CLM-former for enhancing multi-horizon time series forecasting and load prediction in smart microgrids using a robust...

Area of Science:

• Electrical Engineering
• Artificial Intelligence
• Data Science

Background:

• Accurate multi-horizon load forecasting is crucial for smart grid stability and efficiency.
• Transformer models like Autoformer capture periodicity but struggle with real-world data's rapid changes.
• Residential electricity consumption exhibits complex long-term trends and short-term fluctuations.

Purpose of the Study:

• To develop a novel deep learning architecture for enhanced residential load forecasting.
• To improve the accuracy of multi-horizon electricity consumption predictions.
• To address the limitations of existing models in capturing localized and dynamic patterns.

Main Methods:

• Proposed CLM-Former, a hybrid deep learning architecture.
• Integrated time series decomposition, autocorrelation-based attention, and a convolutional-recurrent subnetwork (CLM-subNet).
• Evaluated performance on real-world smart meter data against baseline models.

Main Results:

• CLM-Former demonstrated robust and adaptable performance across multiple forecasting horizons.
• The model effectively captured both seasonal dependencies and high-resolution electricity usage variations.
• Outperformed various Transformer-based and deep learning baselines in comprehensive evaluations.

Conclusions:

• CLM-Former is a promising tool for residential energy forecasting.
• The hybrid architecture successfully models long-term periodic trends and short-term dynamics.
• Findings have significant implications for demand response and smart grid management.