A hybrid power load forecasting model using BiStacking and TCN-GRU.

Area of Science:

• Electrical Engineering
• Data Science
• Artificial Intelligence

Background:

• Accurate power load forecasting is crucial for energy efficiency and grid stability.
• Existing methods may not fully capture complex load patterns.
• The need for robust forecasting models is increasing with grid complexity.

Purpose of the Study:

• To propose a novel hybrid forecasting model, BiStacking+TCN-GRU, for short-term electricity load prediction.
• To leverage ensemble learning and deep learning for improved forecasting accuracy.
• To demonstrate the model's effectiveness using real-world electricity load data.

Main Methods:

• Feature selection using Pearson Correlation Coefficient (PCC).
• Ensemble learning with BiStacking for preliminary predictions.
• Deep learning with Temporal Convolutional Network (TCN) and Gated Recurrent Unit (GRU) for final predictions.

Main Results:

• The BiStacking+TCN-GRU model achieved high accuracy on Panama's 2020 electricity load data.
• Key performance metrics include RMSE of 29.1213, MAE of 22.5206, and R² of 0.9719.
• The model demonstrated superior performance in short-term load forecasting.

Conclusions:

• The proposed hybrid model offers a significant advancement in short-term load forecasting.
• The combination of ensemble and deep learning techniques proves effective.
• The model shows strong practical applicability for improving energy management and grid operations.