Wolf Nature and Neural Network Based WSN Energy Optimization in Dynamic Environment

1. INTRODUCTION

Wireless sensor network is a popular area for research now days, due to vast potential usage of sensor networks in different areas. A sensor network is a comprised of sensing, processing, communication ability which helps to observe, instrument, react to events and phenomena in a specified environment [1]. This kind of network enables to connect the physical world to environment. By networking tiny sensor nodes, it becomes easy to obtain the data about physical phenomena which was very much difficult with conventional ways. The process of getting that information to a central place to process it, and then returning the result to the nodes requires communication. Radio communication used by WSN nodes is the largest consumer of the finite energy resources contained within a WSN node‘s battery [2]. Routing technique plays a vital role in the wireless sensor network. It is extremely difficult to assign the global ids for a large number of deployed sensor nodes. Thus, traditional protocols may not be applicable for WSN. Unlike conventional wireless communication networks (MANET, cellular network, etc.), WSN has inherent characteristics. It is highly dynamic network and specific to the application, and additionally it has limited energy, storage, and processing capability. These characteristics make it a very challenging task to develop a routing protocol [3, 4]. There are mainly two reasons responsible for the dynamic infrastructure. The first reason is the energy; the sensor nodes have limited energy in the form of batteries. If the protocol is unable to balance the load among the nodes, the sensor node could die. It leads to the dynamic network structure. The second reason is the mobility; in many scenarios after the deployment, sensor nodes are static but sink can move within the network. It makes the network dynamic, and the protocol that works for static sink may not be applicable for mobile sink [5]. The sole purpose of this paper is to find the method which is more energy efficient. As Wireless sensor networks are battery operated. Sensor nodes collect the data and pass them on to the network for further use. This passing and receiving of data utilizes most of the energy of the network. So for better operation and increase the lifetime of the network, energy consumption must be the major factor of concern.

Cheng et al. [6] developed a fast and efficient broadcast (FEB) protocol for asynchronous WSNs with mobile sink. The sink traveled along the sensing field in a predefined path depending on the coverage

broadcast delay and reduced the energy consumption. Ho et al. [7] applied particle swarm optimization to obtain the optimal WSN topology and UAV trajectory for reducing energy consumption. The proposed model was compared with a low-energy adaptive clustering hierarchy (LEACH) protocol to evaluate its performance. Though the framework considered a relatively flat terrain to model the radio communication, the radio model used in the literature can also be useful to design propagation models in other environments. Different from the other proposed models, the UAV is also utilized in this architecture to select the CH from the ground sensor nodes. Gharaei et al. [8] developed energy-efficient mobile-sink sojourn location optimization (EMSLO) scheme for heterogeneous home network. Robovac is employed as a mobile sink, and the sojourn location was optimized to solve the energy hole and network coverage problems. The obtained results from their approach enhanced the coverage time and improved the network lifetime. Zhong et al. [9] introduced a hyper-heuristic framework (HHF) that intelligently organized the mobile sink movements based on heuristic rules. Based on the prior knowledge of their networks, predefined low-level heuristics and training networks were designed and assigned as input to the genetic programming (GP) algorithm to automatically built-up highlevel heuristics. As a result, the GP algorithm produced the heuristics with the highest fitness. Wang et al. [10] introduced a trajectory scheduling method based on coverage rate for multiple mobile sinks (TSCRM) that utilized particle swarm optimization (PSO) to find the optimal rendezvous points for the mobile sinks. TSCRM integrated the genetic algorithm (GA) for scheduling the traveling trajectory of the multiple sinks. The authors argued that the network lifetime enhanced due to the reduction in node‘s energy consumption. Mohammed Al Mazaideh et. al. in [11] proposed acognitive sensing algorithm for efficient data transfer through WSNs, which uses multiple objective genetic algorithms (MOGA) to optimize the number of measurements, transmission range, and the sensing matrix. The algorithm aims at striking the right balance between energy-efficiency and accuracy. It constructs a path in a multi hop manner based on the optimized values. Numerical simulations and experiments show that Paretofront, which is the output of MOGA, helps the user to select the right combination of the number of measurements and the transmission range fitting the application at hand, and to strike a good balance between energy efficiency and accuracy. [13], [14], while some studies were performed to localize the sensor nodes [15], [16].

Explanation of the proposed WNNEO (Wolf Nature and Neural based Energy Optimization) model is done in this section. Table 1 shows a notation list to explains various steps of the energy optimization proposed model. Fig 1 is a flow chart of neural network training for the selection of fit nodes on the basis of (energy, position). Fig 2 shows the flow chart of the wolf algorithm for nodes cluster center selection after a fix number of rounds.

Wireless sensor nodes are placed in a fixed area A having dimensions in meters. As work is in dynamic nature hence each node may have a different size of the battery or energy storage, in other words, node energy E is not the same at the beginning of work. Randomly place N number of nodes within the A area of WSN. The position of the node plays an important role in energy dissipation for transferring data to the base station in the network. A fix packet size of L number of bits is considered in the work. So receiving cost (energy) Er of a packet is same for all nodes. But transferring cost of Et directly depend of d distance between source and destination node shown in Eq. 1 [17]. In eq. 1 α is energy required per bit for processing in node. p, s value depends on distance if d is greater than do s is 2 otherwise s is 4. classes first is strong nodes and other is weak nodes. This classification of nodes reduces the cluster head selection time of the model and increase the quality of wolf population. WSN nodes have different features out of those this work utilize current energy value and current position in A area. Apart from this number of packets transfer feature were also evaluate. Estimation of packet count was done by Eq. 2. Where rth round in the network, P is packet count for rth round of nth node. Traiing input feature vector looks like {, d, } where d is distance between nth node position coordinates and base station position coordinates. Desired output is value either 1 or 0, where 1 shows that node is strong and 0 shows that node is weak.

This work considers three layer neural network architecture. Input layer have three neurons as per training vector size (Energy, distance, packet_count). Hidden layer has five neurons. Finally output layer has only one neuron. It‘s an completely connected architecture. This work uses soft-max activation function [17] for triggering the neuron. Output of training gives trained neural (TN) which takes node {, d, } value and classify it either strong or weak node class.

Due to dynamic nature of WSN genetic algorithm is adopt as no prior training required to find good solution. Based on the fitness value wolf genetic algorithm [18] classify chromosomes into four categories, hence chance of getting a good solution is high. This algorithm works on wolf hunting nature where alpha category wolf are need to be search from the group. In this model each chromosome is acting as wolf. Based on the fitness function wolf elements were change to get good cluster center. Generate Wolf population: Wolf is set of k nodes in the network and collection of m number of wolf W is wolf population [19]. Wolf population is a matrix of mxk. Trained neural network TN has reduced the size of N by rejecting weak nodes. k is optimal cluster as per N number of nodes and A Value of k obtained from eq. 3 [20]. Where is ideal amplifier power. W is obtained from Gaussian random distribution function.

Hunting Value Each wolf has its own hunting efficiency, so this function finds the fitness of wolf hunting. Each wolf has set of cluster center nodes. Based on the minimum distance between non cluster center node NCN and cluster center nodes, NCN get clustered. Similarly all set of nodes were clustered into any of wolf set nodes. Now estimate energy dissipate to send one packet by NCN to the cluster center ENCND. After that estimate energy dissipation of sending a packet to the base station from the cluster center node ECCD. Summation of total energy dissipation is hunting value in the work.

Hunting value divide wolf population into four group alpha beta, Gamma, Delta such that Halpha > Hbeta > Hgamma > Hdelta Now wolf change position for prey 0 to 3 by eq.

Final position shifting value estimate by Eq. 9

As per shifting position alpha wolf modifies the other wolf in the population. One of cluster center from the Alpha wolf set was select randomly and place at same position in the other wolf set. This operation done with all other wolf so new wolf increase the size of population. Population need to be update by estimating hunting value of new wolf and compare with parent wolf. If any of parent hunting value if lower than new wolf then remove weak wolf from population. This addition of new or removing of old wolf is population updation. Repeat this steps for T number of iterations to get good set of solution (Cluster center).

Cluster center cluster nodes in the WSN. Once nodes get cluster then communication of sending sensor data start. This sending of data continue for a few number of rounds and update cluster center with new node position, energy values.

Input: N, A, TN, Pos Output: CN // Cluster nodes 1. 2. W = Gaussian(m, k, N, TN) 3. Loop 1:T 4. 5. 6. [Alpha, Beta, Gamma, 7. Delta) 8. 9. 10. End Loop

11.

Implementation of proposed WNNEO was done on MATLAB platform. Experimental values was compared with existing model of WSN energy optimization MSGR [20] and IWDEO [21]. Hardware setup for experimental work have configuration of 4GB RAM, Intel I3 processor. In order to analyze models work various combination of Area (100m,

Table 2 shows round instant when first node discharge in the WSN, it was obtained that proposed WNNEO model has higher number of rounds as compared to other existing models. Use of Wolf genetic algorithm has increases the clustering accuracy of work and this reduces the power requirement for communication. Fisrt node discharge in WSN model was improved by % as compared to IWDEO model proposed in [20].

Total packet transfer parameter values in table 3 shows that proposed WNNEO model has increases the packet count by % as compared to IWDEO and % as compared to MSGR. One more observation obtained from the resuts that increase in area size reduces the packet count. Further use of neurl network for cluster center selection increases the model packet delivery count. When all set of live nodes transfer one packet to the base station is termed as round in WSN. So rounds directly shows that how long a network service when last node send its packet. Table 4 shows that proposed model has increases the percentage of total rounds by % as compared to IWDEO. Hrbrid use of neural network and wolf genetic algorithm increases the rounds count in dynamic WSN environment.

(a)

(b) (c) (d) (e) (f)

(a) (b) (c) (d) (e) (f)

Fig. 3 and 4 shows that packet count of proposed model is always high in different size of area and nodes. Similarly node loss of proposed model is less in low number of WSN round. Use of neural network for cluster center candidate node selection has directly increases the wolf clustering algorithm for any random situation.

1. M. Song and M. Zheng (2018). ―Energy efficiency optimization for wireless powered sensor networks with nonorthogonal multiple access,‖ IEEE Sensors Letters, vol. 2, no. 1, pp. 1–4. wireless sensor networks,‖ Computers, Materials & Continua, vol. 56, no. 3, pp. 433–446. 3. J. Wang, Y. Gao, C. Zhou, R. S. Sherratt, and L. Wang (2020). ―Optimal coverage multi-path scheduling scheme with multiple mobile sinks for WSNs,‖ Computers, Materials & Continua, vol. 62, no. 2, pp. 695–711. 4. Yetgin, H., Cheung, K. T. K., El-Hajjar, M., & Hanzo, L. H. (2017). A survey of network lifetime maximization techniques in wireless sensor networks. IEEE Communications Surveys & Tutorials, 19(2), pp. 828–854. 5. Ari, A. A. A., Yenke, B. O., Labraoui, N., Damakoa, I., & Gueroui, A. (2016). A power efficient cluster-based routing algorithm for wireless sensor networks: Honeybees swarm intelligence based approach. Journal of Network and Computer Applications, 69, pp. 77–97. 6. H. Cheng, L. Tao, and X. Zhang (2019). ―A Fast and Efficient Broadcast Protocol With a Mobile Sink Node in Asynchronous Wireless Sensor Networks,‖ IEEE Access, vol. 7, pp. 92 813–92 824. 7. D.-T. Ho, E. I. Grøtli, P. B. Sujit, T. A. Johansen, and J. B. Sousa (2015). ‗‗Optimization of wireless sensor network and UAV data acquisition,‘‘ J. Intell. Robotic Syst., vol. 78, no. 1, pp. 159–179. 8. N. Gharaei, S. J. Malebary, K. Abu Bakar, S. Z. Mohd Hashim, S. Ashfaq Butt, and G. Sahar (2019). ―Energy-Efficient Mobile-Sink Sojourn Location Optimization Scheme for Consumer Home Networks,‖ IEEE Access, vol. 7, pp. 112 079–112 086. 9. J. Zhong, Z. Huang, L. Feng, W. Du, and Y. Li (2020). ―A Hyper-Heuristic Framework for Lifetime Maximization in Wireless Sensor Networks with A Mobile Sink,‖ IEEE/CAA Journal of Automatica Sinica, vol. 7, no. 1, pp. 223–236. 10. J. Wang, Y. Gao, C. Zhou, R. S. Sherratt, and L. Wang (2020). ―Optimal Coverage Multi-Path Scheduling Scheme with Multiple Mobile Sinks for WSNs,‖ Computers, Materials & Continua, vol. 62, no. 2, pp. 695–711. 11. M. A. Mazaideh and J. Levendovszky (2020). "A multi-hop routing algorithm for WSNs based on compressive sensing and

12. S. K. Haider, M. A. Jamshed, A. Jiang, H. Pervaiz, and Q. Ni (2019). ‗‗UAV assisted cluster-head selection mechanism for wireless sensor network applications,‘‘ in Proc. UK/ China Emerg. Technol. (UCET), pp. 1–2. 13. F. Stoican, D. Popescu, and L. Ichim (2019). ‗‗Trajectory design for effective and secure communication in UAV-WSN systems,‘‘ in Proc. IEEE Radio Antenna Days Indian Ocean (RADIO), pp. 1–2. 14. J. Yang, X. Wang, Z. Li, P. Yang, X. Luo, K. Zhang, S. Zhang, and L. Chen (2016). ‗‗Path planning of unmanned aerial vehicles for farmland information monitoring based on WSN,‘‘ in Proc. 12th World Congr. Intell. Control Autom. (WCICA), pp. 2834–2838. 15. J. Grigulo and L. B. Becker (2018). ‗‗Experimenting sensor nodes localization in WSN with UAV acting as mobile agent,‘‘ in Proc. IEEE 23rd Int. Conf. Emerg. Technol. Factory Autom. (ETFA), pp. 808–815. 16. D. S. Jasrotia and M. J. Nene (2019). ‗‗Localisation using UAV in RFID and sensor network environment: Needs and challenges,‘‘ in Proc. Int. Conf. Comput., Commun., Intell. Syst. (ICCCIS), pp. 274–279. 17. Nomaan Jaweed Mohammed (2020). Neural Network Training by Selected Fish Schooling Genetic Algorithm Feature for Intrusion Detection. International Journal of Computer Applications 175(30): pp. 7-11. 18. S. Mirjalili, S. M. Mirjalili, A. Lewis (2014). "Grey wolf optimizer", Advance in Engineering Software, vol. 69, pp. 46–61. 19. Ashwani Mathur (2020). "Hybrid Combination of Error Back Propagation and Genetic Algorithm for Text Document Clustering" International Journal of Computer Trends and Technology 68.11: pp. 64-68. 20. Suraj Sharma, Deepak Puthal, Sabah Tazeen, Mukesh Prasad, And Albert Y. Zomaya . ―MSGR: A Mode-Switched Grid-Based Sustainable Routing Protocol for Wireless Sensor Networks‖. IEEE access October 12, 2017. 21. Vikas Sakalle PhD Scholar, Dr. Sunil Phulare, Dr. Anoop Chaturvadi. "Wsn Energy Optimization Using Intelligent Water Drop

M.Tech (CSE), PhD Research Scholar, LNCT University, Bhopal, MP vikassakalle@gmail.com