MoADL-TLSTM: Multiobjective Automated Deep Learning-Based Transformer-LSTM for Load Forecasting

Area of Science:

• Electrical Engineering
• Computer Science
• Artificial Intelligence

Background:

• Accurate load forecasting is crucial for optimizing energy allocation and economic operation in cyber-physical power systems (CPPSs).
• Deep learning (DL) models excel at capturing temporal patterns but existing methods often rely on single models, lack automation, and are overly complex.
• Manual DL model design requires significant domain expertise and struggles with balancing performance and complexity.

Purpose of the Study:

• To propose a novel multiobjective automated deep learning (MoADL) method for load forecasting.
• To develop a hybrid and lightweight Transformer-Long Short-Term Memory (TLSTM) model automatically.
• To balance model forecasting performance and complexity in cyber-physical power systems.

Main Methods:

• Combined Transformer's global dependency modeling with LSTM's sequential processing for feature extraction.
• Developed a nondominated sorting genetic algorithm II (NSGA-II)-based evolutionary mechanism for neural architecture and hyper-parameter optimization.
• Utilized variable-length encoding, crossover, and mutation operations for evolving TLSTM models.

Main Results:

• The proposed MoADL-TLSTM method demonstrated superior forecasting performance compared to state-of-the-art methods.
• The model achieved a better balance between forecasting accuracy and model complexity.
• Experimental results on 12 real-world Australian load datasets validated the method's effectiveness.

Conclusions:

• The MoADL-TLSTM method offers an effective solution for automated, lightweight, and high-performance load forecasting.
• This approach addresses limitations of existing DL models in CPPSs by enabling automatic hybrid model design.
• The findings suggest a promising direction for improving energy management through advanced AI techniques.