SqueezeNet for the forecasting of the energy demand using a combined version of the sewing training-based...

Area of Science:

• Electrical Engineering
• Artificial Intelligence
• Data Science

Background:

• The increasing complexity of power systems with dispersed loads necessitates precise forecasting.
• Real-time analysis of power system performance and load changes is vital for efficient energy management.
• Long-term electric load forecasting enhances grid stability, reduces equipment failures, and ensures reliable electricity output.

Purpose of the Study:

• To develop an accurate method for short, medium, and long-term electricity load forecasting.
• To improve the performance of the SqueezeNet model for power demand prediction.
• To validate the proposed forecasting technique against existing methods.

Main Methods:

• Utilized SqueezeNet for initial power demand forecasting.
• Enhanced the SqueezeNet architecture with a customized Sewing Training-Based Optimizer.
• Collected data at 20-minute intervals within a time window for neural network training.

Main Results:

• The proposed method achieved Mean Squared Errors (MSE) of 0.48 for short-term, 0.49 for medium-term, and 0.53 for long-term forecasting.
• Demonstrated superior accuracy compared to other published forecasting techniques in a case study.
• The enhanced SqueezeNet model provided highly accurate predictions across different time horizons.

Conclusions:

• The proposed SqueezeNet enhancement using the Sewing Training-Based Optimizer is a viable and accurate solution for energy consumption forecasting.
• The method effectively addresses the need for precise load forecasting in modern power systems.
• The findings support the adoption of this technique for improving energy management and grid reliability.