10.07.2015   BACKGROUND

Input variable selection

One of the key features in neural network modeling is the selection of the input variables. In general, for the non-linear ANN models there is no systematic approach, which can be followed. Thus, many approaches were conducted in order to identify the ANN architecture that gives the best results. Finally, a three-layer architecture was selected. It has 12 input neurons, 18 hidden neurons and 24 output neurons. Table 2 defines the inputs and outputs of the neural network.

Inputs

Description

1-4

Q(d-1,h)

h=1,4

5-7

Q(d-1,h), Q( d-2,h)

h=24

8

AQ(d-1,h)

h=1

9-11

T(d-1,h), Tmax(d-1),Tmin(d-1)

h=1

12

hour of the day

Outputs

Description

1-24

Q(d, h)

h=1,24

Table 2.: Definition of ANN inputs and outputs d=day index; h=hour of the day; Q=cooling load, AQ=cooling load gradient Q =cooling load forecast. T=temperature, Tmn=minimum temperature and Tmax= maximum temperature.