Page 377 - Contributed Paper Session (CPS) - Volume 2

CPS1889 Subanar
2.2. Neural network for time series modeling
The capability of NN in approximating any types of relationships in data
makes this method received great attention in complex time series forecasting.
NN is not only powerful in modeling nonlinear processes but also in the linear
one (Zhang, Patuwo, & Hu, 1998).

Figure 2: Architecture of NN (24-10-1)

In this study, the architecture of NN (see Figure 2) which consists of a
number of input nodes, hidden nodes and one output node is trained by
backpropagation algorithm based on the Levenberg Marquardt method. The
activation function for the hidden nodes is tansig while for the output node is
purelin/identity function.

As described in Figure 2, the forecast value of the ith subseries ( ) can
()
̂

be calculated by formula
10
()
̂
() = ∑

=1
where is the weight that connects the th hidden node to the output node
and

()
()
() = ( + ∑ 24 ) = 2/{1 − exp[−2( + ∑ 24 )]} − 1.
0
0
=1

=1

Notation denotes the weight connecting the th input node to the th

hidden node while is the weight for the bias node. The function is tansig
0
function provided in Matlab.

366 | I S I W S C 2 0 1 9