Forecasting Day-Ahead Electricity Metrics with Artificial Neural Networks

Area of Science:

• Artificial Intelligence
• Energy Systems
• Computational Science

Background:

• Accurate electricity price and demand forecasting is crucial for energy market operations.
• The increasing efficiency of artificial neural networks (ANNs) makes them attractive for time-series prediction tasks.
• A standardized comparison of ANN methods for electricity forecasting is lacking.

Purpose of the Study:

• To compare the performance of various ANN architectures for day-ahead electricity price and load prediction.
• To establish a standardized basis for evaluating different neural network models in electricity forecasting.
• To identify the most effective ANN approaches for short-term electricity market predictions.

Main Methods:

• Development and comparison of multiple neural network models, including those with fully connected, recurrent neural, and temporal convolutional layers.
• Utilizing the same dataset for all models to ensure a fair and direct comparison.
• Evaluation based on standardized prediction metrics for electricity price (HUPX market) and load (Montenegro).

Main Results:

• Neural networks demonstrate significant efficiency for short-term electricity forecasting.
• Hybrid models integrating fully connected layers with recurrent neural or temporal convolutional layers achieved the best performance.
• Temporal convolutional networks (TCNs) exhibit strong feature extraction capabilities, indicating potential for future research.

Conclusions:

• Artificial neural networks are highly effective for short-term electricity price and demand prediction.
• Combining different neural network layers, particularly temporal convolutional networks, offers superior forecasting accuracy.
• Further research into temporal convolutional networks is recommended for advancing electricity forecasting methodologies.