The topic of this thesis is to overview the implementation, comparison, and usability of Time-series Forecasting Methods, especially in the electricity market. A broad spectrum of methods has been used so far, for example, the nearest neighbour based models, ARMA, ARIMA models, artificial neural networks, regression models, moving averages, exponential refinement etc. The purpose of the thesis is to review them and selecting a group, then implementing and testing the method stack. The final goal is to estimate the next day price of the Hungarian Power Exchange.
Since the launch of the Hungarian Power Exchange (HUPX) in 2009, huge number of clearing benefits data have been stored, which are available to perform tests. In addition, MAVIR data publications are also available for the current state of the Hungarian electricity network.
The main problems, which were investigated:
• Processing and analyzing Hungarian stock exchange data.
• Examining the connection between collected data.
• The function of MAVIR data in price evolution.
In the last eight years, the Hungarian stock exchange has also operated in three different environments: independently, in the Czech-Slovak-Hungarian interconnection and also in the 4M Market Coupling
The European Union is developing a common electricity market. In a constantly changing environment, the most important task will be to better predict prices or to determine the evolution of consumption. Better predictions help better economic decisions.
In the thesis, the problem was investigated in several ways. I predicted the following day’s data using average and median of the previous period. I tried the ARIMA model and implemented the nearest neighbour algorithm that measures the proximity of the day with the absolute error. The highest hourly median calculating algorithm works, with the mean square error of 227,034 HUF.