Electrical Energy Consumption Forecasting Using Gaussian Process Regression

This work presents a Gaussian Process Regression (GPR) for forecasting the monthly electrical energy consumption at Assiut University as a function of weather conditions, activities in terms of time schedule, and human occupancy data. The GPR model has been trained using data sets, which represents the inputs to the model including weather data (temperature, humidity, and solar irradiance), data related to activities in terms of time schedule (conferences days and work days), and human occupancy numbers (students, university members' staff, dormitories' residents, employees, university hospitals' staff, and hospitals' patients). Moreover, autoregressive (AR) model is used to estimate the future inputs to the GPR model using time series prediction. The GPR forecasting results are compared with the forecasting results using traditional multiple linear regression. The results show forecasting accuracy of about 5% mean absolute percentage error (MAPE) using GPR.