Photovoltaics (PV) have seen rapid global penetration into the low voltage (LV) electricity distribution grid year-on-year. The result of high PV penetration levels is grid impacts of voltage fluctuations, harmonic distortions and reverse flow among others. Research that attempts to quantify the maximum allowable PV penetration into the LV grid before experiencing detrimental impacts is an important. The most commonly reported barrier to enabling grid impact analysis is the lacking availability of high-resolution and geographically flexible solar irradiance data. As an alternative, synthetically generated solar irradiance data can be used.
There is a distinct lack of synthetic solar irradiance generators that can derive high resolution and statistically accurate solar irradiance data using only readily available inputs. This thesis presents the development of two synthetic generators: the Solar Irradiance Generator (SIG), and the Spatially Decorrelating Solar Irradiance Generator (SDSIG). The SIG proves the concept that synthetic minutely irradiance time series can be generated using readily available mean hourly observations of total cloud amount, atmospheric pressure, wind speed and cloud base height. The SDSIG presents the first ever methodology to synthetically generate unique and spatially decorrelating minutely irradiance time series for any number of uniquely orientated and tilted houses inside a spatial domain using the same inputs as the SIG. The SDSIG employs (1) Markov chains, to derive stochastic weather variable time series, (2) synthetic representations of clouds in the sky, using a novel method called cloud fields, (3) globally flexible irradiance estimation models, and (4) distributions of clear-sky irradiance by total cloud amount, to create the irradiance time series.
The SDSIG outputs are temporally validated using metrics of ramp rates, variability indices and irradiance magnitude frequencies against real world observations at two UK sites and two USA sites, representing three distinct climates. Daily 2-sample Kolmogorov-Smirnov tests of each metric passed a minimum of 95.34% of the time with a 99% confidence limit. The lowest CDF correlation coefficient between modelled and observed data for all metrics and sites was R=0.908; the mean was R=0.987. The SDSIG outputs are spatially validated at Oahu, HI USA, showing R=0.955, RMSE=0.01 and MAPE=0.865% when comparing modelled and observed spatial correlation versus site separation. The SDSIG outputs are applied to a grid impacts power flow model of an LV grid with increasing PV penetration to test the over voltage metric of daily on-load tap changer (OLTC) operations. Using correlating irradiance time series at each house in the LV grid overestimates OLTC operations in every instance of PV penetration when compared to using spatially decorrelating irradiance time series from the SDSIG.
