John Kim

Energy Consumption Prediction Model

Started June 12, 2024

Abstract or project description

In the past century, energy usage has been spiking rapidly, leaving many under its influence to be negatively affected (Boston University). From an economic standpoint, people who utilize a mass of their energy at peak hours are subjected to higher costs, due to the higher demand for electricity during those times. From an environmental standpoint, an uncontrolled use of energy can lead to a spike in various types of pollution and climate change (ScienceDaily). Hence, these periodic energy fluctuations need to be properly analyzed to predict future energy usage and understand how much more it needs to be reduced over time. To yield an accurate set of predictive data, our machine-learning model will utilize supervised learning to take in past energy usage data from various energy factors to predict future energy usage. This model will maintain the energy consumption rate by analyzing the historical energy usage from January to June of 2007, a time range that encapsulates over 240,000 household energy records. In order to maintain the quantity and quality of the data collection we will utilize data discovery by searching and sharing datasets among a variety of sources on the web. We tested linear regression, polynomial regression, K-nearest neighbors, and neural networks to predict energy usage based on a variety of factors. In this case, we analyzed the global active power, global reactive power, global intensity, and voltage coming from the three submeterings. More specifically, these submeterings focused on the power coming from the kitchen, laundry room, and small portable appliances.