Energy forecasting of the building-integrated photovoltaic facade using hybrid LSTM

Effective building energy management systems need a reliable approach to estimating future energy needs using renewable energy sources. However, nonlinear and nonstationary trends in building energy use data make prediction more challenging for integrating the photovoltaic system. To estimate future energy forecast, this work presents a hybrid approach based on random forest (RF) and long short-term memory (LSTM) using complete ensemble empirical mode decomposition with adaptive noise (CEEMDAN). Initial steps in our suggested procedure include utilizing CEEMDAN to translate the raw energy usage data into multiple components. Then, the component with the most significant frequency is predicted using RF, and the other components are forecasted using hybrid LSTM. Finally, all of the individual parts' predictions are combined to form a whole. Real-world output energy usage data has been predicted to test the suggested strategy. Results from the experiments show that the suggested strategy outperforms the reference methods.

Automated summary

This study proposes a hybrid approach based on random forest (RF) and long short-term memory (LSTM) using complete ensemble empirical mode decomposition with adaptive noise (CEEMDAN) to estimate future energy forecast. The raw energy usage data is translated into multiple components using CEEMDAN. The component with the most significant frequency is predicted using RF, while the other components are forecasted using hybrid LSTM. The predictions from all components are then combined to form the final result. Experimental results show that the suggested strategy outperforms the reference methods.