Get documents about "SOFT COMPUTING IN WIRELESS SENSORS NETWORKS
Document Sample
SOFT COMPUTING IN WIRELESS SENSORS NETWORKS
Averkin A.N. Belenki A.G.
Dorodnicyn Computing Center of the Russian Lomonosov Moscow State University
Academy of Sciences abelenkiy@gmail.com
averkin@ccas.ru
embedded realization in WSN due to limited com-
Abstract munication and power resources of WSN.
Only few companies in the world solve this problem
by embedded soft computing approaches. These ap-
The embedded soft computing approach in proaches for WSN mean a combination of embedded
wireless sensor networks is suggested. This fuzzy logic and neural networks models for informa-
approach means a combination of embedded tion processing in complex environment with uncer-
fuzzy logic and neural networks models for in- tain, imprecise, fuzzy measuring data. It is generali-
formation processing in complex environment zation of soft computing concept for the embedded,
with uncertain, imprecise, fuzzy measuring distributed, adaptive systems. These approaches were
data. It is generalization of soft computing suggested by Russian scientists in 1997 [2], when
concept for the embedded, distributed, adap- hybridization of soft computing and mathematical
tive systems. statistics was used to process the results of heteroge-
neous measurements for environment monitoring
Keywords: embedded fuzzy logic, data fu- applications.
sion, clusterization, aggregation, fuzzy distrib- The first realization of these approaches was made in
uted knowledge base. 2004 [1,2,4]. The main part of our embedded soft
computing and soft computing approaches is Smart
Node (SN) model for WSN [3]. The core of SN is
1 Introduction Fuzzy Engine, which consists of three modules:
knowledge base (a set of fuzzy production rules),
The technology of fuzzy and neuro-fuzzy systems fuzzification and defuzzification modules, which
in WSN uses soft computing approach to increase transform numerical measurements in linguistic form
performance of wireless sensors networks (WSN) and vice-versa. The output of SN can approximate of
and to make them more intelligent. WSN is one of any function of input parameters, e.g. when it is im-
the most promising technologies of the 21st cen- possible or difficult to measure certain parameter, it
tury. For the first time «smart» sensors were im- can be computed by SN with the use of special rules
plemented by Berkeley University of California from knowledge base. Similarly the special rules can
together with INTEL corporation. be created for data fusion, clusterization, aggregation,
The prototype of WSN node is a software- routing and power consumption. The knowledge base
hardware platform for deployment of several spe- of SN can be created as a result of knowledge acqui-
cialized sensors on the base on an autonomous sition from exert or by supervised neural network
wireless controller. WSN consists of a large num- learning. Application knowledge in nodes can sig-
ber of tiny devices, which are deployed in real en- nificantly improve the resource and energy effi-
vironment and function as a united network. To ciency, for example by application-specific data
provide sensor nodes with a possibility of self- caching and aggregation in intermediate node
organization the specialized software was designed SN are realized inside MeshNeticsTM platform [4].
together with IDE for application development. MeshNetics™ is a family of software components,
The specialized software implements the possibili- algorithms, hardware designs and solutions that en-
ties of communication, routing and application able next generation M2M applications [3]. By add-
support for WSN. ing expert system to monitoring, controlling, tracking
The increasing of WSN performance means pre- applications, it enables new generation of solutions
liminary processing of raw data, data fusion, clus- optimized for the needs of end-users. With MeshNet-
terization and aggregation. Intelligent WSN pro- ics™ businesses profit from reduced pricing struc-
vides also distributive decision-making and queries ture, life-cycle time and enhanced competitiveness of
processing, knowledge-based routing and power their new and existing products.
consumption. Methods of decision-making and MeshNetics™ enables expert remote monitoring and
information processing based on symbol models of control of a wide range of processes, assets, systems,
classic artificial intelligence are too complex for and facilities. Built-in expert system with advanced
algorithms, neural networks and fuzzy logic creates a
new generation of WSN’s and frees businesses from
being trapped into costly and complex wired • WSN must continue to operate at all times
“static” systems and enhances the possibilities by even when some of it nodes get physically
listening actively to your environment. destroyed at unpredictable times.
Built-in expert system on the base of SN, that • WSN must continue to operate without in-
allows to support hybrid distributive expert sys- terruption when new nodes are added to the
tem on the nodes of WSN, which can realize, network in order to replace the failed ones or
together with data collection and communication, extend the network.
a large class of existing algorithms of data fusion • As a result, node communication may re-
and aggregation. quire different paths at different times de-
MeshneticsTM SN is the node of MeshneticsTM pending on the state of end-to-end link be-
platform, which include Smart Engine. Smart En- tween communicating parts of the network.
gine is given by a number of its parameters, i.e. • The decrease of battery power is different in
rules patterns, variables, terms, membership func- different nodes.
tions, triangular norms. These parameters can • Data transmission time and power losses in-
transmit across WSN. These transmissions can be crease with the size of WSN.
defined by user, or by Smart Node (SN) itself. Two last problems are of great importance to the
E.g., in goal tracking process these transmissions user. The carrying capacity of network only slightly
can be realized by SN in dependence of mutual depends on number of nodes (it increases as log N,
smart node and goal positions. where N – number of nodes). The most rational out-
Software environment of SN is universal tool for put is decreasing of traffic by adding distributive hi-
intellectual decision support in WSN and it is erarchical data processing inside network and by pro-
strongly connected with following power- viding the user with relevant answers only. There are
aware&networked embedded Computer Systems a number of various algorithms for this processing
research areas: application-driven network archi- realization. These algorithms depends on data types
tectures, emerging platforms and technology, and data generalizations levels. But in traditional
resource constrained real-time OS’s, distributed models of distributive hierarchical data processing
algorithms (broadcast, anycast, multicast, con- each algorithm is strictly connected with certain node
vergecast) in lossy wireless networks, ad hoc for given network topology. Changing of network
multi-hop routing, , in-network aggregation and topology implies reboot of nodes and this process
processing, coverage and density, ranging and needs transmitting of large pieces of code. To solve
localization, resilient aggregators, distributed the problem in SN transmitting of code units has
feature extraction, tracking, and collaborative changed by transmitting of knowledge units. These
signal processing. knowledge units are used for reboot local note with
tuning parameters. The last problem (irregularity of
power consumption) is solved by embedding power
2 Smart Nodes Challenges consumption rules in given sensor node. On the base
of these rules the sensor node can make autonomous
2.1 Data Processing Challenges decision about utility of participation in data collec-
tion and data transmission for given states of the en-
The main goal of WSN activity is collecting a tre-
vironments, neighbor nodes and decision-making
mendous amount of data and transmitting them to
the user. Collected data can be interpreted as dis- node.
As software tool for this purpose we have realized
tributed knowledge base. In this approach both
universal tool of fuzzy sensor shell with production
system user and system designer have the prob-
lems of control for distributed data processing knowledge model, embedded in all nodes of SN. This
processes of uncertain, incomplete or redundant shell may approximate a large class of existing algo-
rithms of data fusion and aggregation. In this ap-
sensor measurements.
proach each sensor becomes intelligent agent with
The main factors, that influence on WSM effec-
knowledge about himself and its’ environment and it
tiveness and make problems for designers are the
is able to autonomous decision-making. Sensor may
following:
control this knowledge and send it to other node. In
• WSN nodes have very restricted computa-
this case WSN can be interpreted as distributed data
tional and storage power.
base and knowledge base with the possibilities of
• Node communication range is limited. In
mobility and adaptability. The fact allows using
most cases nodes can directly communi-
multi-agent technologies and distributive intelligent
cate with immediate neighbors only.
decision support systems.
• WSN consists of a large number of unre-
liable nodes, that produce measurement or
transmission errors
2.2 Middleware Challenges 3.2 Using SM for Fuzzy Data Base Design-
SN technology sits between the operating system ing
and the application and thus belongs to middle- From one perspective sensor networks are similar
ware. Thus the main purpose of middleware for to distributed database systems. They store environ-
sensor networks is to support the development, mental data on distributed nodes and respond to ape-
maintenance, deployment, and execution of sens- riodic and long-lived periodic queries. Data interest
ing-based applications. This includes mechanisms can be pre-registered to the sensor network so that the
for formulating complex high level sensing tasks, corresponding data is collected and transmitted only
communicating this task to the WSN, coordination when needed. These specified interests are similar to
of sensor nodes to split the task and distribute it to views in traditional databases because they filter the
the individual sensor nodes, data fusion for merg- data according to the application’s data semantics and
ing the sensor readings of the individual sensor shield the overwhelming volume of raw data from
nodes into a high-level result, and reporting the applications. Fuzzy query approach can be used to
result back to the task issuer. The most part of reduce this volume of raw data.
these mechanisms were successfully realized in The extension of TinyDB by fuzzy attributes can be
SN. interpreted as fuzzy TinyDB and fuzzy active Ti-
Unique property of SN embedded middleware for nyDB. This possibility can be easy realized but does
WSN is imposed by the design principle applica- not seems very useful because there are only few
tion knowledge in nodes. Traditional middleware possible classes of requests to WSN and representa-
is designed to accommodate a wide variety of ap- tion of fuzzy modifier in TinyDB are rather re-
plications without necessarily needing application stricted. Besides volume of DB is too large for SN
knowledge. SN embedded middleware for WSN memory. So using of extended SQL for fuzzy re-
has to provide mechanisms for injecting applica- quests processing is not effective.
tion knowledge into the infrastructure and the Thus in Fuzzy MeshneticsTM expert WSN on the base
WSN. of SN we can completely substitute functions of
For this purpose SN embedded middleware for fuzzy data base for WSN (possible TinyDB fuzzy
WSN has to provide special knowledge representa- extensions) by SSM data bases functions with special
tion language, special query language, special pro- query processing language for SN. The language can
tocols of query forwarding, special methods of data be used for:
fusion and aggregation and special methods of • Fuzzy queries to WSN (fuzziness can be in
software update management. These new tech- query only and also in WSN);
nologies have been realized on the base of fuzzy • For fuzzy triggers designing;
systems technology. • For active data base designing;
• For communication between SN.
Traditional Event-Condition-Action triggers (active
3 Possible Application Fields of SN database rules) include a Boolean predicate as a trig-
ger condition. As far as WSN can be considered as
3.1 Smart Node in WSN Control. distributive database, we can see that SN with Event-
When we use knowledge (meta-rules) to control Condition-Actions in KB realize embedded fuzzy
WSS we have analogy with active network para- triggers for this distributive database.
digm. The similarity is in sending together using
together with each request special block of rules 3.3 Using of SN for Data Aggregation
(capsule for active networks) to process the request and Fusion.
by SN (server for active networks). The difference
is that SN suppose two types of traffic – knowl- If all raw data is sent to base stations for further proc-
edge traffic and data traffic and in active networks essing, the volume and burst ness of the traffic may
there no knowledge traffic. For knowledge traffic cause many collisions and contribute to significant
and for communication with other subsystems (e.g. power loss. To minimize unnecessary data transmis-
neuro-fuzzy systems) we can use FULL-like spe- sion, intermediate nodes or nearby nodes work to-
cial language for fuzzy knowledge representation. gether to filter and aggregate data before the data
Using SN we can realize: arrives at the destination.
• Routing algorithms Five general, goal-oriented, data fusion methods are
• Optimal control of power consumption in in use today in WSN (ordered by data complexity) -
WSN data association, identity fusion, effect estimation,
• Data traffic control pattern recognition and artificial intelligence. Ten
• Q&S control discrete data fusion techniques can be identified
within these five general categories: figure of merit
and gating technique in the data association, Kalman
filters in the identity fusion, Bayesian decision are to provide distributed accumulation, transmitting
theory and Dempster-Shafer evidentional reason- and using of these knowledge. One of approaches is
ing in the effect estimation, adaptive neural net- to use expert system with knowledge base distributed
works and cluster methods in the pattern recogni- among SNs in WSN. The real data attributes (IPA)
tion, expert systems, blackboard architecture and are processed by SN knowledge base and by knowl-
fuzzy logic in the artificial intelligence. edge bases of neighbor SNs.
The SN data fusion technology focuses on the ac- But the main problem is the cost and the complexity of
quisition of high-level information (artificial intel- data delivery in data fusion SN, because this data fusion
ligence level), i.e. information that is related to SN, because this position of this SN must be fixed and
many conventional physical quantities in a non- close to the user. So the assignment of WSN as point for
analytical way. In these complex cases, fuzzy pro- data fusion must be dynamic procedure and the SN
duction systems and fuzzy neural networks are position should be optimized in regarding of the query,
more effective and they compute and report lin- WSN and environment status. Together with SN’s
guistic assessments of numerically acquired val- assignment its knowledge base should be changed. When
ues. Two methods are proposed to realize the ag- cluster head function is delivered inside cluster of nodes
gregation from basic measurements. The first one
from one SN1 to SN2 , than knowledge base with cluster
performs a combination of the relevant features by
head functions of SN1 should be send to SN2.
means of a rule-based description of the relations
between them. With the second, the aggregation is
Thus fuzzy distributed knowledge base for distributed
realized through an interpolation mechanism that
data base query processing has the following proper-
creates a fuzzy partition of the numeric multi-
ties:
dimensional space of the basic features. This parti-
tion can be realized with fuzzy neural networks. • It functions as distributed expert system.
But SN can also can be used on low levels of data • Knowledge in production form can be
fusion, e.g. for filtering and for pattern recognition. transmitted between nodes.
Data fusion algorithms in production form can be • Knowledge base for inquiry answer can be
easily decomposed and they have hierarchical form send in SN together with the inquiry
by nature. So sensor nodes hierarchy can realize • Special language is used for knowledge base
data fusion inside WSN. The knowledge bases for transmission between nodes.
processing of data in given node is distributed in- • Knowledge-based program of data fusion
side WSN. uses parallel computing algorithms and des-
But a naive placement of the fusion functions on tines for the whole WSM and not only for
the network nodes will diminish the usefulness of certain SN.
in-network fusion, and reduce the longevity of the
network (and hence the application). Thus, manag-
ing the placement (and dynamic relocation) of the References:
fusion functions on the network nodes with a view
to saving power becomes an additional responsibil- [1] Suvorov V., Zubkov G., Averkin A. In for-
ity of the application programmer. Dynamic relo- mation Technologies on the Base of Wire-
cation may be required either because the remain- less Sensors Networks // Proceedings of the
ing power level at the current node is going below International conference on Information
threshold, or to save the power consumed in the problems in the 3rd millennium (Kazan,
network as a whole by reducing the total data 2004).
transmission. Supporting the relocation of fusion [2] Averkin A., Belenki, Suvorov V., Zubkov
functions at run-time has all the traditional chal- G. Soft Computing in Wireless Sensors
lenges of process migration. Networks, International Conference on Soft
Computing and Measurements SCM’2005,
St.Petersburg, 2005.
3.4 Using of SN for Fuzzy Distributed [3] Averkin A.N., Prokopchina S.V. Survey of
Expert System Designing Soft Measurement Conception, St.Peterburg:
Gydrometeoisdat, 1997.
WSNs are huge dynamic databases but for more [4] http://meshnetics.com
effective using of information we need more effec-
tive organization. The most interesting approach is
to use WSN as distributed computing environment
for intelligent data processing methods and as
storehouse of this methods and not only tools for
data measuring and transmitting.. Thus methods
Related docs
