Page 362 - From Smart Grid to Internet of Energy

326 From smart grid to internet of energy

operated in single direction and eliminates extra cycles between neurons. The
regular training time of ANNs may be long lasting, but the efficiency on clas-
sification and regression increases its operating speed [20].
The main objective of linear regression method is to learn a function that is
denoted as f(x,w) where f:φ(x) ! y mapping is obtained and a linear combina-
tion of a linear or nonlinear input parameters of φ i (x). This fundamental func-
T
tion is formed as f(x,w) ¼ φ(x) w where w is the weight vector. The
fundamental linear regression functions are composed of polynomial, radial,
sigmoidal, and Gaussian functions. Widely used training models are known
as Ordinary Least Square, Regularized Least Squares, Least-Mean-Squares
(LMS) and Bayesian Linear Regression approaches. LMS is the most rapid
one among others and it can be scaled for high volume of datasets. The com-
binational models can be formed as Classification/Regression and Cluster-
ing/Classification/Regression approaches. The linear regression,
classification and regression trees, support vector regression, random forests,
and bagging methods are some of widely used combinational machine learning
algorithms [22].
In addition to aforementioned algorithms, there are sophisticated tools pre-
sented by corporates for efficient data analytics. Most of the widely used ana-
lytics tools are operated in Standard Query Language (SQL)-based structures.
Moreover, dedicated data servers are also preferred by corporations to decrease
data storage and data processing costs. The R and Weka are corporate level
machine learning tools that are widely used in big data analytics by deploying
machine learning algorithms and models. It is obvious that big data provides
several benefits to corporates in any area including healthcare, smart environ-
ments, traffic managements, financial sectors, security, banking and many
more. Therefore, another important issue to be considered is security and pri-
vacy for secure data processing [23].

8.4 Overview of smart grid privacy
The smart grid devices generate big datasets due to widely used SCADA,
advanced metering infrastructure (AMI), and IEDs at each 2 or 4 s intervals.
The PMUs provide much faster data stacks since they generate 30 or 60 samples
per second, and attach time stamp to measurement data. The replacement of reg-
ular electricity meters with the advanced meter read (AMR) devices comprised
another important component of smart grid big data. Each AMR device gener-
ates 96 metering data a day which comprises 2880 measurement data for each
consumer in a month. The deployment of AMR, AMI, PMU and IEDs have
accelerated generation of massive data sets. A sample calculation has been
noted as 100 PMUs with 20 measurements a day generates over 100 GB data
at standard sampling rates [24]. In some respects, smart grid systems can be
assumed as integration of internet and numerous IEDs used in the power net-
works. Each section of power network including generation, transmission,