Comfort in Buildings - Comfort Control in Buildings - page 52

Fig. 3.4 Real PMV index and neural network approximation for the winter data VA2a ( top )and
VA 2 b ( middle ), and absolute error (real PMV—neural network approximation) for both data sets
( bottom )
3.3.1.3 Polynomial Approach
The second approach presented in this section is based on a polynomial approxima-
tion for the PMV index. This model has been obtained using a polynomial regression
modelling tool, more specifically, the MATLAB Polyfitn library (D'Errico 2012 ).
Polyfitn is able to solve the coefficients of a polynomial regression model using clas-
sical linear least squares techniques. Several numerical methods have been used to
implement the Polyfitn library. However, to obtain a more stable solution it is worth
highlighting the use of the QR factorisation with pivoting for solving the system
(D'Errico 2012 ). More specifically, within the framework of linear algebra, the QR
factorisation of a matrix is a decomposition of a certain matrix into a product of an