Master thesis presents utilization of demand side response on the ancillary service market in Slovenia and Austria. The relevant regulation on Slovenian and Austrian market is described in detail and a comparative analysis about required reserves, activated volumes and costs of the transmission system operator for the purpose of system balancing is made. In order to optimize the economic result of a given service provider (aggregator), a model for forecasting the auction price for tertiary control in Austria was developed.
The ancillary market in Austria and Slovenia are presented and compared. Analysis shows that the possibility of penetration of the DR on the ancillary services market is heavily affected by the regulation and market design. Despite all difficulties faced by DR in real life operation, DR can reduce the balancing cost born by the system operator.
The developed forecasting model is based on an artificial neural network. The output of the model is the average accepted price of each specific product on the auction for tertiary control in Austrian market, while inputs into the model were carefully selected, representing relevant information affecting the price. There are many factors that have an impact on the product price at the auction as availability of gas generators, solar energy production, day-ahead market price etc. In addition to obvious fundamental factors, there are still many others affecting auction results, such as the price of the previous auction, the number of participants at the auction, offered quantities, offered prices, and many others.
First, an artificial neural network was used for selecting relevant input variables referring to the price of the specific products, which was possible by making a loop that tested predefined sets of input data and the model results attained by them. In the second phase, two models for price prediction have been developed, since price evolution exhibits specific pattern of continuously and slowly declining price with sudden increases (jumps) during winter and summer months, which is associated with unit availability and operation of renewables. The first model using one set of input data predicts the product price at the auction in usual price declining periods, but it is not able to predict price jumps. The second model using different set of input data on one hand produces worse overall results, while on other hand substantially improves prediction of price jumps. This is achieved by utilizing day-ahead price on the German market (EEX spot) and the available capacity of gas production units on German market as input variables. Model for predicting price jumps turned out to be able to predict a jump in 4 out of 5 tests.
Finally, both models are combined into a joint model, which additionally improves the overall result. Results of the developed model were compared with those from naïve model, showing a significant improvement over naïve forecasts.
|
