Areal Solar Irradiance Estimated by Sparsely Distributed Observations of Solar Radiation

Abstract

Photovoltaic (PV) power generation is being decentralized and introduced on various scales in electric power systems that are expanding over wider areas. To understand the effect of PV power generation on these electric power systems (in areas ranging from power generation operation and control to voltage management of the distribution system), areal solar irradiance must be estimated for a target area, particularly, for the position and surface area that correspond to the PV power generation equipment location. By applying spatiotemporal kriging and covariance analysis to observed solar irradiance time series (obtained from sparsely distributed solar irradiance meters), we developed a technique that determines areal solar irradiance for an arbitrary location and surface area to estimate its surficial distribution. The weather conditions under which the spatiotemporal kriging technique can be applied were clarified using solar irradiance and meteorological observations during a one-year period at 25 observation points in the Kansai area.Index Terms-Covariance, dependence on weather conditions, photovoltaic generation, spatial averaging effect, spatiotemporal kriging.

Robust cloud motion estimation by spatio-temporal correlation analysis of irradiance data

“…Therefore the success of two algorithms in detecting cloud motion is estimated from simulated dense ground data: cross-spectral analysis (CSA) and the cross-correlation method (CCM). In CSA, the cloud speed and direction are estimated by cross-spectral analysis of the irradiance data at some given locations (sites) through the domain (Inoue et al, 2012;Shinozaki et al, 2014). The CSA method suggested by Inoue et al (2012) and Shinozaki et al (2014) is restricted by the spatial arrangement of the sites such that the cloud direction may be inaccurate if there are only a few distinct relative angles between the pairs of the chosen sites.…”

“…In CSA, the cloud speed and direction are estimated by cross-spectral analysis of the irradiance data at some given locations (sites) through the domain (Inoue et al, 2012;Shinozaki et al, 2014). The CSA method suggested by Inoue et al (2012) and Shinozaki et al (2014) is restricted by the spatial arrangement of the sites such that the cloud direction may be inaccurate if there are only a few distinct relative angles between the pairs of the chosen sites. To remove the restriction, a new CSA approach for cloud motion direction is proposed by selecting the direction with least variation for all the velocities in the cloud motion direction.…”

“…In CSA, the cloud speed and direction are estimated by analyzing the time series of the irradiance data at the given sites through the domain (Inoue et al, 2012 andShinozaki et al, 2014).…”

“…The cloud speed and direction are estimated by CSA of the observed Kt at some given locations. As suggested by Inoue et al (2012) and Shinozaki et al (2014), for each pair of sites i with i = 1, …, n, the time difference ∆ that maximizes the delayed (cross) correlation between the two Kt time series is obtained. Then, the time delay moment M(θ) for all the sites with angle θ is calculated using the delay times for all pairs of sites as…”

Grouped Gaussian processes for solar power prediction

“…A number of studies have confirmed that multi-task learning approaches can be useful for distributed solar irradiance or solar power forecasting, finding that cross-site information is relevant [33,35]. Several studies build on the early work of [26] and consider co-kriging methods for distributed solar irradiance forecasting or spatial prediction (notably [27,33]). Other approaches include a range of linear statistical methods, shown to be competitive at shorter horizons, and neural network methods [16,30,31].…”