The prediction of household electrical energy consumption is of utmost importance for energy conservation and effective energy management due to the world's rising energy demand, which is largely driven by residential structures. Although physical modelling, commonly referred to as the white box has previously been used to predict the energy consumption of buildings, data-driven modelling has recently gained popularity. This is due to the complexity of physical modelling, which uses engineering procedures and methodologies to make forecasts, as opposed to data-driven modelling, which makes predictions using historical energy data in a simpler manner. The data-driven modelling may be based on statistical methods, machine learning, or deep learning.

In this study, different machine learning models and algorithms including both single and ensemble learning models were implemented for the prediction of household electrical energy consumption. The dataset used was a publicly available historical dataset known as the Smart Meter Energy Consumption Data in London Households and it consists of the energy consumption readings taken at half hourly intervals from the smart meter of different households.

The machine learning models and algorithms implemented were random forest, support vector regression, k-nearest neighbour, voting regression ensemble, stacking regression ensemble, bagging regression, Ada boosting regression and gradient boosting regression. The performance of the models in predicting the electrical energy consumption using the historical data was evaluated and compared using evaluation metrics MAE, MSE, RMSE and MAPE. Based on the low error rate of their predictions, the tuned random forest and gradient boosting regression models outperformed the other models. The voting regression model also performed similarly well. The support vector regression and stacking regression models were the least performed of all the models based on the high error value of their predictions.

A graphical user interface (GUI) was implemented as a web application to present the predictive model created for predicting household electrical energy consumption.

Even though the study's aim and objectives for creating a predictive model to predict household electrical energy consumption were met, there were some drawbacks, such as the difficulty in obtaining data and the time constraint.

Prediction of Household Electric Energy Consumption