Nowadays artifical intelligence is growing both in importance and capabilities. Thanks to machine learning many fields of science and engineering achieved huge breaktroughs. We could mention logic games, such as chess or go, or self-driven cars, voice or face recognition. But after all one of the most important usage is the processing of the ever-growing amount of data produced. In my thesis I estimate the power output of a solar park from measurement data of different sources. Weather and photovoltaic forecasts are developing in great volume, which is needed for the objectives of the future’s energy strategies. Furthermore it could bring a solution for the problems of the electricity market.
For my task first I studied various estimating methods which are used on the electricity market, such as that day or next day power estimation. The efficiency of solar panel systems is greatly influenced by weather, especially by overcast, therefore I examined a few weather models as well, which created the base for the estimations. After this I studied the basic structure and functionality of neural networks. There are many methods for the teaching of the neural network, I searched for and chose the optimal one regarding this task. After finalizing the data structures for the estimation I implemented the network. With neural networks there isn’t a universal model for solving prediction problems, so I used and compared different methods to get the best possible estimation. After this I looked for a cloudy day, when and where I could compare the best result of the created neural networks with a simpler physical estimation. Finally I added some proposals for further development of power estimating and forecast methods.