Scientists have created a new probabilistic model for 5 minutes ahead PV Power forecasting. The method combines a conventional neural network with bidirectional long -term memory, attention mechanism and natural gradient boost.
A research group led by scientists from Hong Kong Polytechnic University has presented a new probabilistic ultra-red-red period PV-current prediction method based on a conventional neural network (CNN) and a bidirectional long-term memory (Bilstm) with a attention mechanism.
The new technology gets patterns from historical weather data and then predicts the output with natural gradient boosting (NGBOOST). “Quantifying the uncertainty of prediction is becoming increasingly essential for reducing risks and supporting informed decision -making in control and the market for electricity market,” the team said. “In this context, probabilistic prediction methods improve the reliability of the prediction by providing insights into the full probability distribution of potential results. This approach makes a more extensive understanding of prediction of uncertainty possible, so that stakeholders can make better informed decisions.”
In the first step of the method, the system uses meteorological observations and historical PV stream measurements as input and feeds it in the CNN-BILSTM-Matteni network. The CNN is then used to find trends in the short term, while the Bilstm find patterns in the long term. The attention mechanism is used to find the most important time steps.
Finally, the CNN-BILSTM-ENTERING provides abstract functions of time series and passes them on to the NGBOOST. The latter is an advanced machine learning technique that can produce both deterministic and probabilistic predictions. The deterministic prediction presents the solar capacity in five minutes, while the probabilistic prediction extracts a series of power outputs and their respective certainty of the event.
“To validate the effectiveness of the proposed model framework for probabilistic solar pv -current prediction, this study compares it with various benchmark models,” said the academics. “The quantile regression (QR) Model, A Widely Used Method in Probabilistic PV Power Forecasting, was selected as the fundamental reference model. MoreOover, Four QR-Based Deep Learning Models Were Included for Comparon, The Quantile Neur-Network Memory Network (QLSTM) Model, The Quantile Bidirectional Long Short -Term Memory Network (Qbilstm) Model, and the Kwating Gated Recurrent Unit (QGRU) model.
The proposed model and the six benchmarks were tested for databases of three Australian sites. The Desert Knowledge Australia Solar Center (DKASC) -7 is a site of 6.96 kW, using 73 W Cadmium Telluride (CDTE) PV modules; The location of Dkasc-9a has 5.2 kW, made from 130 W copper. Indium, Gallium and Diselenide (CIGS) modules; And the DKASC-13 consists of 175 W monocrystalline silicon modules, with a total capacity of 5.25 kW.
The analysis showed that the proposed model reaches a normalized average absolute error (NMAE) of approximately 5%, a normalized root average square error (NRMSE) of approximately 10%, and a forecast score (SS) of approximately 60%, which a reduction of 20.73-411.68%in NME, Gastdent in SS, Gast in In In In In In In In In In In In In In In In In In In In In In In In In In In In In In In In In In In In In In Inae, Gastent in In Inae, Fully – Fasters in In Inae, Gastent in In Inae, Gastends 15.68). Reported on QR-based deplete models and the NGBOOST model. Moreover, it showed the lowest average NMAE and NRM values and the highest average SS values, taking into account the periodic character of weather patterns.
“Turning to probabilistic pv power forecasting, the proposed model Achievers a continuous ranked probability score (CRPS) ranging from 0.0710 to 0.0898 kW, which is 20.60–42.40% lower than the qr-based deep those ngboostels and 29.42 –40.0.09% 10–90% Confidence Intervals, The Prediction Interval Coverage Probability (PICP) and the prediction interval Normalized average width (PINAW) results indicate that the proposed model offers higher coverage opportunities and narrower average prediction interval widths than the benchmark models.
The results of their analysis were presented in “Probabilistic Ultra-Kort-Term Solar photo photovoltaic asset forecast using natural gradient stimulation with attention-strengthened neural networks“Published in Energy and AI. Scientists from Hong Kong Polytechnic University and the Technical University of Denmark conducted the research.
This content is protected by copyright and may not be reused. If you want to work with us and reuse part of our content, please contact: editors@pv-magazine.com.
