Chair: Christoph Graf, University of Vienna, Austria
Feedforward neural network feature selection for electricity price forecasting: evidence from the German and Austrian day-ahead market
Ralph Grothmann (email@example.com), Merlind Weber (firstname.lastname@example.org)
Forecasting electricity prices is important to market participants in order to optimize their generation portfolio and to reduce their risk exposure. On the German electricity market, a priority feed-in regime for renewable energy has been established which resulted in a rapid capacity growth of renewable electricity. This has led to fundamental changes in the dynamics of short-term energy markets causing increased price volatility. The identification of influencing factors to explain the new energy price dynamics is key for the development of forecast models. This research applies a feedforward neural network ensemble to model hourly prices on the German and Austrian day-ahead spot market and analyzes their price components for the years 2010 to 2013. In order to identify the components, an input feature selection technique is presented which is based on a heterogeneous ensemble of neural networks. The heterogeneity is introduced through randomly selected subsets of inputs from total population. Training with stochastic learning algorithms simultaneously an ensemble of heterogeneous models ensures that the best predictors among a superset of inputs can be identified. The feature selection is based on input-output sensitivities to reveal input saliency and changes in the features over the years. With this technique, the dimensionality of necessary information to predict day-ahead electricity prices is reduced while providing a robust forecasting model. Results show that the most salient features are lagged prices, loads, Swiss and French electricity prices which have a positive monotonous input-output relationship. The effect of gas prices on the electricity prices is minor. For wind electricity production, a negative input-output relationship is identified which increases significantly from 2010 to 2013. The effect of solar electricity production was negligible in 2010, whereas a negative relationship for solar electricity becomes evident by 2013. The findings reflect the changes in the price dynamics on the German and Austrian day-ahead spot market.
Keywords: Electricity market, Feedforward neural networks, Feature selection
Stochastic programming and equilibrium models for studying competition in electricity markets
Huifu Xu (Huifu.Xu.email@example.com), Arash Gourtani (firstname.lastname@example.org), Dali Zhang (email@example.com)
In this talk, we discuss some stochastic programming and equilibrium models for studying competition in electricity markets. We start with a one stage stochastic Nash equilibrium model where generators compete in a single node wholesale spot market with forward contracts and investigate existence, uniqueness of the Nash equilibrium and numerical methods for computing an approximate equilibrium. We then move on to discuss a two stage stochastic equilibrium program with equilibrium (SEPEC) model where generators compete in a multiple node spot market with network constraints. Finally we discuss a two stage multi-objective stochastic programming model for a dominant producer's medium term decision making in electricity industry.
Keywords: Stochastic Nash equilibrium, Sample average approximation, Stochastic bilevel programming
The effect of intermittent renewables on the electricity price variance
David Wozabal (firstname.lastname@example.org), Christoph Graf (email@example.com), David Hirschmann (firstname.lastname@example.org)
In this paper, we analyze the effect of intermittent electricity production from renewables on the electricity spot price variance. Using a static market model, we identify the variance of the infeed from intermittent electricity sources (IES) and the shape of the industry supply curve as two pivotal factors influencing the electricity price variance. The model predicts that the overall effect of IES infeed on the variance depends on the produced amount: while small to moderate quantities of IES tend to decrease the price variance, larger quantities have the opposite effect. In the second part of the paper, we test these predictions using data from Germany, where investments in IES have been massive in the recent years. The results of this econometric analysis largely conform with the prediction from the theoretical model. This has implications for policy makers coordinating subsidy schemes for renewables, flexible production capacities, and electricity storages.
Keywords: Electricity Spot Markets, Renewables, Merit Order