Εxtreme Learning Machines and Classification Problems – Application categorizing buildings in energy classes

Postgraduate Thesis uoadl:1319008 298 Read counter

Unit:
Κατεύθυνση / ειδίκευση Διαχείριση Πληροφορίας και Δεδομένων (ΔΕΔ)
Library of the School of Science
Deposit date:
2015-03-30
Year:
2015
Author:
Κόκκινου Άννα
Supervisors info:
Παναγιώτης Σταματόπουλος Επίκ. Καθηγητής
Original Title:
Μηχανές μάθησης υψηλής απόδοσης για προβλήματα κατηγοριοποίησης - εφαρμογή στην ταξινόμηση κτιρίων σε ενεργειακές κλάσεις
Languages:
Greek
Translated title:
Εxtreme Learning Machines and Classification Problems – Application categorizing buildings in energy classes
Summary:
The topic of this thesis is the study and analysis of the extreme Learning
Machine Algorithm and its application on classification problems.
Feed forward neural networks have been extensively used on classification
problems, but their training time with traditional learning algorithms is
extremely slow. Two key reasons for that may be: the slow gradient-based
learning algorithms are extensively used to train neural networks, and all the
parameters of the networks are tuned iteratively by using such learning
algorithms.
The ELM algorithm differs compared to traditional learning algorithms because
it is extremely fast, while achieving high accuracy in classification. In
addition, the parameters of the network are selected randomly and not
recalculated until the end of the training process.
Several ELM variants are presented, which manage to overcome the difficulties
and disadvantages of the classical algorithm achieving better training times
and training accuracy. Some hybrid models are presented as well combining the
classic algorithm ELM with a traditional learning algorithm.
Additionally, some ELM algorithm’s applications are mentioned in classification
problems in multiple scientific fields, such as biomedical, engineering,
biology and music.
Then, a new classification problem is analyzed, concerning the classification
of buildings in an energy class depending on required heating load and cooling
load. It is elaborated how the total primary energy of every building comes out
from heating load and cooling load and there is a thorough description of the
methodology used for buildings classification regarding this total energy
amount.
Finally experimental results are presented and Extreme Learning Machine
Algorithm is compared with other learning algorithms.
Keywords:
Learning algorithm, Data classification, Energy class, Total primary energy, Heating load
Index:
Yes
Number of index pages:
8-10
Contains images:
Yes
Number of references:
42
Number of pages:
66
document.pdf (1 MB) Open in new window