Date of Award

January 2023

Document Type


Degree Name

Doctor of Philosophy (PhD)


Computer Science

First Advisor

Hassan Reza


Day-ahead electricity price forecasting is a critical research area that revolves around predicting prices in wholesale electricity markets. While significant progress has been made in energy price forecasting, the existence of a state-of-the-art method for accurately predicting prices in the USA energy market remains a topic of debate. The wholesale and retail markets in the USA greatly value improvements in the accuracy of electricity price forecasts. It is evident that renewable energy sources have become increasingly influential in the US power market, enhancing their effectiveness. However, existing forecasting models exhibit limitations, such as inadequate consideration of the impact of renewable energy and insufficient feature selection. Furthermore, the reproducibility of research, transparent depiction of input features, and the inclusion of renewable resources in electricity price forecasting are either lacking or loosely attempted.

In this research, we tackle these issues by providing a wide range of input features, including historical price data, weather conditions, and renewable energy generation. These features are carefully engineered to capture the complex dynamics and dependencies within the electricity market. The inclusion of renewable input features like temperature data to catch solar energy effect, and wind speed data to capture wind energy effects in electricity prices in the USA market make our model unique. Additionally, data preprocessing techniques, such as data windowing, data cleaning, normalization, and feature scaling, are employed to ensure the quality and relevance of the input data. We developed four high-performing hybrid deep-learning models to enhance the accuracy and reliability of electricity price predictions. Our proposed model integrates the Variational Mode Decomposition (VMD) technique with the strengths of four deep learning (DL) architectures, including dense neural networks (DNN), convolutional neural networks (CNN), long short-term memory (LSTM) networks, and bidirectional LSTM (BiLSTM) networks, to capture the intricate patterns and temporal dependencies present in electricity price time series data. To deploy the VMD-DL hybrid model, we created four different combinations, namely: (i) VMD-DNN, (ii) VMD-CNN, (iii) VMD-LSTM, and (iv) VMD-BiLSTM. However, in our study, the VMD-BiLSTM model demonstrates superior performance compared to the other models in all window implementations. The VMD-BiLSTM hybrid model with 24 input features shows only 0.2733 mean absolute error with the MISO market data to forecast prices. The findings of this research contribute to the field of electricity price forecasting by providing an advanced and comprehensive solution tailored to the USA energy market. The proposed hybrid deep neural network models offer valuable insights and practical tools for market participants, energy traders, and policymakers, enabling them to make informed decisions, optimize energy efficiency, and navigate the volatile energy market landscape.