AI Data-Driven Models for Short-Term Load Forecasting

The thesis aims to evaluate the use of modern deep learning models with attention mechanisms which have not been explored thoroughly for use in load forecasting in the past, to predict how much energy a building uses, twenty-four hours in advance. Two modern models were compared to traditional methods of load forecasting, a transformer model and a one-dimensional convolutional neural network. Local data was collected and used in this thesis to ensure that the models take local trends into account. It was found that the modern models are able to make more accurate predictions than the traditionally used models.