Loading...
Publication
Photovoltaic power forecast modeling with artificial neural networks
| Name: | Description: | Size: | Format: | |
|---|---|---|---|---|
| 22.16 MB | Adobe PDF |
Advisor(s)
Abstract(s)
Com uma crescente preocupação relativamente ao consumo energético global, a energia
fotovoltaica surge como uma fonte energia renovável promissora. Esta dissertação é constru
ída sob a premissa de que a capacidade de previsão de potência fotovoltaica produzida
possibilita o aumento de performance da rede elétrica local através de um controlo eficiente
da mesma. O trabalho desenvolvido propõe uma estrutura com a capacidade de previsão de
potência produzida por um sistema fotovoltaico ligado a rede elétrica presente na Universidade
do Algarve. A estrutura de previsão proposta é composta por dois modelos dinâmicos,
não lineares, de previsão e um modelo estático não linear. Redes Neuronais Artificiais foram
usadas como modelos. Os modelos de previsão têm como objectivo fazer previsões da temperatura
do ar e irradiação solar em passos incrementais de 5 minutos para um horizonte de
previsão de 4 horas. O modelo estático é construído para estimar a potência gerada pelo sistema
fotovoltaico e é otimizado através de comparação entre vários tipos de redes neuronais
como o perceptrão multicamadas e funções de base radial, e modelos com escalas temporais
diferentes, aplicados a diferentes estações do ano, bem como um modelo anual.
In a growing concern for the world energy consumption, photovoltaic energy sources are a reliable renewable energy alternative. This thesis is built upon the premise that the forecast of photovoltaic power production can increase performance of local electric network through an efficient network management. The work developed proposes a power production forecast structure based on a grid-connected photovoltaic system in the University of Algarve. The proposed forecast structure is composed of two non-linear dynamic forecasting models and one non-linear static model. Artificial Neural Networks were used in the development of these models which are intended to forecast solar irradiance and air temperature using Radial Basis Functions with 5 minutes time steps within a prediction horizon of 4 hours. The static model on the structure was created to estimate the power generated by the photovoltaic system and it was optimized through comparison between several network architectures (MLP and RBF) and several seasonal models, as well as a annual model.
In a growing concern for the world energy consumption, photovoltaic energy sources are a reliable renewable energy alternative. This thesis is built upon the premise that the forecast of photovoltaic power production can increase performance of local electric network through an efficient network management. The work developed proposes a power production forecast structure based on a grid-connected photovoltaic system in the University of Algarve. The proposed forecast structure is composed of two non-linear dynamic forecasting models and one non-linear static model. Artificial Neural Networks were used in the development of these models which are intended to forecast solar irradiance and air temperature using Radial Basis Functions with 5 minutes time steps within a prediction horizon of 4 hours. The static model on the structure was created to estimate the power generated by the photovoltaic system and it was optimized through comparison between several network architectures (MLP and RBF) and several seasonal models, as well as a annual model.
Description
Dissertação de Mestrado, Engenharia Eletrónica e Telecomunicações, Faculdade de Ciências e Tecnologia, Universidade do Algarve, 2016
Keywords
Energia fotovoltaica Redes neuronais artificiais Previsão Photovoltaic energy Artificial Neural Networks Forecast