A0807
Title: HMM-enhanced LSTM for electricity spot price prediction
Authors: Christina Erlwein-Sayer - University of Applied Sciences HTW Berlin (Germany) [presenting]
Stefanie Grimm - Fraunhofer Institute of Industrial Mathematics ITWM (Germany)
Tilman Sayer - Advanced Logic Analytics (United Kingdom)
Abstract: Electricity spot prices are prone to volatile periods and frequently occurring jumps over time. A prediction of intraday spot prices relies on suitable modelling paradigms to capture these changing dynamics. We develop a long-short-term memory model (LSTM) enhanced by an underlying Hidden Markov model (HMM) to capture regime shifts and ensemble these with the deep learning architecture. Market regimes are adaptively filtered out from the data set and utilised to split the spot price series. We develop an n-state HMM-LSTM which is trained on split regime-specific electricity prices. The prediction is then gained by weighting the LSTM forecast with state probabilities. Combing LSTM with filtered Markov chain probabilities increases the interpretability of predictions. Each activated LSTM is dependent on the filtered state of the underlying market. The proposed model is applied to an extensive data set of German spot prices.
