Incorporating causality in energy consumption forecasting using deep neural networks.

Area of Science:

• Energy Science
• Artificial Intelligence
• Machine Learning

Background:

• Accurate energy demand forecasting is vital for planning, distribution, and policy.
• Traditional methods are evolving with machine learning and deep learning advancements.

Purpose of the Study:

• To introduce and evaluate a novel deep learning architecture for energy demand forecasting.
• To leverage weather data and causal relationships for improved prediction accuracy.

Main Methods:

• Developed an "entangled long short term memory" (eLSTM) deep learning architecture.
• Integrated weather data and causal information linking weather indicators to energy consumption.
• Compared eLSTM performance against bidirectional long short term memory (BiLSTM) on three datasets.

Main Results:

• The entangled long short term memory significantly outperformed the bidirectional long short term memory.
• Demonstrated the efficacy of incorporating causal relationships into deep learning models for energy forecasting.

Conclusions:

• The proposed eLSTM model offers a superior approach to energy demand forecasting.
• Findings have significant implications for enhancing energy management and decision-making systems.