Multi-region electricity demand prediction with ensemble deep neural networks.

Area of Science:

• Energy Systems Engineering
• Artificial Intelligence
• Data Science

Background:

• Accurate electricity consumption prediction is crucial for intelligent energy management.
• Power supply companies require reliable short-term and long-term energy forecasts.
• Existing prediction models may not fully capture complex temporal dependencies.

Purpose of the Study:

• To develop and evaluate a deep ensembled neural network for hourly electricity consumption prediction.
• To provide an effective approach for anticipating power utilization.
• To assess the model's performance against established statistical metrics.

Main Methods:

• Utilized a dataset of hourly energy expenditure from 2004-2018 across 13 regions.
• Applied Min-Max scalar for data normalization.
• Implemented a deep ensembled model combining Long Short-Term Memory (LSTM) and Recurrent Neural Networks (RNN).

Main Results:

• The deep ensembled model demonstrated strong performance in predicting energy consumption.
• The model effectively captured long-term dependencies within the sequential data.
• Evaluated using metrics such as RMSE, rRMSE, MABE, R2, MBE, and MAPE.

Conclusions:

• The proposed deep ensembled model offers a highly effective solution for electricity consumption forecasting.
• The model's accuracy surpasses existing methods, enhancing intelligent energy management.
• This approach provides a robust tool for power supply companies' planning and operational needs.