Neural models project for solar radiation and atmospheric temperature forecast

Martins, I.

http://hdl.handle.net/10400.1/3993

Use this identifier to reference this record.

Name:	Description:	Size:	Format:
report.pdf		15.12 MB	Adobe PDF	Download

Send Feedback

Authors

Martins, I.

Advisor(s)

Ruano, A. E.

Ferreira, P. M.

Abstract(s)

This work arises from the necessity of temperature and solar radiation forecast, to improve the Heating, Ventilating, and Air Conditioning (HVAC) systems e ciency. To do so, it was necessary to determine neural models capable of such forecast. The chosen characteristics were solar radiation and temperature because these two characteristics directly a ect the room temperature inside a building. This forecast system will be implemented on a portable computational device, so it must be built with low computational complexity. During this dissertation the various research phases are described with some detail. The applications were developed on Python programming language due to it library collection. In this task several algorithms were developed to determine the cloudiness index. The results of these algorithms were compared with the results obtained using neural models for the same purpose. In solar radiation and temperature forecast only neural models were used. The cloudiness index forecast was not implemented as this is only an intermediate step; instead measured values of cloudiness index were used for the solar radiation forecast. Regarding the solar radiation forecast two neural models were implemented and compared, one of the models has an exogenous input, the cloudiness index forecast, and the other one is simply a time series. This models were compared to determine if the inclusion of the cloudiness index forecast improves solar radiation forecast. In temperature forecast only one model will be presented, a Nonlinear AutoRegressive with eXogenous input (NARX) model, with solar radiation forecast as exogenous input. All the neural models are radial Basis Function (RBF) and there structure was determined using a Multi-Objective Genetic Algorithm (MOGA). The models were used to determine cloudiness index, forecast solar radiation and temperature.