The Latest Update of JMA Numerical Weather Prediction Models and its Solar Power Forecasting Errors

Ohtake, Hideaki; Uno, Fumichika; Oozeki, Takashi; Yamada, Yoshiyuki

doi:10.1541/ieejpes.138.881

Cited by 6 publications

(1 citation statement)

References 9 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The predictive techniques can be roughly classified into three categories: Physical, statistical, and machine learning methods. In the physical method, such as numeric weather prediction (NWP) [9,10], solar irradiance and numerous meteorological parameters, such as cloud coverage and wind speed require to be taken into account to develop a complex physical mathematical model, while the computational complexity and poor anti-interference performance have been an obstacle. Instead, simpler than the physical method, the statistical method exclusively relies on historical data.…”

Section: Introductionmentioning

confidence: 99%

Short-Term Solar Power Forecasting: A Combined Long Short-Term Memory and Gaussian Process Regression Method

Wang

Bao²,

Hua

et al. 2021

Sustainability

View full text Add to dashboard Cite

Solar power is considered a promising power generation candidate in dealing with climate change. Because of the strong randomness, volatility, and intermittence, its safe integration into the smart grid requires accurate short-term forecasting with the required accuracy. The use of solar power should meet requirements proscribed by environmental law and safety standards applied for consumer protection. First, time-series-based solar power forecasting (SPF) model is developed with the time element and predicted weather information from the local meteorological station. Considering the data correlation, long short-term memory (LSTM) algorithm is utilized for short-term SPF. However, the point prediction provided by LSTM fails in revealing the underlying uncertainty range of the solar power output, which is generally needed in some stochastic optimization frameworks. A novel hybrid strategy combining LSTM and Gaussian process regression (GPR), namely LSTM-GPR, is proposed to obtain a highly accurate point prediction with a reliable interval estimation. The hybrid model is evaluated in comparison with other algorithms in terms of two aspects: Point prediction accuracy and interval forecasting reliability. Numerical investigations confirm the superiority of LSTM algorithm over the conventional neural networks. Furthermore, the performance of the proposed hybrid model is demonstrated to be slightly better than the individual LSTM model and significantly superior to the individual GPR model in both point prediction and interval forecasting, indicating a promising prospect for future SPF applications.

show abstract

Section: Introductionmentioning

confidence: 99%