Bagging Ensemble of Multilayer Perceptrons for Missing Electricity Consumption Data Imputation

Area of Science:

• Energy Management
• Artificial Intelligence
• Data Science

Background:

• Accurate energy consumption forecasting is vital for energy management systems (EMSs).
• Missing data in electric load datasets, due to malfunctions or transmission errors, hinder accurate forecasting.
• Existing imputation methods struggle with long missing periods and high historical data dependency.

Purpose of the Study:

• To propose a novel missing-value imputation scheme for electricity consumption data.
• To address the limitations of current imputation methods in handling complex energy data.
• To improve the accuracy of energy consumption data for forecasting applications.

Main Methods:

• A novel imputation scheme based on a bagging ensemble of multilayer perceptrons (MLPs).
• The ensemble utilizes a softmax function to determine the weight of each MLP.
• The network learns from electric energy consumption data and explanatory variables to impute missing values.

Main Results:

• The proposed softmax ensemble network demonstrates superior performance in imputing missing electric energy consumption data.
• Experiments on real-world data confirm the effectiveness of the novel imputation scheme.
• The method successfully handles long periods of missing data and high historical dependencies.

Conclusions:

• The developed imputation scheme offers a significant improvement over existing methods for electric energy data.
• This technique enhances the reliability of data used in energy management systems.
• The softmax ensemble network provides a robust solution for missing data challenges in energy forecasting.