Forecasting electrical energy consumption using efficient gaussian processes : A case study

Abstract EN

This paper presents an application of Gaussian Processes (GPs) for forecasting electrical energy consumption, a case study is considered, which is the main campus of Assiut University located in Assiut city in the middle of Egypt.

GPs are a tool that can handle uncertainties in prediction.

The study carried out here incorporates the effect of several variables on the prediction process, which constitute the inputs to a GP model.

We consider three significant factors affecting the energy consumption including weather conditions, schedules related to work and human activities, and occupancy.

Each factor is further divided into a number of timedependent variables that are used as the inputs to the GP model, which infers the energy consumption per month.

Based on historical data of these variables, the hyper parameters of the GP model are optimized offline and the resulted model is tested for 12-month-ahead prediction.

In order to enhance the GP forecasting model, a nonlinear autoregressive (NAR) model based on neural networks is used to predict the future values of the inputs in order to use the GP model for prediction.

Different types of the GP kernels are examined.

The performance of the GP models with the different kernels is validated using different validation criteria and compared with a feed-forward neural network for predicting the electrical energy consumption.

This work leads to the best forecasting accuracy results in mean absolute percentage error (MAPE) of about 5 % during a whole year using GP approach.