A sustainable system for predicting appliance energy consumption based on machine learning

Area of Science:

• Sustainable Energy Systems
• Data Mining and Machine Learning
• Environmental Sustainability

Background:

• Energy conservation is crucial for sustainable societies facing fossil fuel scarcity and climate change.
• Accurate energy consumption forecasting is essential for effective energy management and planning.
• Big data-driven Machine Learning (ML) models offer improved predictive capabilities for energy distribution.

Purpose of the Study:

• To develop and evaluate a comprehensive ML model for accurate energy consumption prediction.
• To compare the performance of different classification algorithms in an energy consumption forecasting context.
• To leverage data reduction techniques for optimizing the ML model's efficiency.

Main Methods:

• Utilized MATLAB for Principal Component Analysis (PCA) to reduce data dimensionality.
• Employed Orange 3, a data mining tool, to build a classification model.
• Implemented and compared four classifiers: AdaBoost, Logistic Regression (LR), Naive Bayes (NB), and Stochastic Gradient Descent (SGD).
• Trained the model on a 4.5-month energy consumption dataset collected every 10 minutes using m-bus energy meters.

Main Results:

• AdaBoost demonstrated superior performance, achieving 100% accuracy in energy consumption prediction.
• Logistic Regression (LR) achieved 99.8% accuracy.
• Naive Bayes (NB) and Stochastic Gradient Descent (SGD) showed high accuracy at 99.7% and 99.4%, respectively.
• The confusion matrix was used to evaluate model performance.

Conclusions:

• The developed ML model, particularly with the AdaBoost classifier, is highly effective for accurate energy consumption forecasting.
• PCA-based data reduction in MATLAB enhances the efficiency of ML models.
• The findings support improved energy management strategies and contribute to sustainable development goals.