Despite all of its obvious benefits the widely spreading solar energy production constitutes several challenges for the grid operators. Photovoltaic power forecasting is an important issue both in terms of grid stability and economical impact. The photovoltaic system with its data logger at BME Department of Electric Power Engineering is perfectly suitable for examinig the power production under different circumstances. The aim of my thesis is to implement a photovoltaic forecasting method with this system that will be appropiate for research or educational use in the future.
In the first part of my dissertation I examined different methods commonly used for photovoltaic forecasting and divided them into groups considering their features. I wanted to choose a method that can be connected to the original system the most easily and avoiding major alterations and infrastructural investments. These were the reasons why I decided to use artificial neural networks which are practical for predicting problems. Their great advantage is that the users are not required to have complex knowledge of the physical context of the process because the network is capable of learning from previous data and recognition of patterns.
The software I made operates with power values from the past and takes in consideration meteorological datas besides datas from sun path calculation, also it simulates that datas from numerical wheater prediction (NWP) are available.
Consideration of the weather forecast has proved to be useful especially at longer time horizons. During the testing of the system the forecast error at 6 hours time horizon with 30 min. sampling rate was 5,1% RMSE%.
At the end of my dissertation I made a suggestion for a student laboratory measurement syllabus based on the accomplished system and I also suggested several ideas about the development of the system in the future.