Residential Electrical Load, Solar Energy and Electricity Bill Forecasting Using Hybrid Machine Learning Models with Time-of-Use Tariffs: A Case Study of Durban, South Africa

Accurate forecasting of energy consumption, renewable power output and utility expenditure is essential for sustainable planning of residential buildings and improving smart grid integration. This study presents several techniques such as random forest, gradient boosting regression, extreme gradient boosting, deep belief networks, random vector functional link, multi-layer perceptron and hybrid ensemble for forecasting of residential load demand, electricity bills, solar energy generation and solar irradiance. Electricity bills under Time-of-Use tariffs are introduced in the paper to accomplish realistic evaluation of economic implications and facilitation of optimized energy usage and cost savings using real-time residential energy data collected from Durban, South Africa. The performance of the forecasting model is assessed by root mean square error (RMSE), mean absolute error (MAE), mean squared error (MSE), coefficient of determination (R2) and mean absolute scaled error (MASE). The outcomes of the study show that the hybrid ensemble model accomplished the highest forecasting accuracy of the electricity bill with MAE, RMSE, MSE, MASE and R2 of 0.018126, 0.022961, 0.00052719, 0.30006 and 0.97978 when compared to other models. The findings of the research can be used as potential benchmarks for intelligent tariff forecasting, demand response planning, smart energy management and renewable energy integration in residential buildings.