientation is completed, the node will calculate the selection operator of each orientation, and select the orientation represented by the largest selection operator for perception; S6. After a round of sensing operation is over, the node will wake up other nodes in the sensing position and transmit request information; S7. After the transmission operation is completed, each node will calculate the node's ability value, modify the wake-up probability through the node's ability value, and screen out some inefficient nodes. 2. a kind of partition cooperative perception method facing large-scale internet of things as claimed in claim 1, is characterized in that: the attribute of described node is as follows: (Nid, N_X, N_Y, Rc, Rs, PreOri, Pwake, Per, Sleep, RSF, Wtime, Em) Among them, Nid represents the node number; N_X and N_Y represent the geographical coordinates of the node; Rc and Rs represent the communication radius and perception radius respectively; PreOri represents the number of the preferred orientation; Pwake represents the wake-up probability of the node; Per and Sleep represent the perception and sleep flags respectively , and satisfy the following relationship:

RSF represents the perception marks of eight orientations; Wtime represents the waiting time of eight search areas, and Em is the energy of nodes. 3. A partition cooperative sensing method oriented to large-scale Internet of Things as claimed in claim 1, characterized in that: in step S1, each partitioned area has the same size. 4. A kind of large-scale Internet of Things-oriented partition cooperative sensing method as claimed in claim 1, characterized in that: in step S3, each round of sensing operation, node N _i will perceive or receive a service request, and node N _i will Record the request information in the request storage table RST _i ; the recorded information is represented as a set in the node, and its elements are as follows: (Rid,R_X,R_Y,Ori,Rw,t) Among them, Rid is the number of the service request; R_X and R_Y represent the geographic coordinates of the service request; Ori represents the orientation number of the perceived Rid; Rw represents the way to obtain the service request; t represents the time the request is stored in the node;

5. A kind of large-scale Internet of Things-oriented partition cooperative sensing method as claimed in claim 1, characterized in that: in step S4, during the sensing process, the node Ni perceives the service request with preference orientation, and the specific process is as follows: first Update the request coordinates in RST _i according to the following formula; then mark the corresponding position as 1 according to the coordinates in the lattice; use the HNN network to determine the preferred orientation of the node; convert the lattice into a vector noi by row, use The HNN network simulates noi and the lattice of eight search orientations divided in S1, and uses the probability p to modify the results of the HNN network with a small probability. The result obtained is the preferred orientation of the next round of node N _i

Where R_X and R_Y represent the request coordinates in RST _i , N_X and N_Y represent the coordinates of node N _i , and Rs _i represents the perception radius of node N _i . 6. A kind of large-scale Internet of Things-oriented partition cooperative sensing method as claimed in claim 1, characterized in that: in step S5, the value of the selection operator of the node area is determined according to the search frequency and waiting time of each node, The formula is as follows:

Among them, S(i,k) represents the value of the selection operator of the kth search area of node N _i , where Rnum represents the number of requests stored in node N _i ; Freq(i,k) represents the request in area k Total number; PFreq(i,k) indicates the total number of requests whose area is k and Rw is 1; RSF(i,k) indicates the perception flag of area k in node N _i ;

Indicates the waiting time of node N _i 's search area k. 7. a kind of partition cooperative perception method facing large-scale internet of things as claimed in claim 1, is characterized in that: utilize the value of selection operator S to calculate the capability value Ab of node according to the following formula:

Among them, M is the control factor that controls the degree of difference between Ab, T represents the time from the beginning of the model to the current perception, S(i,m) represents the value of the selection operator of the mth search area of node N _i ; Freq( i, m) represents the total number of requests in area m in node N _i ;

Indicates the minimum waiting time among the eight search orientations in node N _i ; after calculating the selection operator, open the orientation represented by the largest selection operator for perception. 8. A kind of large-scale IoT-oriented partition cooperative sensing method as claimed in claim 7, characterized in that: in step S7, the wake-up probability of the node is calculated by the following formula, and the wake-up probability is controlled at [a, b] between:

Among them, Ab _i represents the ability value of node N _i , and Ab _min and Ab _max represent the smallest ability value and the largest ability value among all nodes; Then use the convolution method to screen inefficient nodes, obtain the Ab of all nodes, and combine them into a 12×12 lattice, called feature map C1; then proceed as follows: Step a: Use a 3×3 convolution kernel

Perform convolution to obtain a feature map of size 10×10 named C2, which obtains the average ability value of each 3×3 region; Step b: The 2×2 normative mode pooling operation is performed on C2, and the obtained result is used to represent the horizontal efficiency of the 2×2 region, and the obtained feature map is called C3; Step c: Obtain the feature map C3 through convolution and mode pooling; by comparing the data in C3, it can be determined that the feature value with a smaller feature value is an area with weaker perception; Step d: According to the corresponding relationship in step c, select the nodes with lower ability values in C1, and form their ability values into lattice C4; Step e: Divide each C4 into 4 feature maps with a specification of 2×2, and each feature map contains the ability values of 4 nodes; perform convolution operation on each feature map; convolution operation is to remove its own ability value The sum of the ability values of the remaining three nodes; Step f: By comparing the capability value in C5, select the node number with the smallest capability value to C6; C6 contains 4 results selected from each C4; Step g: The node number recorded in C6 forms the final output result C7. According to the node number in C7, the model sets the Pwake of the recorded node number to 0.

Images