Dayahead PV plant forecast

OData support
Supervisor:
Olaszi Bálint Dávid
Department of Electric Power Engineering

Nowadays machine learning is used in almost every area. Handwritten addresses are analysed with it at post offices, and artificial intelligence is used in translating websites from one language to another to achieve a better result.

The main objective of the thesis was to estimate the day-ahead performance of a given solar panel farm in a 15 minute resolution. First I’ve researched the usual prediction timer intervalls and reviewed the general operation of the electric energy markets. After that I’ve presented the NWP (Numerical Weather Prediction) cloud index model and the Clear Sky model, because these provided the network inputs. In the next chapter I’ve showed the basics of neural networks. During the solution of the task I’ve created a neural network in Python with the help of the Keras library. For the solution I chose the LSTM model. I tried to enhance the prediction by adjusting the parameters like the activation function or changing the number of layers in the network. In the end I’ve compared the results and made some suggestions about possible further developments.

Downloads

Please sign in to download the files of this thesis.