The advent of photovoltaics (PV), while promising near limitless green energy, has a serious drawback: the influence of weather conditions on PV output. Accurate and reliable PV power prediction is crucial. Deep-learning algorithms are state-of-the-art methods to counter the non-linear and stochastic nature of PV output. However, such deep-learning models manifest acceptable forecasts only for specific forecast-horizons and data trends. The current study proposes two unique methods: LSTM-ensemble and Meta-Boosting-Network (MBN) using adaptive-weighting and data segmentation-technique. These methods can precisely and consistently predict day-ahead PV output generation. The results establish that the MBN outperforms benchmark models by an average of 50%.
|Qualification||Doctor of Philosophy|
|Award date||6 Jul 2022|
|Publication status||Unpublished - 2022|