Explainable attention-based neural network for load forecasting in super smart grids using socioeconomic and power...

Area of Science:

• Electrical Engineering
• Computer Science
• Data Science

Background:

• Super Smart Grids (SSG) require advanced load forecasting for stability and equitable energy distribution.
• Current forecasting methods lack socioeconomic understanding and regional specificity for SSG planning.
• Addressing multidisciplinary challenges in SSG implementation is crucial for successful deployment.

Purpose of the Study:

• To propose a hybrid load forecasting approach for SSG integrating time-series power consumption and socioeconomic metrics.
• To develop a novel deep learning algorithm combining Artificial Neural Network (ANN) and Luong's Attention Mechanism (LAM) for improved generalization.
• To enhance the interpretability of the forecasting model using SHapley Additive exPlanations (SHAP) and Local Interpretable Model-Agnostic Explanations (LIME).

Main Methods:

• Utilized two parallel ANNs to extract load demand features.
• Implemented LAM to dynamically fuse ANN features based on an attention score function.
• Integrated socioeconomic metrics with time-series power data for training.
• Employed SHAP and LIME for model interpretability.

Main Results:

• Achieved a Mean Absolute Percentage Error (MAPE) of 1.78% on 92 Australian suburban zones.
• Outperformed Bidirectional Long-Short Term Memory (BiLSTM), Long-Short Term Memory (LSTM), and Recurrent Neural Network (RNN) models.
• Demonstrated the effectiveness of incorporating socioeconomic metrics in load forecasting.

Conclusions:

• The proposed hybrid forecasting method enhances demand and supply management for SSG.
• Socioeconomic awareness in forecasting is critical for optimizing pricing strategies and ensuring equitable energy distribution.
• This approach provides a foundation for robust and socially conscious SSG deployment.