Forecasting Production and Demand for PV-Integrated System with LSTM and Comparison with ANN Model : A case Study at KTH Live-In Lab

Sammanfattning

A growing population requires efficient usage of energy systems to achieve a sustainable path moving forward. With the increasing implementation of renewable energy sources, modeling their production becomes imperative. To maintain a balanced network, the forecast of demand becomes equally important for the system. The improvement of production and load demand forecasting is key for optimization models and can lead to efficient use of energy. This study analyses the use of LSTM to make predictions about PV-production and demand. Furthermore, the model is compared with an ANN model with the same goal for forecasting. Both models are investigated to see which method yield more accurate results and what their respective advantages and disadvantages are for this case. The main finding from the compared models is a higher accuracy for the LSTM model, both with respect to production and demand of the building. On average, the model performs more precise than the ANN model in most metrics used to evaluate the models. Although the LSTM model achieved good performances compared to the ANN model, the hyperparameters could be further adjusted. It could therefore be beneficial to investigate other, more extensive, methods for hyperparameter tuning.