GOSSIP ALGORITHMS FOR SMART GRIDS

Bibliographic Review

D. Koukoula

,N. Hatziargyriou

, Aris Dimeas

and A. Krkoleva

School of Electrical and Computer Engineering, National Technical University of Athens, Athens, Greece

Faculty of Electrical Engineering and Information Technologies

Ss Cyril and Methodius University, Skopje, Fyrom, Macedonia

Keywords: Gossip algorithms, Ad-hoc networks, Power grid.

Abstract: Motivated by the need of application of reliable distributed algorithms in modern power systems, we study

gossip algorithms. The inspiration of gossiping can be adapted to communication, computation or spreading

information needs of power grids related to the control of voltage and frequency, to the load balancing and

to every aspect of control and measurement in a power network. A short reference on related applications is

the evidence of capabilities of gossip algorithms.

1 INTRODUCTION

Modern power grids integrate gradually an

increasing number of dispersed micro-generators

and new kind of loads, such as electric vehicles.

Apparently, customers tend to become more

conscious of their electricity consuming behaviours

and they want to fully understand and take control of

the costs related to energy, while they are waiting to

enjoy uninterruptible power. Therefore, power grids

tend to become smart grids and dispersed generation

must be followed by distributed intelligence

enabling grids to take advantage of ICT

infrastructures.

In this paper we attempt a bibliographic

approach of gossip algorithms, as fully distributed

algorithms of processing and disseminating

information, and we argue that they should be

implemented on smart grids of the future.

Specifically, gossip algorithms could be applied to

power quality control, to information dissemination

of smart meters, to local load balancing, to

determination of the best operation point of a

dispersed generation unit.

2 GOSSIP ALGORITHMS

Gossip Algorithms were firstly described, under the

name epidemic algorithms, in 1987 (Demers, 1987)

as a tool for rapid update awareness among database

replications. The core idea of Gossip Algorithms is

the fast information dissemination among a group of

nodes, such as a rumour is spread within a social

network or like a virus infects a group of humans. So

far Gossip Algorithms have been variously used in

routing in ad-hoc networks, in data aggregation, in

peer sampling. Gossip Algorithms may not offer the

optimally best solution to a problem, but they give

distributed systems a powerful tool for solving a

distributed problem in a good inexpensive way. In a

smart grid, a candidate to participate in gossiping

could be every part having the ability to

communicate with other parts and holds some kind

of data.

2.1 Description

A Gossip Algorithm consists of a sequence of steps,

which are periodically followed by all nodes

interconnected in a network. The scope of the

procedure is the spread of new information in

exponential rate.

A common Gossip Algorithm includes three

main steps: peer selection, decision about the

exchanged data and data processing. Each of these

steps includes a certain procedure, which can be

deterministic or not. The peer selection for the

information propagation can be either probabilistic

or defined under a specified metric. Let the node A

to be the initiator of communication. Node A has a

154

Koukoula D., Hatziargyriou N., Dimeas A. and Krkoleva A..

GOSSIP ALGORITHMS FOR SMART GRIDS - Bibliographic Review.

DOI: 10.5220/0003978701540158

In Proceedings of the 1st International Conference on Smart Grids and Green IT Systems (SMARTGREENS-2012), pages 154-158

ISBN: 978-989-8565-09-9

 2012 SCITEPRESS (Science and Technology Publications, Lda.)

partial knowledge of the candidate nodes, with

which a connection can be established. This set of

nodes is a subset of the nodes of network, which is

known to node A through a network map or through

a periodical procedure, in which each node identifies

its neighbourhood nodes within its communication

range area. Therefore a type of peer selection

algorithm may contain an internal priority list

according to the distance of neighbours. Another

option is to leave the nodes select their peers

randomly according to a possibility density function.

In any case gossiping ensures the new information

will be spread exponentially. The second step of the

gossiping procedure includes a decision about the

data exchanged in each round of communication.

The content of the message sent depends on the final

scope of the algorithm. If the final scope is data

aggregation, the message may contain an average

value. If the scope is the control of a critical growth,

the message may contain the current measurement of

the growth. The third step of gossiping is apparently

connected to the second step and deploys a number

of comparisons or calculations internally.

Formally expressed, gossip algorithms include

operations at any node of a network of n nodes,

which satisfy the following properties (Shah, 2009):

(1)The algorithm should only utilize information

obtained from its neighborhood

(2)The algorithm performs at most O(di logn)

amount of computation per unit time.

(3) Let |Fi| be the amount of storage required at

node i to generate its output. Then the algorithm

maintains O(poly(logn) +|Fi|) amount of storage at

node i during its running.

(4)The algorithm does not require

synchronization between node and its neighbors

(5)The eventual outcome of the algorithm is not

affected by ‘reasonable’ changes in the

neighborhood during the course of running of the

algorithm.

The classification of gossip algorithms is based

on the final scope of the algorithm and on the time

between gossiping rounds. The final scope of the

algorithm can be information dissemination, simple

overall computations (e.g. averaging) or locally

computations of separable functions (e.g. load

balancing). According to the timing of the message

transmissions, a gossip algorithm can be either

synchronous or asynchronous. Furthermore, the

direction of information transmitted is an important

parameter of a gossip algorithm. When gossip

initiator nodes are asking for information from their

partners, we have pulling type of gossiping, while

the action of pushing new information to their

partners, means pushing type of gossiping. There is

also a mixed type of gossiping, which is names as

push-pull, when gossiping nodes inform each other.

The research so far presents best convergence for

push-pull gossiping type.

According to the convergence criterion there are

two different types of gossip algorithms (Renesse,

2008), (Jelasity,2007) : anti-entropy and rumor-

mongering. Anti-entropy gossiping includes

information propagation until it is replaced by newer

information, while rumor-mongering means that

new information is spread until there is high

probability that all the nodes have received the

information. The information spreading with rumor-

mongering can be stopped by when the receiver is

aware that rumor has spread sufficiently

Gossip algorithms gather a variety of

characteristics, which make them ideal for

distributed applications in a changing environment

like a power system.

Gossip algorithms are by default implemented

distributed and there is no need for central

coordination. They do not rely on a static network

topology and are resistant to communication

failures. They are fast and scalable to large systems.

3 RELATED APPLICATIONS

We provide a short description of some of the

related applications. We describe a group of power

grid problems or generally sensor network problems

solved by gossip algorithms.

3.1 Peer-to-Peer Architectures

In (Beitollahi, 2007) authors describe four peer-to-

peer architectures and evaluate them according to

their ability to deploy an overlay network over nodes

of the power grid. As nodes we can see physical part

of the grid, such as circuit breakers or transformers,

or agents representing distributed generators and

loads. Some of the proposed architectures and

routing algorithms are proved to meet better the

needs of modern power grids, to fulfil basic

operations like demand and production matching,

intelligent load shedding, secondary and tertiary

control.

3.2 Decentralized Aggregation

3.2.1 Accelerated Gossip Algorithms for

Distributed Computing

In (Cao, 2006) authors propose an accelerated gossip

GOSSIPALGORITHMSFORSMARTGRIDS-BibliographicReview

155

algorithm with application in averaging

measurements of sensor networks. The need of

acceleration in gossip algorithms for sensor

networks proves their importance.

The core of the algorithm is based on the idea

that each sensor has at its disposal local memory

(e.g. a register), where the sensor stores the current

and previous measurements. The content of the

message transmitted in every round of the gossiping

is deployed by a mixture of these stored values. The

accelerated gossip algorithm offers a tenfold

convergence rate by decreasing only the local

computations, and performs quite well when applied

to overlay networks of more than 20.000 nodes.

3.2.2 Randomized Gossip Algorithms

In (Boyd, 2006) authors have designed a gossip

algorithm suitable for averaging and for

decentralized optimization in an arbitrarily

connected network. We can see the results both of

the synchronous and the asynchronous

implementation of the algorithm. The asynchronous

model of gossiping shows better convergence speed.

We can find a description of a fast aggregation

algorithm for wireless sensor networks, modelled as

Geometric Random Graphs.

3.2.3 Geographic Gossip

In (Dimakis, 2008) authors propose a gossip

algorithm, which exploits the geographic

information contained in nodes. The application

evaluated is on data aggregation of sensor networks.

The algorithm differs from random gossiping in the

way of determination of gossiping peer. The initiator

of gossiping searches the candidate partner having

knowledge of the geographic properties of its

neighbourhood.

3.3 Control of Microgrids

In (Brabandere, 2007) authors implement a set of

distributed control algorithms based on gossiping.

The nodes participating in gossip are distributed

energy resources of a microgrid and the final scopes

of gossip algorithms include the voltage/frequency

control to ensure stable operation, power generation

following local load deviations and the economic

optimization of power production based on a totally

distributed matching. The results show a very

encouraging view of the capabilities of gossip

algorithms as decentralized computational entities

serving the need of modern power systems.

3.3.1 Control of Microsource Generation

Electric generation within a microgrid can consist of

different renewable and conventional fuel

technologies. The selection of operating period and

power level of generators is quite complicated and

depends on cost of fuel, cost of deferral of electric

power, and impact of emissions and differs on

respect with the most important goal of the

microgrid (e.g. energy saving, load-generation

balancing). Modern power systems encompass

dispersed generators with different technical

limitations and interests. The control of this diverse

portfolio matches well to gossip algorithms as the

core of distributed computation.

3.3.2 Domestic Loads Control

Controllable domestic loads can participate in the

optimization process followed by the microgrid.

They can form a group of loads, which can follow a

specified schedule. However, in future power grids

controllable domestic loads should be able of

participating themselves in a more dynamic way in

the energy market.

3.3.3 Load Shifting and Shedding

The microgrid undertakes the load and generation

balancing employing different policies. The control

of generation is a choice already described. Apart

from that, microgrid should find smarter ways to

avoid a black-out during peak hours. Shedding of

loads could be an option, while an elegant way

would be to utilize energy storage or renewables

with pump-storage facilities so as to store energy

during value periods, when conventional generation

is inexpensive and when there is excess of

renewable energy production. The scheduling and

coordination of these complex operations could be a

very interesting field of application of gossiping. A

mesh network could be constructed, containing a

group of nodes (e.g. generators, loads, renewables,

storage facilities).

3.3.4 Ancillary Services

The other side of the microgrid should be selling

power to the main utility grid during overall peak

load period aiming to maximize its profit. Microgrid

could provide valuable ancillary services to the main

utility grid. However, providing these ancillary

services means that the microgrid as an entity has

established a fast and reliable communication with

the main grid, as well as that every node of the

SMARTGREENS2012-1stInternationalConferenceonSmartGridsandGreenITSystems

156

microgrid itself communicates well with each other..

Microgrids may participate in deregulated energy

markets market as both suppliers, and customers of

electricity services, leading to global optimization of

exploitation of natural resources. The key of the

success of this operation is the deployment of

reliable distributed algorithms resistant to

communication failures, such as gossip algorithms.

4 MESH NETWORKS OVER

POWER SYSTEMS

A prerequisite of gossiping is the existence of mesh

networks. The main idea of introducing gossip

algorithms into power grids is to build mesh overlay

networks above the existing structures. The wide

spread of smart meters in low level voltage will

allow the structure of new mesh networks over

power grids. However, there is a need of careful

selection of the adopted communication

technologies. An improved selection could be the

exploitation of wireless low-range communication

technologies, such as Radio Frequencies or

WiMAX, to enable the construction of an overlay

mesh network.

According to Gupta and Kumar an ad-hoc

wireless network can be modeled by a Geometric

Random Graph. The wireless ad hoc network of

nodes is modeled with transmission radius. Nodes

are placed randomly on a grid and are connected

with other nodes in their neighborhood within their

transmission range. An example of such a graph is

shown in the following figure.

Figure 1: Geometric Random Graph.

The representation of power grid parts as nodes

of an overlay mesh network, transform it to smart

grid and enable the implementation of gossip

algorithms to improve performance. Gossip

algorithms should prove themselves as a powerful

distributed tool for information exchange and

decentralized computations in a dynamically

changing network relied on external changes (e.g.

increased load, environmental conditions, unit and

grid failures).

5 CONCLUSIONS

In the context of liberalised energy markets, modern

power grids should play an active role. Distributed

generation and renewable energy resources should

be integrated in a manner that does not harm the

stability and economic operation of the system. The

available options of maintaining the stability and the

cost effectiveness of modern power systems include

flexible management of controllable loads, the

curtailment or shifting of the production of

renewable resources, the use of energy storage

devices. The central control and coordination of

these operations does not always ensure the best

outcome, as it stresses the communication resources

and lacks on reliability due to communication

failures. The choice between the different available

solutions should economic notions, having in mind

the technical limitations. The application of gossip

algorithms on mesh networks, which have been built

over power systems, gives the wanted distributed

impetus and integrates a fairly new ICT

infrastructure with the domain of smart grids.

6 FUTURE WORK

As a future work, we intend to make an evaluation

of different communication technologies for smart

meters and elaborate the results of implementation

of gossip algorithms over these communication

technologies. The simulation of communication is

intended to be held on Network Simulator 3. The

main problem is that the communication simulation

is based on discrete time events, which is in contrast

to the continuous character of the power grid

operation.

REFERENCES

Demers, A., Greene, D., Hauser, C., Irish, W., Larson, J.,

Shenker, S., Sturgis, H., Swinehart, D., Terry, D.,

1987. Epidemic algorithms for replicated database

maintenance. In PODC '87 Proceedings of the sixth

annual ACM Symposium on Principles of distributed

computing

GOSSIPALGORITHMSFORSMARTGRIDS-BibliographicReview

157

Shah, D., 2009. Network gossip algorithms. In ICASSP '09

Proceedings of the 2009 IEEE International

Conference on Acoustics, Speech and Signal

Processing

Renesse, R., Dumitriu, D., Gough, V., Thomas, C., 2008.

Efficient Reconciliation and Flow Control for Anti-

Entropy Protocols. In LADIS '08 Proceedings of the

2nd Workshop on Large-Scale Distributed Systems

and Middleware

Jelasity, M., Voulgaris, S., Guerraoui, R., Kermarrec, A.-

M., van Steen, M., 2007. Gossip-based peer sampling.

In ACM Transactions on Computer Systems Volume

25, Issue 3, Article 8

Beitollahi, H., Deconinck, G., Guerraoui, R., Kermarrec,

A.-M., van Steen, M., 2007. Peer-to-Peer Networks

Applied to Power Grid. In Proceedings of the

International conference on Risks and Security of

Internet and Systems (CRiSIS)' in conjunction with the

IEEE GIIS'07m.

Cao, M., Spielman, D. A., Yeh, E.M., 2006. Accelerated

gossip algorithms for distributed computation. In

Proc. of the 44th Annual Allerton Conference on

Communication, Control, and Computation

Boyd, S., Ghosh, A., Prabhakar, B., Shah, D., 2006.

Randomized Gossip Algorithms. In IEEE

Transactions on Information Theory, Volume 52, p.

2508

Dimakis, A., Sarwate, A., Wainwright, M., 2008.

Geographic gossip: Efficient aggregation for sensor

networks. In Proceedings of 5

ACM/IEEE

International Conference on Information Processing

in Sensor Networks, p.69

De Brabandere, K.., Vanthournout, K., Driessen, J.,

Deconinck, G., Belmans, R., 2007. Control of

Microgrids. In General Meeting of IEEE Power

Engineering Society

SMARTGREENS2012-1stInternationalConferenceonSmartGridsandGreenITSystems

158