| Work Detail |
Scientists have created a novel probabilistic model to predict photovoltaic power five minutes in advance. The method combines a convolutional neural network with bidirectional short-term memory, an attention mechanism, and natural gradient boosting. A research group led by scientists at The Hong Kong Polytechnic University has proposed a novel probabilistic ultra-short-term solar PV forecasting method based on a convolutional neural network (CNN) and a bidirectional long-short-term memory (BiLSTM) with an attention mechanism. The new technique extracts patterns from historical weather data and then predicts the outcome using natural gradient boosting (NGBoost). “Quantifying forecast uncertainty is increasingly essential to mitigate risks and support informed decision-making in demand-side management and electricity market bidding,” the team explains. “In this context, probabilistic forecasting methods improve forecast reliability by providing insight into the full probability distribution of possible outcomes. This approach provides a better understanding of forecast uncertainty, enabling stakeholders to make better-informed decisions.” In the first step of the method, the system uses meteorological observations and historical PV power measurements as input data and feeds them into the CNN-BiLSTM-Attention network. The CNN is then used to find short-term trends, while the BiLSTM finds long-term patterns. The attention mechanism is used to identify the most important time steps. Finally, the CNN-BiLSTM-Attention algorithm extracts abstract features from the time series inputs and passes them to NGBoost. The latter is an advanced machine learning technique that enables both deterministic and probabilistic forecasts. The deterministic forecast presents solar energy production over a five-minute period, while the probabilistic forecast extracts a range of energy outputs and their respective certainty of occurrence. “To validate the effectiveness of the proposed model framework for probabilistic solar PV power forecasting, this study compares it with several benchmark models,” the academics state. “The quantile regression (QR) model, a widely used method in probabilistic PV power forecasting, was selected as the fundamental benchmark model. Additionally, four QR-based deep learning models were included for comparison, namely the quantile convolutional neural network (QCNN), the quantile long short-term memory (QLSTM), the quantile bidirectional short-term memory (QBiLSTM), and the quantile recurrent unit (QGRU) model. A separate NGBoost model was trained as an additional benchmark.” The proposed model and the six reference models were tested on databases from three Australian centers. Desert Knowledge Australia Solar Centre (DKASC)-7 is a 6.96 kW site, using 73 W cadmium telluride (CdTe) PV modules; DKASC-9A is a 5.2 kW site, consisting of 130 W copper indium gallium diselenide (CIGS) modules; and DKASC-13 is composed of 175 W monocrystalline silicon modules, with a total capacity of 5.25 kW. The analysis showed that the proposed model achieves a normalized mean absolute error (NMAE) of approximately 5%, a normalized root mean square error (NRMSE) of approximately 10%, and a forecast skill (SS) score of approximately 60%, indicating 20.73–41.88% reductions in NMAE, 15.68–37.24% reductions in NRMSE, and 15.51–61.22% improvements in SS compared to recently published QR-based deep learning models and the NGBoost model. Furthermore, it showed the lowest mean NMAE and NRMSE values ??and the highest mean SS values, considering the periodic nature of weather patterns. “Moving to probabilistic PV power forecasting, the proposed model achieves a continuous ranked probability score (CRPS) ranging from 0.0710 to 0.0898 kW, which is 20.60–42.40% lower than QR-based deep learning models and 29.42–40.09% lower than the NGBoost model. Within 10–90% confidence intervals, the prediction interval coverage probability (PICP) and prediction interval normalized half-width (PINAW) results indicate that the proposed model provides higher coverage probabilities and narrower prediction interval half-widths than the benchmark models. The Winkler score (WS) of the proposed model ranges from 0.2182 kW to 0.7155 kW, consistently outperforming the benchmark models,” the researchers conclude. The results of their analysis are presented in “ Probabilistic ultra- short-term solar photovoltaic power forecasting using natural gradient boosting with attention-enhanced neural networks,” published in Energy and AI . Scientists from The Hong Kong Polytechnic University and the Technical University of Denmark led the research. |
