Using Temporal Fusion Transformer Predictions to Maximise Use of Renewable Energy Sources

The environmental transition led by the opportunities that arise with the digital transformation is engaging most industry sectors and bringing several unprecedented challenges to energy consumption and sustainability. The increasing adherence to renewable energy sources by the industrial sector not only underscores a significant shift towards sustainability but also reflects a strategic alignment with global efforts to mitigate climate change impacts, thereby marking a pivotal transition in energy consumption patterns and environmental stewardship within the industry. On the other hand, the increasing amount of renewable energy sources is disrupting the energy grid, due to their fluctuating nature and issue that energy has to be instantly consumed. Our contribution to this paper is a novel method that applies the Temporal Fusion Transformer (TFT) to energy data to make accurate multi-target predictions of load demand, energy price and carbon emissions. By utilising TFT, the predictions of national load demand for Serbia were as low as 1.87% MAPE, which is 20% lower than the baseline model. Similarly, for the Slovenian load demand in 2022, the iterative model achieved a MAPE as low as 2.37%, representing a 13.1% improvement over the baseline. We were also able to predict carbon footprint with a MAPE of 18.14% and energy price with a MAPE of 16.28%. Using a multi-target approach, we were able to improve our approach by an additional 2%. Overall the results show an improvement in load demand prediction scenarios based on data from Slovenian and Serbia energy suppliers. The predictions are used as part of a novel scheduler designed to find optimal execution times of energy-intensive tasks, resulting in decreased carbon emissions and utility costs of operations.