CROSS-MARKET ANALYSIS OF DEEP LEARNING MODELS FOR ELECTRICITY PRICE FORECASTING


Doç. Dr. Claudıa Fernanda YAŞAR

Tez Türü: Yüksek Lisans

Tezin Yürütüldüğü Kurum: Yıldız Teknik Üniversitesi, Elektrik-Elektronik Fakültesi, Kontrol ve Otomasyon Mühendisliği, Türkiye

Tez Danışmanı: Claudıa Fernanda Yaşar

Tezin Onay Tarihi: 2024

Tezin Dili: İngilizce

Özet:

Electricity price forecasting is a crucial factor for the power industry to operate reliably and cost-effectively. However, the complex and non-linear structure of

electricity prices presents challenges for energy management. To tackle this issue, Artificial Intelligence models such as SARIMAX, LSTM, and CNN-LSTM

have been developed to predict short-term electricity prices. These models are known for their strong analysis capabilities in time series data, enabling the

discovery of hidden patterns. The accuracy of these models was tested using Mean Absolute Error, Root-Mean Squared Error, Mean Absolute Percentage Error, and

percentage accuracy, to compare forecasting methodologies. The Diebold-Mariano test was used to confirm the statistical significance of the difference between the

forecast errors of the two models. Ensemble learning was used to optimize a CNN-LSTM model, which automatically selects the best model by using CNN to

extract valuable characteristics and LSTM to recognize data dependency in time series. This study sheds light on the relationship between natural gas prices and

electricity prices in the Turkish electricity market, where natural gas contributes to 22.91% of the total energy generation. Additionally, the study highlights the

importance of electrical consumption as an exogenous variable to electricity price forecasting in the German market. The EXIST Transparency Platform provided

5-year historical data, including hourly records for electricity prices and natural gas prices in Turkey from 2018 to 2023. The SMARD application provided

hourly historical data from the German electrical market, including electricity prices and electrical consumption for the years 2019, 2020, and 2021. These data were used to validate the models’ prediction performance and to analyze the unexpected pandemic effect on forecasting electricity prices. The results showed

that the LSTM and CNN-LSTM models outperformed the SARIMAX model in terms of accuracy and simplicity, with the CNN-LSTM technique having significant

forecasting advantages. The methods utilized in this study exhibit the potential for intelligent optimization in forecasting electricity prices, as indicated by their

successful outcomes and automated development structure.