Solar and Wind Power and Energy Forecasting

This paper proposes and validates a data-driven minute-ahead forecast model for photovoltaic (PV) generation, which is essential for real-time micro-grid scheduling. Unlike day-ahead PV forecasts that heavily rely on weather forecast information, our proposed model does not require such data as it operates in an ultra-short-term time domain. Instead, the model leverages the generation data of the target PV sector and its adjacent sectors to capture short-term factors that affect electricity generation, such as the movement of clouds. The proposed model employs a long short-term memory (LSTM) network to process the data. By conducting experiments with real PV site data, we demonstrate that the information from adjacent PV sectors improves the accuracy of minute-ahead PV generation forecasts by 3.66% in the mean squared error index and 1.19% in the mean absolute error index compared to the model without adjacent sector information. Full article

Effective operation of a microgrid depends critically on accurate forecasting of its components. Recently, internet forecasting competitions have been used to determine the best methods for energy forecasting, with some competitions having a special focus on microgrids and COVID-19 energy-use forecasting. This paper describes forecasting for the IEEE Computational Intelligence Society 3rd Technical Challenge, which required predicting solar and building loads of a microgrid system at Monash University for the month of November 2020. The forecast achieved the lowest error rate in the competition. We review the literature on recent energy forecasting competitions and metrics and explain how the solution drew from top-ranked solutions in previous energy forecasting competitions such as the Global Energy Forecasting Competition series. The techniques can be reapplied in other forecasting endeavours, while approaches to some of the time-series forecasting are more ad hoc and specific to the competition. Novel thresholding approaches were used to improve the quality of the input data. As the training and evaluation phase of the challenge occurred during COVID-19 lockdown and reopening, the building demand was subject to pandemic-related effects. Finally, we assess other data sources which would have improved the model forecast skill such as data from different numerical weather prediction (NWP) models, solar observations, and high-resolution price and demand data in the vicinity of the campus. Full article

Future wind energy potential over Portugal is assessed, using wind speed data from a WRF regional simulation under the SSP5-8.5 scenario for 2046–2065 and 2081–2100. Data from a CMIP6 multi-model ensemble were also used to assess future changes in the Euro-Atlantic large-scale circulation. CMIP6 results have shown a southward displacement of the mid-latitude jet stream during winter, and a northward displacement during spring, summer, and autumn, which causes the northern winds to strengthen during summer along the north-western Iberian coast. Furthermore, in 2046–2065 the wind power density (WPD) should increase between 25% and 50% off the northwest coast of Portugal and in the Serra da Estrela mountain range during summer, which is in agreement with the CMIP6 global ensemble projections. Analyses of WPD’s 2046–2065 daily variability of offshore north-western Portugal reveal a variability increase during winter, spring and summer, as well as more intense extreme WPD events, and less intense extreme events during autumn. The WPD’s 2046–2065 inter-annual variability should increase off the northwest coast, and decrease along the central western and southern coasts, whereas it should increase in the entire studied area in 2081–2100, apart from the northern mountain regions and Cape Raso. Full article

To increase the accuracy of photovoltaic (PV) power prediction, meteorological data measured at a plant’s target location are widely used. If observation data are missing, public data such as automated synoptic observing systems (ASOS) and automatic weather stations (AWS) operated by the government can be effectively utilized. However, if the public weather station is located far from the target location, uncertainty in the prediction is expected to increase owing to the difference in distance. To solve this problem, we propose a power output prediction process based on inverse distance weighting interpolation (IDW), a spatial statistical technique that can estimate the values of unsampled locations. By demonstrating the proposed process, we tried to improve the prediction of photovoltaic power in random locations without data. The forecasting accuracy depends on the power generation forecasting model and proven case, but when forecasting is based on IDW, it is up to 1.4 times more accurate than when using ASOS data. Therefore, if measured data at the target location are not available, it was confirmed that it is more advantageous to use data predicted by IDW as substitute data than public data such as ASOS. Full article

With the development of predictive management strategies for power distribution grids, reliable information on the expected photovoltaic power generation, which can be derived from forecasts of global horizontal irradiance (GHI), is needed. In recent years, machine learning techniques for GHI forecasting have proved to be superior to classical approaches. This work addresses the topic of multi-horizon forecasting of GHI using Gaussian process regression (GPR) and proposes an in-depth study on some open questions: should time or past GHI observations be chosen as input? What are the appropriate kernels in each case? Should the model be multi-horizon or horizon-specific? A comparison between time-based GPR models and observation-based GPR models is first made, along with a discussion on the best kernel to be chosen; a comparison between horizon-specific GPR models and multi-horizon GPR models is then conducted. The forecasting results obtained are also compared to those of the scaled persistence model. Four performance criteria and five forecast horizons (10 min, 1 h, 3 h, 5 h, and 24 h) are considered to thoroughly assess the forecasting results. It is observed that, when seeking multi-horizon models, using a quasiperiodic kernel and time as input is favored, while the best horizon-specific model uses an automatic relevance determination rational quadratic kernel and past GHI observations as input. Ultimately, the choice depends on the complexity and computational constraints of the application at hand. Full article