3.1 Characterisation of solar power variability
Materials discussing the variability, correlation with load and spatial diversity aspects of distributed andare covered in this section. CST is addressed only briefly, reflecting the small amount of published material currently available on the topic.
Clouds are the main reason Table 2.generation experiences intermittency (excluding diurnal intermittency). PV generation can drop by 60% within seconds  due to a reduction in solar insolation. The time taken for the cloud to pass is dependent upon cloud height, sun elevation and wind speed. These factors need to be considered in solar power production forecasting . In , it is noted that the variability of solar systems can be characterised across two dimensions, temporal and spatial. The analysis in the existing literature covered in this section investigates intermittency at different timescales, and the significance of these timescales to grid operation and their potential power system impacts in the temporal domain are included in
|Timescale of Intermittency||Potential Power System Impact|
|Seconds||Power quality (e.g. voltage flicker)|
|Minutes to hours||Load following|
|Hours to days||Unit commitment|
The spatial impacts of variability on the electricity grid will vary because of the ‘smoothing’ effect, which describes the solar power production decorrelation that occurs as spatial separation is increased. Despite this, even if this aggregated generation is smoothed, some potential impacts act at a local level and some at a global level. For example, while power quality concerns may be limited to a single feeder (tens of square kilometres), load balancing is an issue which occurs across a much larger area (say, an entire network segment). For this reason, it is clear that an understanding of variability at both tens of kilometres as well as thousands of kilometres (or however large the load balancing area) is required.
According to , the variability of PV generation is reasonably insignificant on cloudless and consistently overcast days (albeit at reduced power output). It is on partly cloudy days that variability in insolation becomes a major concern. At present, although these variations can cause local problems, the bulk grid is only concerned if the outputs of a large number of PV installations vary at the same time, due to the modest amount of PV installed compared to total electrical load.
Figure 2 shows PV output profiles representing relatively large regions in the western USA.
Although reliant on direct beam irradiance, concentrating solar thermal (CST) is inherently less variable than PV due to the thermal mass present in the energy distribution system, which can be water, oil or molten salts . The graph below in Figure 3 is a normal day for a large CST plant in California. The CST plant in the study had six hours of thermal storage which was dispatched to a typical utility load pattern.
3.1.1 Correlation of solar and load
Neither PV nor CST correlates greatly with the morning and afternoon load (demand) peaks. Figure 4 shows correlation between CST, PV production profile and load. Peak PV production occurs between the morning and afternoon load peaks. The CST system shown includes thermal storage, reducing the time to the peak afternoon load to about 3 hours. Without thermal storage the production profile would be similar to PV . These profiles vary throughout the year with PV and CST generating for longer periods during the summer months. At the height of summer (June/July for the Northern Hemisphere and January/February for the Southern Hemisphere), both PV and CST generation begin to overlap with the afternoon load peak, CST (with thermal storage) in particular, as can be seen in Figure 5
3.1.2 Variability of renewable sources and net load
An interesting finding is the variability of monthly solar energy production over the course of three years as discussed in . Figure 6 shows the variation year-to-year for each month over the entire footprint of the study. Variation is likely to be greater over smaller areas. Knowing what kind of variability to expect at certain times of the year will be important for forecasting and planning purposes.
One of the main challenges to the power system is characterised by the instantaneous penetration of intermittent solar generation, i.e. the fraction of total system load being provided by solar generation at a given instant in time. Further analysis in  looked into net load. Solar generation is viewed as negative load, and when this is combined with the actual system load yields a ‘net load’, which corresponds to the power that must be supplied by other resources on the system. The variability of load and solar will drive the change in production of the other generation resources on the system. When load and solar power are both coincidentally increasing or decreasing, the need for other generation to vary their output power will decrease. Of concern to utilities is when load and solar power move in opposite directions (e.g. increasing solar output power and decreasing load, and vice versa) at the same time creating large changes in the net load. Figure 7 shows hourly changes of solar generation and load over a year with 25% solar penetration (relative to total annual load energy). The four quadrants can be described as:
Q1: concurrent increase in solar and load
Q2: increase in solar and decrease in load
Q3: concurrent decrease in solar and load
Q4: decrease in solar and increase in load
A noteworthy aspect is how much greater the variation in solar power is compared to load with solar showing variations of up to 15 GW and load barely 4 GW. Summer (pink dots) sees the most extreme variations. Hours above and below the 5000 MW dotted lines correspond respectively to net load decreases and increases greater than 5 GW. These large variations in the net load have the potential to stress the grid and operating reserves. Looking at Q1 and Q3 (concurrent changes), load changes regularly by over 2 GW at the same time as solar power changes by more than 10 GW (Q1) and 5 GW (Q3). This results in a change in net load of 8 GW up and 3 GW down. According to  utilities in California are concerned with changes in net load over 5 GW. The data for solar generation (25% penetration level) simply scaled up a 5% penetration scenario, and does not take into account the effect of increased spatial diversity on the reduction of solar power variability. Realistically, the hourly variability of solar power would be lower with a more geographically diverse installation of solar generation plants.
A study on the net load variability in the Californian grid was reported in . Figure 8 illustrates the distributions of hourly change in total load (delta) for load-only and net load (total load taking into account wind and solar generation as negative load) for a model of the Californian power grid assuming 33% penetration of renewable energy, wind and solar in this case. Net load is indicated in the figure by ‘L-W-S’. The projection in this study is based on 2006 data. The data is split up into deciles. The 1st decile is a measurement of delta when load is between 90% and 100% of peak load (top 10% of peak load hours) with the 10th decile being a measurement of delta for loads up to 10% of peak load. The thin lines above and below the bars represent the standard deviation of the positive and negative deltas respectively. An increase in hourly net load variability can be seen in majority of the deciles, more pronounced in the higher deciles, with the integration of 33% renewable energy. The standard deviation of net load variability in the 10th decile is seen to increase by 47% with the integration of 33% intermittent renewable energy penetration. According to , the required flexibility to account for the variability of the renewable resource is three standard deviations. System flexibility can be described as the general characteristic of the ability of an aggregated set of generators to respond to the variation and uncertainty in net load.
Figure 9 shows the results of a similar analysis of net load variability, but in this case for each hour of the day. At 6am, we have similar maximum and minimum deltas but the average hourly variation is seen to increase by 25% from 2000MW to 2500MW in the scenario with 33% intermittent renewable penetration. For the afternoon peak, the averages are similar but the maximum and minimum deltas are significantly larger for net load with solar and wind energy integrated. The standard deviation is also noticeably larger for net load, increasing from approximately 1300 to 1600 MW (around 23% increase).
3.1.3 Ramp rates
A ramp rate analysis for a PV system at the NREL/SRRL site in Golden, Colorado, USA, was performed in  where data was accumulated over a 1-year period at 1-minute intervals. Figure 10 shows the distribution of 1-minute and 15-minute ramp rate events and the number of high ramp rate events, both positive and negative, was seen to be much greater for the 1-minute data compared to the 15-minute data. Data of this nature would assist in determining the degree of flexibility required in the network to compensate for the variability of PV systems.
3.1.4 Spatial diversity
Changes in the position of the sun affect the output of all PV plants in a nearly uniform and highly correlated way . Changes in PV output due to clouds are not driven by a similar uniform process. Clouds move across plants affecting one part of a plant before another, or leaving some parts of the plant unobstructed as the cloud passes. The relative reduction in solar irradiance ramps for the aggregate of multiple plants relative to a single point is demonstrated in Figure 11. The reduction in variability due to spatial diversity is seen within large-scale PV plants as well. The graph in Figure 12 compares the variation of a single insolation point to the variation of the output of an entire PV plant rated at 13.2MW. Unfortunately, information on the layout of the PV panels and the area of the plant were not available. The output power did not experience a change of more than 20% in any 10-second period, whereas there were variations of nearly 80% in the irradiance level. The average correlation between the output of one inverter and all the other inverters in the same plant was found to be about 50%, which is an indication of diversity of output within the same PV plant. The smoothing of PV output across plants when aggregated is also investigated in  for 1-minute and 10-minute timescales. The study comprised of six individual PV plants within an approximate 200 square kilometre area in Las Vegas, USA. For the 1-minute timescale, the maximum ramp rate of the six plants ranged from 30-50% and this was seen to reduce to around 20% when the outputs of all plants were aggregated. A similar reduction in variability due to aggregation is seen for the 10-minute timescale.
Source: Hoff and Perez, 2009
Source: Carl Lenox, SunPower Corporation. Adaptedfrom presentation at PV Variability Workshop.
The distance between plants before correlation of irradiance ramps is lost for various timescales is discussed in . These are for plants located in the Great Plains. Irradiance ramps over timescales of 30 minutes were uncorrelated for sites around 50 km apart. Ramps over timescales of 60 minutes were uncorrelated for sites in the order of 150 km apart and ramps over timescales of 15 minutes and shorter were uncorrelated for all distances between sites down to the minimum spatial resolution of 20 km between sites.
|Distance between plants (km)||Required time scale (min) for loss of correlation|
|20 < d < 50||> 15|
|50 < d < 150||> 30|
|> 150||> 60|