Simulation of electricity consumption data using multiple artificial intelligence models and cross validation...

Area of Science:

• Energy Management
• Data Science
• Machine Learning

Background:

• Global electricity production often exceeds consumption, leading to significant financial and energy waste.
• Accurate electricity consumption forecasting is crucial for optimizing resource allocation and reducing losses.

Purpose of the Study:

• To present a comprehensive dataset of monthly community-level electricity consumption in Dubai, UAE (May 2017-December 2019).
• To evaluate the performance of various machine learning models in predicting electricity consumption.
• To analyze the impact of data formatting and size on prediction accuracy.

Main Methods:

• Collected monthly electricity consumption data from Dubai Pulse and demographic/environmental data from Dubai Statistics Center and Dubai International Airport.
• Engineered additional features like expatriate ratio and building occupancy.
• Implemented and compared multiple machine learning models including linear regression variants, SVM, decision trees, ensemble models, and neural networks.
• Trained models on temporally ordered and randomly split datasets, varying test data size, and employed rolling and moving cross-validation (CV) methods.

Main Results:

• Evaluated model performance using metrics such as R-squared, root mean squared error, mean absolute error, and computational time.
• Assessed the reliability of models through rolling and moving CV procedures.
• Identified the dependence of model accuracy on the amount of training data.

Conclusions:

• The presented dataset and model evaluations provide a foundation for accurate electricity consumption forecasting in Dubai.
• The findings can inform strategies to mitigate energy waste and optimize resource management.
• The data can be further utilized to analyze the impact of events like COVID-19 on electricity usage patterns.