Evaluation of multistep forecasting model in a home energy management system

Thumbnail Image
Journal Title
Journal ISSN
Volume Title
Insinööritieteiden korkeakoulu | Master's thesis
Sustainable Biomass Processing
Degree programme
Environomical Pathways for Sustainable Energy Systems
Home Energy Management Systems (HEMS) offers to forecast energy consumptions in a household, optimize and operate flexible appliances, as well as the generations, such as the photovoltaic (PV) panels and the battery storages. The forecasting model in a HEMS aims to anticipate the correct consumption for the battery, in order to have the sufficient amount of capacity for solar generation or to have sufficient charge for evening peaks. There is little research on how the accuracy of a forecasting model would affect the energy cost after the battery operation is optimized. In this thesis, a simple HEMS model is simulated with the programming language Python. Several forecasting models are simulated in combination of 3 tariff schemes, 2 seasons, and with 3 sets of household data. The forecasting results were evaluated with common error metrics and compared with the simulated energy cost. The results shows that the common error metrics don’t give a good indication how a forecasting model would perform monetarily. Addition experiments were conducted to explore the more important timesteps in a HEMS to lower the overall cost. The results show the effect of accurately forecasting the consumption decreases, the further the timestep is from the time of forecasting. The first timestep for the next 24 hours has the most significant effect of decreasing the total electricity cost, when predicted perfectly. A custom weighted error metric is tested with the forecasting models. The custom metric only performs marginally better in indicating the monetary performance of a forecasting model. The conclusion of thesis is the error metrics tested does not give a universal indication of how the forecasting model would perform without the optimization part of the HEMS. Secondly, the closer a timestep is to the time of forecast, the more important it is to have the forecast accurate. Therefore, when developing a forecasting algorithm for an HEMS, the focus should be on improving the accuracy of the first few timesteps.
Lahdelma, Risto
Thesis advisor
Lahdelma, Risto
forecasting metrics, HEMS, energy forecasting, energy optimization
Other note