Mining and Forecasting Energy Consumption Based on Weather Data

Nathaniel Giesbrecht, Owen A. Hnylycia, Carson K. Leung, Junyi Lu, Thanh Huy Daniel Mai, Fan Jiang, Alfredo Cuzzocrea

Abstract: For a modern grid to be reliable, energy efficiency and identifying consistent energy consumption patterns are becoming essential. In this paper, we present a data science solution that analyzes and predicts temporal energy consumption patterns using techniques like frequent pattern mining, traditional machine learning and deep learning. Specifically, our data science solution mines and forecasts energy consumption based on some meteorological and environmental conditions over time series (e.g., hourly or daily data), and examines how weather conditions affect energy usage variation. Evaluation results on a real-world dataset show that our data science solution identified several distinct frequent patterns when using frequent pattern mining with equally distributed bins. These patterns reveal a significant relationship between irradiance and energy consumption, as well as a positive correlation between temperature and energy usage. Furthermore, our solution predicts and compares energy consumption for a specific year using hourly and daily weather data with decision tree, gradient boosting, linear regression, and random forest. Additionally, we applied a long short-term memory (LSTM) model to view energy consumption as time-series data, uncovering patterns in the energy data based on given time steps. These results demonstrate the practicality of our data solution in mining and forecasting energy consumption based on weather data.