KR101072448B1 - Wireless sensor network system and clustering method thereof - Google Patents

Abstract

The present invention relates to a wireless sensor network system and a method for designing the same, and specifically, to minimize the energy transmission distance between sensor nodes in consideration of resource constraints of the sensor network, thereby minimizing the energy consumption of the sensor nodes to maximize the lifespan. A wireless sensor network system and a clustering method thereof. As described above, the present invention provides a clustering method of a wireless sensor network system clustered by a cluster head node connected to a sink node and a plurality of sensor nodes transmitting data to the cluster head node. Inputting, setting parameters, and randomly generating a set number of particles per swarm to perform an initial solution generation process; And calculating an evaluation value of each of the randomly generated particles, evaluating a solution in consideration of the calculated evaluation value, and calculating a probability of becoming a cluster head node by applying the evaluation value to a sigmoid. Using the sigmoid function, we derive the solution of the next generation and update it with the best value among the particles up to the current generation and the best particle among the best particles up to the current generation, and reflect it to the next generation to converge on the optimal solution. Provided are a clustering method of a wireless sensor network system for determining the number and location of cluster head nodes by performing a step, and a wireless sensor network to which the method is applied.

Description

Wireless sensor network system and its clustering method {WIRELESS SENSOR NETWORK SYSTEM AND CLUSTERING METHOD THEREOF}

The present invention relates to a wireless sensor network system and a method for designing the same, and specifically, to minimize the energy transmission distance between sensor nodes in consideration of resource constraints of the sensor network, thereby minimizing the energy consumption of the sensor nodes to maximize the lifespan. A wireless sensor network system and a clustering method thereof.

Micro-wireless sensor networks, which are being actively researched, are expected to be very useful in mobile computing and communication. This technology can be used in various fields as a representative of the ubiquitous environment, and is currently used mainly in military, environmental surveillance, and polar monitoring.

Ubiquitous is an information and communication environment that allows users to connect to the network freely regardless of the location of the network or computer. Each recent report demonstrates the unlimited marketability of wireless sensor networks. In February 2007, EE Times cited market research company INSTIT, which reported that the global RFID market is expected to grow 25 times in 2010 to 33 billion units in 2010. In addition, in the 'NFC Phone Market Forecast and Company Trends' report published by the Korea Institute of Information and Communications Policy (KISDI) on February 12, 2008, the domestic mobile RFID market is estimated at 26.9 billion won in 2007, with an annual average increase of 196%. It is expected to form a large market of 70 billion won a year. In such a ubiquitous environment, surrounding physical information is collected through sensors existing on the sensor network, and the collected data is collected to the sink through the air interface. The collected data is transmitted to a remote server or user through a broadband network such as the Internet or cellular. In a ubiquitous environment, the wireless sensor terminal may be a terminal device such as a notebook, mobile phone, PDA, etc., carried by a user. The RFID system consists of a database, a reader, and a tag, which can be viewed as a wireless sensor terminal. This can be applied to a wide range of everyday life and industrial systems, such as car navigation, location-based content, mobile services, transportation cards. However, wireless sensor terminals have high resource constraints such as limited computing power such as low power and low memory, and must be implemented at low cost.

On the other hand, the sensor network collects data in the field and transmits it to Sink, and development needs are increasing due to the needs in various fields of society.

In general, a sensor network is a structure in which a plurality of sensors disposed at a long distance are grouped together and connected to each other as a node. Here, connected means that the signal is coupled to the wireless communication.

For example, a monitoring system by a sensor node is connected to a main computer with a monitor over a communication network consisting of wireless communication, satellite communication, and the like. It is in a constitution. The sensor node has peripheral components such as a microprocessor, memory connected thereto, RAM, ROM, memory, transceiver, and the like. In the early days, sensor networks were mainly used for unmanned reconnaissance systems that monitor military trends in areas that were difficult to access, but recently, their applications are intelligent traffic systems, automatic control of production processes, It is expanding to remote sensing of patient status and environmental control in intelligent buildings.

In this sensor network system, a plurality of spatially distributed sensor nodes are configured. Each of the distributed sensor nodes enters a monitoring operation state and causes an abnormal shape or transmits the status of the monitoring area to the data collection device through a communication network at a predetermined time. Therefore, sensor nodes used in these fields are mainly operated by a limited energy source, and there is a need to operate without replacement of power for months to years.

This necessity requires the development of efficient sensors that can operate with minimal limited computing power, as well as efficient deployment and grouping of wireless sensor networks to connect the sensors.

However, many researches conducted in the past have been conducted only for academic research purposes, and technology development that can keep pace with the rapidly changing related market is urgent as it does not proceed to commercially available technology to be applied to the actual industry.

The present invention has been made in view of the above problems, and in consideration of the resource constraints of the sensor network, by minimizing the data transmission distance between the sensor nodes, the wireless sensor network system can minimize the energy consumption of the sensor nodes to maximize the lifetime. And its clustering method.

Another object of the present invention is to provide a wireless sensor network system and its clustering method that can search for an optimal solution even in a dynamic environment by using a heuristic algorithm Particle Swarm Optimization (PSO) for optimal clustering of sensor nodes. In providing.

The objects of the present invention are not limited to the above-mentioned objects, and other objects not mentioned can be clearly understood by those skilled in the art from the following description.

According to an aspect of the present invention, there is provided a wireless sensor network system including: at least one cluster having sensor nodes other than cluster head nodes to collect and condense sensing data; And a sink node that finally collects the sensed data abbreviated from the cluster head node, wherein the cluster head node derives a next generation solution using a sigmoid function to obtain a PSO. : Particle Swarm Optimization (Particle Swarm Optimization) algorithm calculates particle-specific evaluation values, and the number and position of the particles are determined by the optimal solution obtained by considering the calculated evaluation values, and the sensor nodes are connected to the nearest cluster head node. It is characterized by.

Preferably, the evaluation value, input the position data of the sink node and each sensor node, set the parameters of the following equations (1) and (2), and by the number set per swarm (swarm) to generate the initial possible naval It is characterized in that the particles of the randomly generated to calculate the evaluation value of each particle.

(Equation 1)

(Equation 2)

Where c ₁ and c ₂ are the weights of the stochastic accelerations toward which each particle is directed to pbest ⁱ _j (best of particle j up to generation i) and gbest ⁱ (best of pbest ⁱ _j up to generation ⁱ ), r ₁ and r ₂ are random constants in [0, 1], and ω is the inertia weight.

Preferably, the evaluation value is characterized in that calculated using the following equation (3).

(Equation 3)

Where TD is the total distance when all sensors are initially connected to the sink node, N is the number of all nodes, a is the weight, index k is the k-th clustering solution, and d _k is the k-th clustering solution. The total distance is shown, and the number of cluster head nodes at that time is _denoted by H _k .

Preferably, the optimal solution is to calculate the probability of becoming a cluster head node by applying the evaluation value, and then reallocate '0' and '1' by using the sigmoid function, and repeats generations. It is characterized in that the excellent solution is updated to the best value among the particles (pbest, gbest) and reflected in the next generation to converge.

According to another aspect of the present invention, there is provided a clustering method of a wireless sensor network system, comprising: a cluster head node connected to a sink node and a clustering method of a wireless sensor network system clustered into a plurality of sensor nodes transmitting data to the cluster head node. Inputting the sink node and position data of each sensor node, setting a parameter, and randomly generating a predetermined number of particles per swarm to perform an initial solution generation process; And calculating an evaluation value of each of the randomly generated particles, evaluating a solution in consideration of the calculated evaluation value, and calculating a probability of becoming a cluster head node by applying the evaluation value to a sigmoid. Using the sigmoid function, we derive the solution of the next generation and update it with the best value among the particles up to the current generation and the best particle among the best particles up to the current generation, and reflect it to the next generation to converge on the optimal solution. Perform the steps to determine the number and location of cluster head nodes.

Preferably, the step of converging to the optimal solution includes inputting the sink node and the position data of each sensor node, setting the parameters of Equations 4 and 5 below, and using a swarm to generate an initial possible naval solution. Randomly generating a set number of particles per step and calculating an evaluation value of each particle.

(Equation 4)

(5)

Where c ₁ and c ₂ are the weights of the stochastic accelerations toward which each particle is directed to pbest ⁱ _j (best of particle j up to generation i) and gbest ⁱ (best of pbest ⁱ _j up to generation ⁱ ), r ₁ and r ₂ are random constants in [0, 1], and ω is the inertia weight.

Preferably, ω of the inertia weight is a control parameter for adjusting the influence of the existing velocity on the current velocity, and a large value of ω is set at the beginning of the search process to enhance global search, and at the end of the search for local search. decreases ω.

Preferably, the evaluation value is characterized in that calculated using the following equation (6).

(6)

Where TD is the total distance when all sensors are initially connected to the sink node, N is the number of all nodes, a is the weight, index k is the k-th clustering solution, and d _k is the k-th clustering solution. The total distance is shown, and the number of cluster head nodes at that time is _denoted by H _k .

Preferably, the step of converging to the optimal solution includes calculating the probability of becoming a cluster head node by applying the evaluation value, and then probably reallocating '0' and '1' using a sigmoid function. Steps.

According to another aspect of the present invention, a computer-readable recording medium includes inputting position data of a sink node and each sensor node to a computer, setting parameters, and setting a number of particles per swarm. Randomly generating and performing an initial solution generation process; And calculating an evaluation value of each of the randomly generated particles, evaluating a solution in consideration of the calculated evaluation value, and calculating a probability of becoming a cluster head node by applying the evaluation value to a sigmoid. Using the sigmoid function, we derive the solution of the next generation and update it with the best value among the particles up to the current generation and the best particle among the best particles up to the current generation, and reflect it to the next generation to converge on the optimal solution. Run the steps to determine the number and location of cluster head nodes.

In accordance with the above-described problem solving means, the present invention provides a PSO method which is recently applied to information communication technology and system design in order to reduce the energy consumption of sensor nodes and extend the life of sensor network by the optimal design of clustering of wireless sensor networks. Among the sensor nodes, the most suitable position and number of cluster head nodes connected to the sink node are determined, and the other sensor nodes are assigned to the nearest cluster head node, so that the transmission distance of the sensor network and the position of the cluster head node and By optimizing the number, the energy consumption of the sensor network can be reduced to extend the life.

In addition, data transfer speed and accuracy can be improved to reduce costs and increase customer satisfaction.

1 is an exemplary view showing a wireless sensor network system according to an embodiment of the present invention.
Figure 2 is an exemplary view showing a possible solution stored on a computer with respect to FIG.
3 is an exemplary diagram representing a possible solution for applying the PSO method according to an embodiment of the present invention.
4 is an exemplary diagram for explaining an example for generating x ₁ ² by applying a solution representation of x ₁ ¹ , pbest ₁ ¹ , gbest ¹ and the PSO method according to the present invention.
5 is an exemplary diagram arbitrarily explaining the positions of predecessors for 100 nodes.
6 is an exemplary diagram for explaining a process of convergence of evaluation values in 100 node wireless sensor networks.
7 is an exemplary diagram showing an optimal clustering result of 100 node wireless sensor networks.
8 is an exemplary diagram showing a position graph of clustering results of optimal / best clustering of 100 node wireless sensor networks;
9 is an exemplary view illustrating a convergence process of evaluation values in a 200 node wireless sensor network.
10 is a diagram illustrating a convergence process of evaluation values in 400 node wireless sensor networks.
Figure 11 is an exemplary view comparing the results of applying the conventional clustering optimization design method and the PSO method of the present invention with the result of the optimal solution of CPLEX.
12 is a flowchart illustrating a clustering method of a wireless sensor network system according to an embodiment of the present invention.

In the wireless sensor network system and its clustering method according to the present invention, in transmitting information collected from each wireless sensor node to the BS, the cluster head node is an intermediate information collection step in consideration of data transmission speed, power consumption efficiency, and communication distance. In order to determine the optimal number and location of sensor nodes and to design the optimal clustering of sensor nodes, we propose a technical scheme to search for optimal solutions in dynamic environments using heuristic algorithm PSO (Particle Swarm Optimization). .

In the following description, the basic concept of PSO is an evolutionary process of finding the optimal solution by sharing the previous experience of each particle and the information of the surrounding groups as the time progresses in the calculation of n-dimensional space. Algorithm.

In the following description, specific details of the wireless sensor network system of the present invention and its clustering method are presented to provide a more comprehensive understanding of the present invention, and the present invention may be readily implemented without these specific details and also by their modifications. It will be apparent to one skilled in the art.

Hereinafter, with reference to the accompanying drawings, preferred embodiments of the present invention will be described in detail, focusing on the parts necessary to understand the operation and action according to the present invention.

1 is an exemplary view showing a wireless sensor network system according to an embodiment of the present invention.

Referring to FIG. 1, the wireless sensor network system 100 according to the present invention includes a plurality of clusters 110, 120, 130, 140, and a sink node 20.

First, clusters 110, 120, 130, and 140 are configured independently of each other. Each cluster 110, 120, 130, 140 includes cluster head nodes 4, 7, 10, and 12 and sensor nodes 1 to 3, 5, 6, 8, 9, 11, 13 to 15. do. In such clusters 110, 120, 130, 140, the cluster head nodes 4, 7, 10, 12 receive from each sensor node 1-3, 5, 6, 8, 9, 11, 13-15. Data reduction is performed on the sensed data. The cluster heads 4, 7, 10, and 12 deliver the shortened sensing data to the sink node 20.

The sink node 20 collects the abbreviated sensing data received from each cluster head node 4, 7, 10, 12. The sink node 20 may be configured in plural in a wireless sensor network system.

The wireless sensor network system 100 described above uses the PSO, a heuristic algorithm that is widely used in information and communication technologies and systems, to find the optimal solution even in a dynamic environment. Perform clustering optimal design to determine the number and location of.

1. For clustering design PSO  Technology

The PSO algorithm searches and finds the best evaluation value of each of the j particles in the i th generation, stores the result in the pbest ⁱ _j (particle best) matrix, stores the result, and stores the best value in the ⁱ _th generation. Select pbest ⁱ _j with good evaluation and store it in gbest ⁱ (global best). The movement of each particle is pbest ⁱ _j Using the result stored in gbest ⁱ and the current position vector and velocity vector, it is expressed as Where pbest ⁱ _j is the best value of particle j up to generation ⁱ , and gbest ⁱ is pbest ⁱ _j up to generation ⁱ Means the best value.

c ₁ and c ₂ are the weights of the stochastic accelerations for each particle towards pbest ⁱ _j and gbest ⁱ . r ₁ and r ₂ are random constants in [0, 1]. ω is an inertia weight and is a control parameter for adjusting the influence of existing speed on the current speed. The value of ω is set to increase the global search early in the search process and ω for local search later in the search process. Decreases.

The modification of the position vector uses Equation 2 below using the current position and the corrected speed.

By repeating this process, we find the optimal solution.

In the wireless sensor network clustering design, the PSO method is applied to determine the number and location of cluster head nodes and to which cluster head nodes other sensor nodes are connected.

The sink node and the position data of each sensor node are input, and the parameters ω, c ₁ , c ₂ , r ₁ , and r ₂ of Equations ₁ and ₂ are set. In order to generate the initial possible navy, a random number of particles is set for each swarm, and the solution of each particle is evaluated using the evaluation function of Equation 3 below.

Where TD is the total distance when all sensors are initially connected to the sink node, N is the number of all nodes, a is the weight, index k is the k-th clustering solution, and d _k is the k-th clustering solution. The total distance is shown, and the number of cluster head nodes at that time is represented by H _k .

The next generation uses the Sigmoid function to probably rewrite 0 and 1 to reflect the previous solution, the solutions of pbest ⁱ _j and gbest ⁱ at appropriate ratios, prevent them from falling into local solutions, and increase the diversity of solution search. Assign. Over generations, the best solution is updated to pbest ⁱ _j , gbest ⁱ and reflected in the next generation to converge to the optimal solution.

The following describes the representation of the solution and the evaluation function, followed by the procedure for performing the PSO method.

2. Expression and Evaluation Functions of Solutions

Referring to the solution of the cluster-based hierarchical model used in FIG. 1, nodes 4, 7, 10, and 12 connected to the sink node 20 become cluster head nodes, and the remaining nodes are sensor nodes (1). 3, 5, 6, 8, 9, 11, 13-15).

Figure 2 is an exemplary view showing a possible solution stored in a computer with respect to FIG.

Referring to FIG. 2, since the cluster head node is connected to the sink node to transmit data, the cluster head node is assigned an index 0 of the sink node, and sensor nodes other than the cluster head node are located closest to their positions. It is connected to and assigned the number of the corresponding cluster head node.

 That is, sensor nodes 1, 2, 3, and 5 are connected to cluster head node 4 to receive 4, sensor nodes 6 and 8 are cluster head node 7, and sensor nodes 9 and 14 10. The cluster head node, sensor nodes 11, 13, and 14 are connected to the cluster head node 12 and are assigned the number of the corresponding cluster head node, respectively.

As described above, since the representation of the solution of FIG. 2 is easier in terms of information and interpretation of the attached solution, the present invention stores the solution in the form of FIG.

The solution representation of FIG. 3 was simply used as a solution representation because it is possible to simply indicate whether each sensor node is a cluster head node with '0' and '1', so that the BPSO optimization method used in the present invention can be smoothly applied. When storing the solution information, the form of FIG. 2 was used.

Meanwhile, in the present invention, Equation 3 is used as an evaluation function, and the weight a is 0.7 to give more weight to reduce the total distance, but the parameter value may be adjusted according to the importance of evaluation.

That is, maximizing the evaluation value of Equation 3 reduces the cluster head node if possible while reducing the total transmission distance.

3. PSO Design

The method of generating the initial solution of the PSO method must first determine whether or not it is a cluster head node for each sensor node. At this time, through random number generation, each node becomes a cluster head node with probability. Next, sensor nodes other than the sink node and the cluster head node are calculated from the cluster head nodes and assigned to the cluster head node located at the shortest distance.

That is, the sensor network design solution is represented by a value of [0, 1]. The clustering result of the generated solution is called x _j ⁱ , and whenever the generation progresses sequentially, the clustering result having the best evaluation value for each particle number is stored in pbest ⁱ _j . Similarly, clustering results with the best evaluation of pbest ⁱ _j are stored in gbest _i .

x ₁ ² is generated by applying x ₁ ¹ , pbest ₁ ¹ and gbest ¹ to Equations 1 and 2 above.

Compare the evaluation function f (x ₁ ¹ ) with f (x ₁ ² ) and store the better value in pbest ₁ ² . Thus ^{_{pbest 2 2, pbest 3 2,}} ... Find pbest _j ² and store the best of these values in gbest ² . In addition, x ₁ ³ is generated by applying x ₁ ² , pbest ₁ ² and gbest ² to Equations 1 and ² above. In this way, generation is performed to find the optimal solution.

The process of generating x ₁ ² by applying x ₁ ¹ , pbest ₁ ^1, and gbest ¹ to equations (2) and (3) can be explained by the following example.

If the evaluation function value ^{_{f (x 1 1) = 2}} , f (pbest 1 1) = 3, f the <Equation 1> and when If (gbest ¹⁾ = 5 la to create new x ₁ ² <Equation 2>, x ₁ ¹ , pbest ₁ ¹ , gbest ¹ can be calculated as shown in Equation 4 below how closely x ₁ ² similarly.

According to the probability according to each evaluation value, the probability that a specific node of x ₁ ² may become a cluster head node may be calculated as a value between 0 and 1, respectively.

That is, v _i ^{t + 1} of Equation 1 is derived by calculating the sum of probabilities that the genetic factor “1” (cluster head node) of each year is stochasticly accepted.

The sum of the probabilities of becoming a cluster head node v _i ^{t +1} is between 0 and 1, and between v _i ^{t +1} min and max v _min = -4, v _max = 4 (typically v _i The range of ^{t +1} can be converted into [-4, 4]) and the Sigmoid function can be applied to Equation 5 as follows.

In v _i ^{t + 1} The larger the value, the cluster head node, that is, the higher the "1" The higher the chance that the smaller the value of v _i ^{t + 1} are likely to be "0", not the cluster head node.

4 is an exemplary view for explaining an example of applying the solution representation of the expression x ₁ ¹ , pbest ₁ ¹ , gbest ¹ and the PSO method according to the present invention.

가 된다. 다시 말하면 x₁ ²의 5번째 노드가 클러스터 헤드 노드가 될 확률은 98.2%가 된다.
Referring to FIG. 4, ^first , if the fifth node of x ₁ ¹ , pbest ₁ ¹ , and gbest ¹ is the cluster head (that is, if the genetic factors are all “1”) as in ⓐ of FIG. 4, the fifth node of x ₁ ² If the probability of receiving the gene of 1 is 1, and converts it to a value between -4 and 4, v _i ^{t +1} of Equation ¹ becomes 4, and substitutes this in Equation 5

Becomes In other words, the probability that the fifth node of x ₁ ² becomes the cluster head node is 98.2%.

이 된다. 다시 말해서 x₁ ²의 6번째 노드가 클러스터 헤드 노드가 될 확률은 1.7986%가 되는 것이다. As shown in ⓑ of FIG. 4, if the sixth node of x ₁ ¹ , pbest ₁ ¹ , and gbest ¹ is not a cluster head node (ie, all of the genes are “0”), the gene of the sixth node of x ₁ ² is set to 1. If the probability is 0 and converted to a value between -4 and 4, v _i ^{t +1} of Equation 1 becomes -4, and this is substituted into Equation 5

Becomes In other words, the probability that the sixth node of x ₁ ² becomes the cluster head node is 1.7986%.

과 같은 계산 과정을 거쳐 0.8이 되고, -4와 4사이의 값으로 변환하면 상기 <수학식 1>의 v_i ^{t +1}이 2.4가 되어, 이를 상기 <수학식 5>에 대입하면

가 된다. 다시 말해서 x₁ ²의 7번째 노드가 클러스터 헤드 노드가 될 확률은 91.6827%가 된다.As shown in ⓒ of FIG. 4, if the 7th node of x ₁ ¹ is not a cluster head node and the pth ₇ ¹ node of pbest ₁ ¹ and gbest ¹ is a cluster head node (ie, the genetic factor of x ₁ ¹ is “0”, pbest ₁ ¹ , if the best factor of gbest ¹ is "1", the probability of getting the gene of the 7th node of x ₁ ² as 1

Through the calculation process as follows, the value is 0.8, and when converted to a value between -4 and 4, v _i ^{t +1} of Equation 1 becomes 2.4, and this is substituted into Equation 5 above.

Becomes In other words, the probability that the seventh node of x ₁ ² becomes the cluster head node is 91.6827%.

Through this process, the solution of particle 1 x ₁ ² , x ₁ ³ ,. Is generated and pbest ₁ ⁱ with the best evaluation value up to generation ⁱ is updated. Thus, if one is pbest ⁱ _j j i to the three updates for each particle, the best evaluation value among pbest ⁱ _j is determined to gbest ^i.

The experimental results of applying the clustering problem of the sensor network through the wireless sensor network system and the clustering method thereof according to the present invention were analyzed. The experiment was performed by creating a network of 100, 200, and 400 nodes. In addition, the sink nodes were located at the origin (0, 0) of each network in the center to compare the results.

In the experiment, the following initial parameter values were used to apply the PSO method. The number of particles generated per generation generates 30 solutions. Inertia weight 2.5, pbest weight 3.3, and gbest weight 0.2 of all particles of Equation 1 were applied. These parameter values were selected through several experiments.

The present invention generates initial data by generating random numbers at the position of each node. In the case of 100 nodes (including sink nodes), randomly generated 1 to 99 for the X and Y coordinates of each node, and all nodes were not duplicated.

5 is a diagram in which positions of predecessors are arbitrarily assigned to 100 nodes, and are initial data used in the present invention.

As a result of applying the PSO method to 100 nodes, the convergence process as shown in FIG. 6 was clearly seen, and when the sink nodes were (0, 0) and (50, 50), the evaluation values were found to be CPLEX, respectively. 4159.84 and 1774.39, the same as the best solution, was able to find the best solution.

7 and 8 (a) and (b) are optimal results of a 100 node wireless sensor network, showing how many cluster head nodes are located, where they are located, and which cluster head nodes are assigned to each sensor. If the sink node is (0, 0), the cluster head node is 3 (73, 81), 16 (82, 35), 17 (18, 95), 32 (74, 18), 38 (36, 7) 11 nodes, 41 (34, 29), 46 (17, 38), 91 (11, 4), 92 (52, 43), 96 (50, 67), 98 (12, 67), were selected.

If the sink node is (50, 50), the cluster head node is 2 (81, 17), 4 (44, 70), 7 (36, 14), 16 (82, 35), 17 (18, 95). , 24 (47, 54), 28 (85, 64), 33 (58, 50), 46 (17, 38), 48 (33, 30), 53 (65, 11), 58 (58, 69), 63 (27, 66), 64 (73, 89), 67 (12, 10), 82 (32, 46), 84 (7, 73), 90 (45, 30), 92 (52, 43) node 19 The dog was chosen.

If the sink node is (50, 50), the cluster head nodes are 2 (81, 17), 4 (44, 70), 7 (36, 14), 16 (82, 35), 17 (18, 95). , 24 (47, 54), 28 (85, 64), 33 (58, 50), 46 (17, 38), 48 (33, 30), 53 (65, 11), 58 (58, 69), 63 (27, 66), 64 (73, 89), 67 (12, 10), 82 (32, 46), 84 (7, 73), 90 (45, 30), 92 (52, 43) node 19 The dog was chosen. The remaining nodes were assigned to the closest cluster head node of the selected cluster head nodes.

In FIGS. 8A and 8B, it can be seen that the cluster head nodes are evenly distributed around the sink nodes of the origin and the center point.

FIG. 9 shows the apparent convergence form of the evaluation values of each experiment when 200 nodes and FIG. 10 have sink nodes located at the origin or the center of 400 nodes.

FIG. 11 shows the clustering optimization of ACO (name: Optimal Design of Wireless Sensor Network using Ant Group Optimization Method) by Sung-Soo Kim and Seung-Hyun Choi on the Clustering Optimization Problem of Sensor Network, Korean Institute of Management Science, Vol. 26, No. 3, 11, 2009, pp.55- 65.) and the results of applying the PSO method proposed in the present invention are compared with the results of the CPLEX optimal solution.

The results of the PSO method were much better than those of the ACO method in all 100, 200, and 400 node cases. In particular, for 100 nodes, the PSO method was able to find the optimal solution.

Based on the data of 100, 200, 400 nodes generated in the present invention, the sink node is the origin (0, 0) and the center point. In all 100, 200 and 400 nodes, the goodness-of-fit value was analyzed to be relatively better than the center point when the sink node is the origin, and the value close to the optimal solution was found.

In addition, when compared to the ACO results presented by Sung-Soo Kim and Seung-Hyun Choi, as shown in Fig. 11, the 100-node problem was able to find the optimal solution, and the 200 and 400 node problems were also significantly improved compared to the ACO result. I could do it.

As with the results of ACO, we could search for a clustering solution with better fit than when the sink node is at the origin.

12 is a flowchart illustrating a clustering method of a wireless sensor network system according to an embodiment of the present invention.

Referring to FIG. 12, first, a sink node and position data of each sensor node are input (S1201). For example, as shown in FIG. 5, the positions of the sensors are arbitrarily assigned.

After that, a parameter for moving particles is set (S1203). At this time, the weight of the parameter used to apply the PSO method is set.

Next, to generate an initial possible navy, a random number of particles set per swarm is randomly generated (S1205). Here, how to create the initial possible navy must first determine whether or not it is a cluster head node for each sensor. At this time, through random number generation, each node becomes a cluster head node with probability. Next, sensor nodes other than the sink node and the cluster head node are calculated from the cluster head nodes and assigned to the cluster head node located at the shortest distance. That is, the sensor network design solution is represented by the value of [0, 1].

After the initial possible naval is created, the solution of each particle is evaluated using the evaluation value (S1207). At this time, the evaluation value for each particle uses the above Equation 3.

After that, the solution generated for each particle is updated to pbest of each particle, and the solution having the best evaluation value among all particles is updated to gbest (S1209 to 1215).

After performing the above process, it is determined whether the end condition is reached (S1217). In this case, the termination condition is, for example, terminated when the process of evaluating and optimizing the solution for 4000 generations.

If the determination result is not the end condition, the next generation proceeds to steps 1219 to 1223 (S1219 to S1223), reflects the previous solution, the solution of pbest and gbest at an appropriate ratio, and generates a new solution using the sigmoid function. .

That is, the probability of becoming a cluster head node is calculated by applying evaluation values f (x _j ^{i −1} ), f (pbest _i ^{j −1} ), and f (gbest _j ^{i −1} ) (S1219).

Thereafter, the probability range [0,1] is proportionally extended to [Vmin, Vmax] (S1221), the next generation solution is derived by applying the Sigmoid function (S1223), and the process returns to the above-described step 1207 (S1207). Evaluate the newly created solution for each particle in the generation using the evaluation value, compare the existing solution with the newly created solution for each particle, update the solution with the better evaluation value to the pbest of the particle, and The solution with the best evaluation value among the solutions is updated to gbest (S1209 to S1215). The same method is used to generate the next generation of solutions.

After this process, if the end condition is satisfied, the process ends. Otherwise, the above process is repeated. Here, the termination condition is the termination condition of the number of iterations of the experimental data obtained by repeating a predetermined number of times for the optimal design of clustering.

While the present invention has been described in connection with what is presently considered to be the most practical and preferred embodiment, it is to be understood that the invention is not limited to the disclosed embodiments, but is capable of various modifications within the scope of the invention. Therefore, the scope of the present invention should not be limited by the illustrated embodiments, but should be determined by the scope of the appended claims and equivalents thereof.

Claims (10)

At least one cluster having sensor nodes other than the cluster head node to collect and condense sensing data; And
And a sink node that finally collects the sensing data abbreviated from the cluster head node.
The cluster head node derives a next generation solution using a sigmoid function to calculate particle-specific evaluation values through a particle swarm optimization (PSO) algorithm. The number and position are determined by an optimal solution obtained in consideration of the estimated evaluation value, wherein the sensor node is connected to the nearest cluster head node.
The method of claim 1, wherein the evaluation value,
Input the sink node and the position data of each sensor node, set the parameters of Equation 1 and 2 below, and randomly generate the set number of particles per swarm to generate initial possible naval Wireless sensor network system, characterized in that for calculating the evaluation value of the particles.
(Equation 1)

(2)

Where c ₁ and c ₂ are the weights of the stochastic accelerations toward which each particle is directed to pbest ⁱ _j (best of particle j up to generation i) and gbest ⁱ (best of pbest ⁱ _j up to generation ⁱ ), r ₁ and r ₂ are random constants within [0, 1], ω is the inertia weight, Vi ^t is the change in time t, Vi ^{t + 1} is the change in time t + 1, Xi ^t is the position in time ^t , Xi ^{t + 1} means the position at time t + 1.
The method according to claim 1 or 2, wherein the evaluation value is
Wireless sensor network system, characterized in that calculated using the following equation (3).
(3)

Where TD is the total distance when all sensors are initially connected to the sink node, N is the number of all nodes, a is the weight, index k is the k-th clustering solution, and d _k is the k-th clustering solution. It represents the total distance, and means the number of cluster head nodes at that time as H _k .
The method of claim 3, wherein the optimal solution is
After calculating the probability of becoming a cluster head node by applying the above evaluation value, it is possible to reallocate '0' and '1' by using sigmoid function. Wireless sensor network system characterized in that the convergence is updated to the value (pbest, gbest) reflected in the next generation.
A clustering method of a clustered wireless sensor network system clustered by a cluster head node connected to a sink node and a plurality of sensor nodes transmitting data to the cluster head node,
Inputting the sink node and position data of each sensor node, setting parameters, and randomly generating a predetermined number of particles per swarm to perform an initial solution generation process; And
Calculate the evaluation value of each randomly generated particle, evaluate the solution in consideration of the calculated evaluation value, calculate the probability of becoming a cluster head node by applying the evaluation value, and sigmoid Using the (Sigmoid) function to derive the solution of the next generation, update it with the best value among the particles up to the current generation and the best particle among the best generation up to the current generation, and reflect it to the next generation to converge on the optimal solution. The clustering method of the wireless sensor network system, characterized in that for determining the number and location of the cluster head node.
6. The method of claim 5, wherein converging to the optimal solution comprises:
Input the sink node and the position data of each sensor node, set the parameters of Equation 4 and 5 below, and randomly generate the set number of particles per swarm to generate initial possible naval Comprising a step of calculating the evaluation value of the particle clustering method of a wireless sensor network system.
(4)

(5)

Where c ₁ and c ₂ are the weights of the stochastic accelerations toward which each particle is directed to pbest ⁱ _j (best of particle j up to generation i) and gbest ⁱ (best of pbest ⁱ _j up to generation ⁱ ), r ₁ and r ₂ are random constants within [0, 1], ω is the inertia weight, Vi ^t is the change in time t, Vi ^{t + 1} is the change in time t + 1, Xi ^t is the position in time ^t , Xi ^{t + 1} means the position at time t + 1.
The method of claim 6, wherein ω of the inertia weight is
As a control parameter to control the influence of the existing speed on the current speed, the wireless sensor is characterized by setting a large value of ω to enhance global search at the beginning of the search process and reducing ω to search the area at the end of the search. Clustering method of network systems.
The method according to claim 5 or 6, wherein the evaluation value,
Clustering method of a wireless sensor network system, characterized in that calculated using Equation 5.
(5)

Where TD is the total distance when all sensors are initially connected to the sink node, N is the number of all nodes, a is the weight, index k is the k-th clustering solution, and d _k is the k-th clustering solution. It represents the total distance, and means the number of cluster head nodes at that time as H _k .
6. The method of claim 5, wherein converging to the optimal solution comprises:
Calculating a probability of becoming a cluster head node by applying the evaluation value, and then reallocating '0' and '1' by using a sigmoid function. Clustering method.
Inputting position data of a sink node and each sensor node to a computer, setting parameters, and randomly generating a predetermined number of particles per swarm to perform an initial solution generation process; And
Calculate the evaluation value of each randomly generated particle, evaluate the solution in consideration of the calculated evaluation value, calculate the probability of becoming a cluster head node by applying the evaluation value, and sigmoid Using the (Sigmoid) function to derive the solution of the next generation, update it with the best value among the particles up to the current generation and the best particle among the best generation up to the current generation, and reflect it to the next generation to converge on the optimal solution. A computer-readable recording medium having recorded thereon a program for determining the number and location of cluster head nodes by executing a.

Images