Home         Authors   Papers   Year of conference   Themes   Organizations        To MES conference

Support methods of reliability on the wireless sensor networks by criterion of the network load
Authors
	Yuldashev M.N.
	Adamov A.P.
	Adamova A.A.
Date of publication
	2016
Abstract
	Growth of computing opportunities of the modern program-driven devices, and also existence of a set of different sensors made possible creation of the sensor networks of information processing are providing automation of various processes. Similar systems allow to collect various data from environment and then as appropriate to react to them. Great attention and hardware such leading companies as Philips, Ember, Samsung, IBM, Motorola, Freescale Semiconductor, LG, , etc. pay to the analysis and the solution of problems of development of the reliable BS. The active researches in the field of sensor networks speak about relevance of this subject. Scientists of the whole world try to improve this system. In article one of tasks taking place on the sensor networks is provided: classification of a status of system in general. The operator needs to know in what status there is a system at the moment to accept the appropriate actions. At the same time sensor networks represent unstable systems where the node which is responsible for data transfer to the principal module at any time can fail. Use in these cases of the branching algorithms doesn't yield desirable results. The main objective when using machine training on the sensor networks is classification of a status of system in general. The system of a sensor network can be described by means of a limited number of statuses. In the elementary case it is enough to designate two statuses: "operational" and "faulty". Each set of values is compared with a specific status of system, not crossed subsets thereby are formed. This problem of classification is well solved with the help of a decision tree. Leaves of a tree will be the given statuses of system, and a branch – search criterions. For determination of parameters of predicates it is necessary to collect learning selection. Quality of the constructed decisive tree directly depends on quality of this selection. As technology of creation of a decisive tree it is possible to use algorithm of ID3. This algorithm makes locally optimal solutions at each stage therefore the number of computation in ID3 is minimum that positively affects energy consumption of nodes of a sensor network. Advantages of use of the decisive trees constructed on algorithm of ID3 is that: labor input of algorithm of decision-making is linear on selection length, there are no failures from classification, decision-making on the basis of incomplete information from sensors is possible (objects with passes. Use of methods of machine training for the task of classification of a status of system is offered. In addition to classification, the solution of such tasks as anticipatory decision-making, noise immunity increase, adaptation to changes of environment will become possible. The offered algorithm of creation of a decision tree of ID3 has the advantages and shortcomings. Advantages of algorithm are considerable for an objective, and defects are eliminated by analytical methods, such as truncation of the constructed tree.
Keywords
	wireless sensor networks, machine learning, decision making, optimization of networks, data analysis, network load, classification.
Library reference
	Yuldashev M.N., Adamov A.P., Adamova A.A. Support methods of reliability on the wireless sensor networks by criterion of the network load // Problems of Perspective Micro- and Nanoelectronic Systems Development - 2016. Proceedings / edited by A. Stempkovsky, Moscow, IPPM RAS, 2016. Part 2. P. 248-251.
URL of paper
	http://www.mes-conference.ru/data/year2016/pdf/D104.pdf