Comparative Study of Various Cluster Formation Algorithms in Wireless Sensor Networks

Transcription

1 Comparative Study of Various Cluster Formation Algorithms in Wireless Sensor Networks Zhan Wei Siew, Yit Kwong Chin, Aroland Kiring, Hou Pin Yoong and Kenneth Tze Kin Teo Modelling, Simulation & Computing Laboratory, Material & Mineral Research Unit School of Engineering and Information Technology Universiti Malaysia Sabah Kota Kinabalu, Malaysia Abstract Wireless sensor networks (WSNs) is one of the rising technologies that will be widely used in the future. It is a challenging problem to select suitable cluster heads (CHs) in WSNs among large number of sensor nodes as the size affects the network lifetime performance. To achieve an energy efficient clustering protocol, the selection of CH should consider various critical parameters. There are a few methods in CH formation such as, low-energy adaptive clustering hierarchical (LEACH) protocol which randomly rotate CHs, fuzzy logic in base station (BS) for suitable CH selection and particle swarm optimization (PSO) to select suitable set of CH. In this paper, simulation results for LEACH protocol, fuzzy logic based clustering protocol and adaptive PSO based clustering protocol will be demonstrated and analyzed. Comparison of performance metrics such as network lifetime, energy consumes per round and data received by BS will be discussed in the paper. Index Terms LEACH, fuzzy logic, adaptive-pso, CH, wireless sensor networks. I. INTRODUCTION Wireless sensor technologies are widely used in military, healthcare monitoring and high end industrial sectors [1]. In the near future, the linking of wireless sensor networks all over the world will form a global monitoring system. An application example, users can access to weather station located on the other side of the globe for updated information. The usage of wireless sensor networks is not only limited to that specific sector. The large numbers of potential users in public sectors can contributed and accelerate towards the development of wireless sensor network global monitoring system. Basically, wireless sensor networks contain hundreds or thousands of sensor nodes. Large number of sensor nodes increases the difficulty to form energy efficient networks [2]. Sensor nodes which overhear the wireless channel will drain the battery faster. Therefore, it is essential to design a smart and energy efficient sensor node. The sensor nodes should also be cost efficient and robust towards fault tolerance while maintaining the least energy consumption [3]. In addition, the sensor nodes should be able to perform self-management and collaboration between other sensor nodes. In environmental monitoring, clustering protocol is simple yet efficient. Since the sensor nodes are randomly deployed over sensing fields. The chances of having sensor nodes that are close to each other are very high. It is possible that, even the environment data from two sensor nodes is almost the same. Furthermore, the diversity for parameters is small in the same area. Data aggregations in clustering protocols provide an energy efficient method for data collection. Each cluster member transmits one packet to the CH, and then only one combined message packet will be transmitted to BS by CH. In essence, the collected data will go through data aggregation process, and then only one packet message will be produced. In nodes communication inside a network, network coding can reduce transmission count and transmission energy usage [4]. In clustering protocol, the selected CH will transmit the sensed data to BS instead of every sensor node. Thus, heavy work load is concentrated in the CH due to information collection from cluster members and transmission to BS. In other words, cluster formation will drastically affect the CH energy. Therefore, a method to randomly rotate the role of CH to evenly distribute the work load has been introduced in [5]. By using this method, the selected CHs will be possibly located near to each other or the distribution of CH will not be even. The contradicting solution and problem lies in the randomness in CH selection [6]. Introduction of artificial intelligence in CH selection shows an improvement of network lifetime compared to classical random selection method. An energy efficient CH formation should consider critical parameters such as sensor node remaining energy, sensor node coordinates and sensor node density. The use of artificial intelligence with parameters consideration eases the process of CH selection. In previous works, fuzzy logic is used to select suitable CH among sensor nodes in the network based on critical parameters [7]. Gupta s literature shows an improvement of network lifetime over LEACH protocol, but it only selects one CH in each round [8]. Chin described the use of fuzzy logic to harvest energy from solar panel, which can be use for sensor nodes with solar power [9]. It is complicated to select few strategic CHs among large number of sensor nodes [1]. By using swarm intelligence, it can search for the best CHs formation through an iterative process [11]. Through modeling of the CH selection problem

2 into PSO algorithm, the algorithm will explore the given solution and each particle in the algorithm co-operate with each others to search for the best solution [12]. Adaptive PSO further reduce the iterative cycle compared to basic PSO [13]. II. RELATED WORK A. LEACH Protocol LEACH protocol aims to evenly distribute the energy load or work load among sensor nodes in the network. The use of Eq. 1, provides each sensor node with a global reference that is used for CH election. First, each sensor node generate a random number between to 1, if the generated number is less than a threshold value given by Eq. 1, the corresponding node will become a CH for the current round. p, if n G 1 T ( n) = 1 p ( r mod ) (1) p, otherwise Where p is the desired percentage of CHs, r is current round and G is the set of sensor nodes which have never been CHs in the last 1 / p rounds. LEACH protocol prevents the work load from concentrating in individual sensor nodes by randomly rotating among CHs. Sensor nodes will self-elect to become CH based on Eq. 1. The threshold value increase over round, it reset to original setting when the cycle is reached. By doing so, sensor nodes in the networks will become cluster nodes at least one in the same cycle. The structure of LEACH protocol can be divided to: Set-up phase Advertisement phase Cluster set-up phase Steady phase Schedule creation Data transmission In set-up phase, elected CH will broadcast an advertisement message and then neighboring sensor nodes will join into the cluster based on receive signal strength. For steady phase, CH will go through schedule creation phase, creating communication time slots for cluster members. In data transmission phase, sensor nodes start to transmit data to CH based on the given time slots. Lastly, the CHs will transmit aggregated packet message to the BS. B. Overview of Fuzzy Logic Fuzzy logic is a problem solving control system that provides a simple way to obtain definite result only based on imprecise or incomplete information. The easy to use feature of fuzzy logic influences various disciplines from high end industrial such as aerospace to small home applications such as washing machines. Figure 1 shows the four elements in fuzzy logic, fuzzification, fuzzy rule base, inference engine and deffuzification. Fuzzification is the element that transforms the system inputs which are crisp x y Fuzzification Rule Base Inference Engine Defuzzification Fig. 1. The four elements in the operation of fuzzy logic. values into fuzzy sets. For example, value x and y are converted into linguistic fuzzy sets using fuzzy membership function. Fuzzy rule base evaluation is a compilation of IF- THEN rules that contains the experience and expertise knowledge of the model. The inference engine is the main element that formulates logical decisions based on the fuzzy rule base and unification of the fuzzy rule base into fuzzy linguistic output. Lastly, defuzzification transforms the fuzzy linguistic set obtained by the inference engine into a single crisp value denoted by z. For defuzzification process, Eq. 2 can be use to transform fuzzy set in the inference engine to the single crisp value. Centre of area (COA) is a method that is used to calculate the final result from the output membership function. output = ( x. ua ( x) dx xdx) (2) C. Overview of PSO PSO is a swarm intelligence algorithm that is inspired by mimicking the behavior of bird flocking. The theory steps that underlines PSO are evaluate, compare and imitate [14]. In evaluation stage, the solution will be rated base on pre defined measure schemes. For compare stage, each solution will compare themselves with their neighbors based on the evaluation performance. Lastly, in the imitate stage. The solution imitates only neighbors who are better compared to themselves. Basically, PSO begins with a group of particles which are randomly generated. Each particle will go through evaluation process to obtain the fitness value. The particle aims to search for the optimum solution by interacting with one another while learning from their experience (local best) P k. All particles tend to move towards a better solution discovered by particle (global best) P G. Eq. 3 is used by PSO to update its particle velocity. The equation consists of three parts, which are inertia weight denoted by wv k, cognitive component denoted by c1r ( P k xk ) and social component denoted by c2r( P G xk ). Where r is a random number from to 1; c 1 and c 2 are the learning factors. Inertia weight is used to control the exploration and exploitation of the search space based on the contribution of previous velocity. Cognitive component is based on their own experience and social behavior component which is based on collaboration with other particle. vk+ 1 = wvk + c1r( Pk xk ) + c2r( PG xk ) (3) z

3 Eq. 4 is used by PSO to update the particle location. The new location is the sum of current location and new velocity. Eq. 5 shows the dynamic inertia weight that linearly decreases after rounds [15]. The use of dynamic inertia weight in PSO demonstrates the exploration ability towards the search space in the beginning of iterative process and exploitation to the final solution when it reaches the end of the iterative process. x (4) k + 1 = xk + vk+1 k bit packet k bit packet E Tx (d ) Transmit Electronic s Tx Amplifier E elec k 2 kd E amp E Rx Receive Electronic s E elec k d w max wmin w iter = wmax iteration (5) iterationmax III. SYSTEM MODEL In this section, previous work on CHs selection will be discussed. There are system models of fuzzy logic based CHs selection [7] and adaptive PSO based CHs selection [13]. The development of proposed CHs selection methods uses the same structure as LEACH protocol with different CHs selection mechanism. A. Radio Model Fig. 2 shows the network radio model used in this paper. The model includes the energy usage for data communication, such as energy for data transmission and reception. The model is assumed to have the transmit power control (TPC) module and the data transmission uses minimum transmit power to achieve acceptable signal-to-noise ratio (SNR). Eq. 6 shows the formula used for data transmission. It represents transmission energy for k bits of data over distance d. E Tx ( k, d) = E ( k) E ( k, d) (6) Tx elec + Tx amp In this paper, two transmit amplifiers are used as described [4]. The use of each transmit amplifier is based on the threshold distance denoted by Eq. 7. Free space propagation and multipath fading propagation is shown in Eq. 8. If the distance between transmitter and receiver is smaller than threshold d, the model ε fs is used. Fig. 2. Network radio model. To receive k bits of data, the received energy of the radio model is expressed as shown in Eq. 9. E ( k) E k (9) Rx = B. Fuzzy Logic Based CHs Selection BS with fuzzy logic will select CHs in every early stage of the round. The selection is based on fuzzy parameters such as sensor node remaining energy and distance. Te details of the parameters are shown below: elec Energy Level remaining energy in the sensor node. Distance the distance between sensor node and the BS measured by radio signal strength (RSS). In simulation study of wireless sensor CHs selection via fuzzy logic control, MATLAB fuzzy toolbox is used as a simulation tool. Fig. 3, 4 and 5 are the membership functions for the simulation study. Fig. 3. Membership functions of distance. d = ε fs ε mp (7) Eq. 8 is the transmitter energy equation together with the data aggregation energy E DA. In clustering protocol, CH will perform data aggregation to compress received data from cluster members to produce a single packet message. The modified model will only be applied into CH for the data transmission to the BS. ke = ke elec elec + kε d + kε fs mp d ke + ke DA DA if d < d if d d (8) Fig. 4. Membership functions of energy level

4 Start Initiate particles velocity and position Fitness evaluation, update P i and P g Fig. 5. Membership functions of node fitness. Update particle velocity and particle location TABLE I. FUZZY RULE BASE No Energy Level Distance Node Fitness 1 low far very small 2 low near small 3 low middle rather small 4 medium far medium small 5 medium near medium 6 medium middle medium large 7 high far rather large 8 high near large 9 high middle very large Fig. 3 shows the membership functions of distance. The linguistic variables used to represent the distance are divided into three levels; near, medium and far. Fig. 4 shows the membership functions of energy level. The linguistic variables used to represent the energy level are divided into three levels; low, medium and high. Finally, Fig. 5 shows the output membership function which is the node fitness. It consists of nine linguistic variables; very small, small, rather small, medium small, medium, medium large, rather large, large and very large. Table I shows the fuzzy rule base used in the paper, it consists of nine rules. C. Adaptive-PSO Based CHs Selection PSO based CH selection is an algorithm that is implemented in the BS. In this paper, it is assumed that the BS is equipped with unlimited power supply with sufficient computing power to run the CHs selection algorithm. After the deployment of sensor nodes, the BS will receive sensor node location and initial battery level for each sensor. In each early stage of the round, BS will calculate suitable CHs formation via PSO. PSO will obtain the parameters used in the calculation. The location of sensor nodes are fixed after deployment and BS can model the remaining energy for each sensor node over time. It is hard to select the best CH formation among large number of sensor nodes. Therefore, the use of swarm intelligence to search for best CHs formation exhibits suitability for practice. Basically, particle in PSO represents a set of possible CHs formation solution. The fitness value for each particle can be evaluated by Eq. 1. The parameter α denoted the contribution of each component. The smaller the fitness value are, the better the particles are. no Fitness evaluation, update P i and P g Fig. 6. Flow chart of PSO with re-selects mechanism. node fitness = αf ( f (1) α ) Eq. 11 shows the sum of non CH node energy over sum of CH energy. Eq. 12 shows the maximum average distance of cluster members to CH from the entire cluster in the network. The parameter N represents the total number of sensor node while the parameter m represents the number of cluster form. N M 1 i= 1 i m= 1 m f Particle=maximum number of particle? Iteration=maximum iteration? End yes yes Announce CHs candidate = E( n ) / E( CH ) (11) f 2 = max d( ni, CH m ) / C m= 1,2,... M i m 2 Reselect mechanism m (12) In PSO, most of the authors set the cognitive and social learning factors to constant numbers. Large c 1 and c 2 will increase the weight of cognitive and social components. In other words, it increases the exploration ability of the search space. The high exploration ability near the end of iterative process may cause the solution harder to converge. The use of Eq. 13 in Eq. 3 may help to solve the above problems. Adaptive learning factor improves the exploitation by no

5 considering diversity between local best and global best value and also the current iterative cycle. P = = i PG c1 c2 2 exp (13) iteration The use of adaptive learning factor can accelerate exploitation speed, but the solution will possibly trap in local maximum. Particle re-select mechanism in adaptive PSO can avoid the solution of being trapped in local maximum. If the value for global best is continuous same for 7 cycles, 25% of particles will be re-initialized. By doing so, it can increase the diversity of the search space to avoid being trapped in local maximum. Figure 6 shows the adaptive PSO flow chart used in the simulation. IV. RESULTS AND DISCUSSIONS To evaluate the performance of three different protocols, the simulation experiments are carried out in MATLAB. The evaluations of the protocols are based on the performance metrics such as network lifetime, energy consume over round and data received by BS over round. LEACH protocol, fuzzy logic based CHs selection protocol and adaptive particle swarm optimization based CH selection protocol used the same network topology for the sake of fair comparisons. The topology contains 1 sensor nodes randomly deploy over 1 m x 1 m network area. The BS is located at x-5 m, y- 2 m. The parameters used in the simulation are shown in Table III. TABLE II. SIMULATION PARAMETERS Parameter Value Network size 1 x 1 m 2 BS location x = 5 m, y = 2 m Simulation round 2 Number of node, n 1 CH probability, p.5 Fig. 7 shows the network lifetime comparisons for LEACH protocol, fuzzy logic protocol and adaptive PSO protocol. First node die round (FND) is used as performance measure for network lifetime. LEACH protocol experiences FND during 53rd round. Fuzzy logic protocol, with the performance of 72rd round shows an improvement of 35.8% over LEACH protocol in FND round. The improvement is contributed from the consideration of critical parameters by the CHs using fuzzy logic protocol. By considering the sensor nodes remaining energy and distance to BS, the remaining energy in all sensor nodes are almost equal. With the score of 89rd round in adaptive PSO protocol, it shows drastic improvement over LEACH protocol by 67.9% and 23.6% over fuzzy logic protocol. The result is due to adaptive PSO protocol selecting the best combination of CH formation through iterative process. High quality data is defined as the data received before FND round. The data aggregation process with sufficient number of cluster members will produce quality data. Fig. 8 shows the data received at the BS. It is very important to prolong the network lifetime while maintaining acceptable quantity of data received by BS. In Fig. 8, the result clearly shows that fuzzy logic and adaptive PSO outperforms LEACH protocol before FND round. Number of Alive Nodes Simulation Time (round) Fig. 7. Network lifetime. LEACH Fuzzy APSO Initial energy, c 1.5 J 1 Packet Size, k 4 bit Transceiver energy, E elec 5 nj/bit Data aggregation energy, E DA 5 nj/bit Amplifier energy Free space, ε 1 pj/bit/m 2 fs Amplifier energy Multipath, ε.13 pj/bit/m 4 mp PSO Number of particle 2 Number of Cluster 5 Simulation round 5 α.5 Number of Alive Nodes LEACH 1 Fuzzy APSO Data Received by Base Station (bits) x 1 6 Fig. 8. Data received at the BS

6 Energy Consume per Round (J) LEACH Fuzzy APSO Simulation Time (round) Fig. 9. Network energy consumption. The simple performance evaluation method for cluster formation can be related to energy consumption per round in the network. With the desired CHs location, the total energy used by the network can be minimized. To achieve energy efficient cluster formation, overall energy consumed should be as low as possible. Adaptive PSO protocol shows the lowest average energy consumption before FND which is.489 J per round. Fuzzy logic protocol achieves.58 J per round which is 3.88% higher than adaptive PSO protocol. Finally, unstable energy uses from LEACH protocol achieved.515 J per round, 5.32% higher than adaptive PSO protocol and 1.38% higher than fuzzy logic protocol. V. CONCLUSION In conclusion, protocols that select CHs based on critical parameters outperform LEACH protocol. Adaptive PSO based CHs selection perform better than fuzzy logic based CHs selection in terms of network lifetime, total data received by BS and less energy consume per round before FND. Optimized formation of CHs in the network will result in longer network lifetime. For future work, fuzzy logic can be used as an adaptive tuning system that modifies PSO inertia weight and learning factors. Hybrid fuzzy logic based PSO can decrease the iteration cycle on finding best solution. ACKNOWLEDGMENT The authors would like to acknowledge the financial assistance of the Universiti Malaysia Sabah Research Grant Schemes (SGPUMS), grant no. SLB14-TK-1/211, and scholarship support by Ministry of Higher Education of Malaysia (MoHE) under MyMaster program. REFERENCES [1] F. Hu and X. Cao, Wireless Sensor Networks: Principles and Pratice. USA: CRC Press, 29. [2] J. Zheng and A. Jamalipour, Wireless Sensor Networks A Networking Perspective. USA: Wiley, 26. [3] I. Mahgoub and M. Ilyas, Sensor Network Protocols. USA: CRC Press, 26. [4] W. Heinzelman, A. Chandrakasan, and H. Balakrishnan, Energy-efficient communication protocol for wireless microsensor networks, Proceedings of the 33 rd Annual Hawaii International Conference on System Sciences (HICSS), Maui, Hawaii, pp , 2, doi: 1.119/HICSS [5] W. Heinzelman, A. Chandraksan, and H. Balakrishnan, An Application-Specific Protocol Architecture for Wireless Sensor Networks, IEEE Trans. on Wireless Communications, vol. 1, no. 4, pp , 22. [6] S.E. Tan, H.T. Yew, M.S. Arifianto, I. Saad and K.T.K. Teo, Queue Management for Network Coding in Ad Hoc Networks, 3rd International Conference on Intelligent Systems Modelling and Simulation, pp , 212, doi: 1.119/ISMS [7] Z.W. Siew, A. Kiring, H.T. Yew, P. Neelakantan and K.T.K. Teo, Energy Efficient Clustering Algorithm in Wireless Sensor Networks using Fuzzy Logic Control, Proc. 211 IEEE Colloquium on Humanities, Science and Engineering Research (CHUSER 211), pp , 211, doi: 1.119/CHUSER [8] I. Gupta, D. Riordan, and S. Sampalli, Cluster-head election using fuzzy logic for wireless sensor networks, Proceedings of Communication Networks and Services Research Conference (CNSR2), Halifax, Nova Scotia, Canada, pp , 25, doi: 1.119/CNSR [9] C.S. Chin, P. Neelakantan, H.P. Yoong, and K.T.K. Teo, Optimisation of fuzzy based maximum power point tracking in PV system for rapidly changing solar irradiance, Global Journal of Technology and Optimisation (GJTO), vol. 2, no. 2, pp , 211. [1] R.V. Kulkarni, A. Forster and G.K. Venayagamoorthy, Computational Intelligence in Wireless Sensor Networks: A Survey, IEEE Communications Survey & Tutorials, vol. 1, no. 2, pp , 211, doi:1.119/surv [11] M. Saleem, G. A. Di Caro and M. Farooq, Swarm intelligence base routing protocol for wireless sensor networks: Survey and future directions, Information sciences, vol. 181, no. 2, pp , 211. [12] J. Cai and J. Sun, A Clustering Routing Algorithm Based on Adaptive PSO in WSNs, Proc. Fourth International Conference on Wireless Communications, Networking and Mobile Computing, pp. 1-4, 28, doi:1.119/wicom [13] Z.W. Siew, C.H. Wong, C.S. Chin, A. Kiring and K.T.K. Teo, Cluster Heads Distribution of Wireless Sensor Networks via Adaptive Particle Swarm Optimization, Proc. 4th International Conference on Computational Intelligence, Communication Systems and Networks Proceeding (CICSyN 212), pp , 212, doi: 1.119/CICSyN [14] J. Kennedy and R. Eberhart, Swarem Intelligence.USA: Morgan Kaufmann Publishers, 21. [15] J. Xin, G. Chen, Y. Hai, "A Particle Swarm Optimization with Multistage Linearly-Decreasing Inertia Weight," International Joint Conference, Computational Science and Optimization (CSO 29), pp , 29, doi: 1.119/CSO