Deep learning-based neural networks for day-ahead power load probability density forecasting.

Area of Science:

• Electrical engineering
• Climate science
• Data science

Background:

• Energy efficiency is vital for greenhouse gas (GHG) emission reduction pathways.
• Electrical overloads in distribution networks cause outages, reducing efficiency and posing safety risks.
• Accurate load forecasting is challenging due to fluctuating, nonlinear power demand and climate conditions.

Purpose of the Study:

• To develop a novel deep learning model for accurate multi-step-ahead probability density forecasting of power load.
• To address error propagation and accumulation issues in traditional forecasting methods.
• To enhance energy utilization efficiency and mitigate GHG emissions by preventing electrical overloads.

Main Methods:

• Proposed a deep learning-based monotone composite quantile regression neural network (D-MCQRNN) model.
• Extracted multiple non-crossing and nonlinear quantile functions for forecasting.
• Validated the model using hourly power load and meteorological data from Henan Province, China.

Main Results:

• The D-MCQRNN model significantly reduced time-lag and prediction bias.
• Demonstrated noticeable improvements in the accuracy and reliability of multi-step-ahead probability density forecasts.
• Effectively alleviated issues of error propagation and accumulation in forecasting.

Conclusions:

• The D-MCQRNN model offers a robust solution for accurate power load forecasting.
• Improved forecasting reduces the risk and impact of electrical overload faults.
• The model contributes to enhanced energy utilization efficiency and supports cleaner energy production by mitigating GHG emissions.