Loading...
Please wait, while we are loading the content...
Similar Documents
Modeling and Forecasting the Nord Pool Day-Ahead Power Market through Deep-Learning
| Content Provider | Semantic Scholar |
|---|---|
| Author | Lysfjord, Magnus J. Walker |
| Copyright Year | 2017 |
| Abstract | One of the great challenges faced in the Nord Pool Market by companies and traders is the frequency of behaviour modification due to government policy such as Carbon tax, demand changes, supply changes, as well as changes faced in marginal cost. These behavioural changes have been colloquially described as price spikes or market regimes and are noted as the overall key influential factor in forecasting accuracy. This thesis undergoes a hidden markov model to categorize the hidden states within the most influential stochastic datasets provided by Nord Pool and utilizes this output as input to train the LSTM algorithm. It was found that the methodology in implementing the HMM did not allow for the LSTM to recognize the differing volatile regimes; however, a significant accuracy of 0.0082 mean absolute error was found and bench marked to other results. Recurrent neural networks (RNNs) are suitable for time series phenomena because of their effective dynamic relationship in utilizing changing temporal information. The long-short term memory (LSTM) algorithm arose to solve the RNNs fundamental problem of vanishing gradient problem. The LSTM is thus utilized as the main algorithm in learning the Western European’s largest Market for Electricity, Nord Pool. A key issue in previous deep learning studies has been the identification of key features that are explanatory for the relationship of Nord Pool’s notorious spike regimes. |
| File Format | PDF HTM / HTML |
| Alternate Webpage(s) | https://brage.bibsys.no/xmlui/bitstream/handle/11250/2454314/Lysfjord,%20Magnus%20Johann%20Walker.pdf?isAllowed=y&sequence=1 |
| Language | English |
| Access Restriction | Open |
| Content Type | Text |
| Resource Type | Article |
