Many of the activities that aretaking place in today’s world are dependent on services offered by computernetworks, and MANET networks are of the networks that have many applications.
Load balancing and network congestion are major problems in MANET networkrouting, and therefore the issue that becomes important is creating loadbalancing in this kind of network so that all data is transmitted intact. Forthis purpose, it is necessary to perform routing in the network so that the loadis balanced in all directions.It is therefore necessary that allthe deficiencies that may unbalance the network be identified. Characteristicssuch as high mobility of the nodes and thus dynamic topology of the network,low bandwidth, and even limited power and energy lead to the complexity ofrouting algorithms in ad hoc networks. All of these characteristics make mostof the ideas contained in routing algorithms on other networks not be applicablefor ad hoc networks. Algorithms that perform routing as a single route and areof algorithms based on demands have not generally solved the problem of load.
Among ad hoc routing-networksalgorithms, some algorithms perform routing as a multi-route operation usingmulti-route routing where fault tolerance effectively increases. Most multi-routerouting algorithms after route discovery process where they found multiple routesfrom source to destination choose one of these routes as the main route andbegin sending data via the same route. They keep the other routes asalternative routes and in case of failure of the main route, they use one ofthe alternative routes for sending data, and until a route is established and itsnodes have energy, the same route is used eventually, leading to end nodesenergy.Ending of the node energy on a routeand failure of nodes create gaps in the network and the network is fragmentedthat has a direct impact on network performance and routing. This is seen bothin one-path and multi-route algorithms.Given the need for load balancingin MANETs and its importance, in this paper, the aim is to improve loadbalancing in this type of networks by considering upgrading parameters thatdirectly affect the type of balance. To this end, here a framework based onlearning automata is proposed for load balancing by considering nodes energy,signal strength, number of steps, and the amount of nodes for each route forMANETs.2.
RESEARCHBACKGROUNDMany studies have been proposed inthe field of load balancing in MANET, the most important of which are asfollows:In 1, a method has been proposedto balance load that distributes traffic on three key metrics: 1) Route Energy,2) queue traffic, and 3) the number of steps with the corresponding weight. In2, priority-scheduling technique based on fuzzy is used to determine effectsof this method on reactive routing protocols AODV and DSR. In this method, apriority index is added to each packet in the queue nodes, and this priorityindex is based on queue length, data rate, and packet expiry time.In 3, stable routing algorithmsbased on fuzzy are provided for MANET whose aim is to strengthen QOS.
One ofthe major problems in QOS routing in MANET is ensuring that one route isestablished in the network until the end of data transmission, to reduce thenumber of broken routes, a reliable new routing algorithm that uses fuzzy logicis presented. When routing, this algorithm chooses a sustainable route based onthe position of nodes and data rate. Moreover, a new protocol has beenpresented for maintenance/repair routes so that when a route breaks, a newroute is established. In 4, Energy Efficiency in MANET routing protocols loadbalancing is studied.
The work presented in this paper is: 1) energy efficiencyin routing protocols of storage of node based on AODV with volume of work loadbalancing (AODV-NC-WLB), 2) a new application of parameters of energyefficiency in routing protocols MANET, 3) implementation and simulation of thestudy in NS2 in energy efficiency AODV-NC-WLB, significant maintaining ofimprovement in throughput, overflow, delivery rates, and delay more than AODVstandard for high-volume scenarios. In 5, multiple adaptive routing has beenoffered for load balancing. In this method, an algorithm is used that detectsmultiple routes to the destination called Fail-Safe routing. There is one main route,in all of which there are intermediate nodes and multiple routes are composed ofnodes with minimal remaining load, time, power, and energy, and have thehighest bandwidth.
In routing when the average load of a node has an increasebeyond the threshold, to reduce the traffic load on that congested route,traffic is distributed over multiple other routes. In 6, a new method ofrouting MANETs by considering energy by learning automata is suggested. In thismethod, for MANETs based on best route selection using learning automatatechnique is proposed.
The proposed protocol is able to apply effectively ontrack with regard to limited energy. This method is applied in a version ofAODV routing algorithm (i.e., routing AODV with learning automata) (AAODV).Arepresentative of learning automata runs on each node runs, and this representativedynamically trains the best route. The protocol consists of three modes, andeach node trains the best route to the destination. In 7, the efficiency ofrouting protocols of traffic-oriented load balancing in MANET has been compared.In 8, Load Balancing ParallelRouting Protocol (LBPRP) has been provided that solves the problems of previousmultiple routing protocols and distributes data in parallel in all directionsat the same time.
In this paper, a simple test scenario has been presented toensure that the model is effective and credible. LBPRP creates load balancing,reduces delay, increases packet delivery ratio, and throughput. In 9, theanalysis of the performance of load balancing in MANET using multiple routingprotocols on demand is offered. In 10, the protocols of load balancing inMANET are tackled. In addition to the above-mentioned methods that are more important,other methods are provided in the literature for load balancing in MANET.3.
THE PROPOSED METHODThe processduring which organisms learn various subjects has long been of interest toexperts. Studies done in this area are generally concentrated in two branches:1)Understanding the process where living things learn.2) Obtaining methods byusing which this learning capability can be created in machines.A research group called Tsetlin inthe Soviet Union introduced the concept of random automaton for the first timein the early 1960s. After that in later research, several examples of theapplication of learning methods in engineering systems were developed of which routing in phone, patternrecognition, Object Partitioning, and Adaptive Control could be named.Each Stochastic Learning Automata (SLA)is made up of two main components: 1) a stochastic automata with a limitednumber of operative and is in interaction with a stochastic environment. 2) A learningalgorithm by using which the automaton learns doing the optimal action.Each automaton can be seen as aFinite State Machine (FSM) that can be observed by the following pentamerous: Eq.
(1.3)The parameters in this pentamerousare:• Set of automaton operations • Set of the automaton input • Function that maps input and the currentstates to the next state • Output function that maps thecurrent state to the next output • The set of the internal states ofthe automaton at the n-th moment Set ?includes applying the automaton where the automaton in each iteration, chooses oneof them. Inputs ? define automaton inputthe details of which will be discussed in the next section.
F and G mappings convertthe current state and input to the next output (action) selected by theautomaton. If mappings F and G are decisive, the automaton is calledDeterministic Automata.In such a case, by having primary andinput modes, output and the next state are uniquely obtained. If mappings F andG are random, automata is called stochastic automata. In that case, only theprobabilities related to the next state and corresponding outputs could bedetermined.
Stochastic Automata is divided into two categories: 1) Automatonwith a fixed structure 2) Automaton with Variable Structure.Learning automata with variablestructure a used here is LR-P where the probability selecting p foreach action of ? is updated according to the following equations. In case ofreceiving a favorable response from the environment ?(k)=0, possibility of thataction is rewarded by the following equation: (1) In case of receiving an unfavorableresponse from the environment ?(k)=1, possibility of that action is fined bythe following equation: (2) In both of the above equations: pj(k + 1) is probability automaton at the time k + 1, pj (k) isthe probability automaton at time k, a is reward, b is fine and r is number ofmeasures.After creating the existing nodesin the network stochastically, two nodes are selected randomly, one of which isa resource and the other plays the role of the destination. If two nodes aredirectly connected to each other, routing does not make sense and packettransmission is performed. Otherwise, the routes between two nodes aredetermined. An example of this is shown in Figure 1.