Short-term power load forecasting based on the CEEMDAN-TCN-ESN model

Area of Science:

• Electrical Engineering
• Data Science
• Time Series Analysis

Background:

• Efficient power system operation relies on accurate load forecasting to balance supply and demand.
• Inaccurate forecasting leads to energy waste and economic losses.
• Existing models struggle with the complex, dynamic features of power load data.

Purpose of the Study:

• To develop a novel hybrid model for enhanced short-term power load forecasting.
• To improve the accuracy and efficiency of electricity load predictions.
• To address the challenges posed by complex load data characteristics.

Main Methods:

• Proposed a hybrid model combining Complete Ensemble Empirical Mode Decomposition with Adaptive Noise (CEEMDAN), Temporal Convolutional Network (TCN), and Echo State Network (ESN).
• Utilized CEEMDAN for decomposing load data into intrinsic mode functions (IMFs) representing different frequencies.
• Reconstructed higher and lower frequency components and applied TCN and ESN for forecasting.

Main Results:

• The CEEMDAN-TCN-ESN model achieved high accuracy on Panama's national electricity load data.
• Achieved Root Mean Square Error (RMSE) of 15.081 and Mean Absolute Error (MAE) of 10.944, with an R-squared (R2) of 0.994.
• Demonstrated a 9.52% reduction in RMSE and a 17.39% reduction in MAE compared to the CEEMDAN-TCN model.

Conclusions:

• The hybrid CEEMDAN-TCN-ESN model effectively captures complex features in short-term power load data.
• The model successfully merges subseries based on similar features, learning from both high-frequency and low-frequency components.
• This approach shows significant potential for real-world applications in short-term power load forecasting.