Effect of probabilistic inputs on neural network-based electric load forecasting.

Area of Science:

• Electrical Engineering
• Artificial Intelligence
• Data Science

Background:

• Accurate electric load forecasting is crucial for power system operation and stability.
• Traditional forecasting models often struggle to account for weather-related uncertainties.
• Neural networks offer powerful predictive capabilities but require robust handling of input variability.

Purpose of the Study:

• To develop a novel method for electric load forecasting that integrates weather uncertainties into neural network models.
• To enhance the accuracy of electric load forecasts by considering variable weather conditions.
• To provide confidence intervals for load forecasts, quantifying prediction uncertainty.

Main Methods:

• A hybrid approach combining traditionally trained neural networks with a set of equations.
• The equations are designed to calculate both the mean value and confidence intervals of the forecasted electric load.
• The method was validated using daily peak load forecasts over a one-year period on modified data from a large power system.

Main Results:

• The novel method successfully generated confidence intervals for the forecasted electric load.
• The method demonstrated a more accurate mean forecast compared to using multilayer perceptron networks alone.
• Incorporating weather-related uncertainties significantly improved the predictive performance.

Conclusions:

• The proposed method offers a significant advancement in electric load forecasting by addressing weather uncertainties.
• The inclusion of confidence intervals provides valuable insights into forecast reliability.
• This approach enhances the overall accuracy and robustness of neural network-based load prediction models.