For Educational Purposes Only
Power production from PV power plants is strongly determined by short-term variability of clouds, and it depends on geographic conditions. The impact of cloud variability on large-scale PV power plant output is critical for determining the spatial distribution and sizing of PV plants in order to ensure optimal grid integration and operation.
Eight years of high-resolution SolarGIS solar and meteorological data, and PV simulation tools were applied for the territory of South Africa. Three aspects of cloud impact were quantified on PV production:
(ii) variability statistics of 15-minute power production profiles, and
(iii) occurrence and persistence of daily power production sums.
By studying four geographical levels of aggregation, the Researchers provided an evidence that the integrated power output from more PV power plants, distributed over wider area results in smoother power production profiles, which are also more stable and less fluctuating.
Similarly, persistence (continuous duration) of low power production during cloudy days is weaker if PV power plants from a wider region are connected into one integration point.
Photovoltaic (PV) technology offers electricity with high value at rapidly improving technological and economic terms. PV technology has the advantage of having short lead times, and thus can be quickly deployed to close the energy deficiency gap. However, large capacities of photovoltaics integrated into the transmission and distribution grids, may pose challenges in terms of grid management/operation and planning when there is intermittency and variability due to clouds.
This challenge was identified by Eskom South Africa (Transmission Grid Planning) and therefore initiated the study for the benefit of electricity supply industry in South Africa (PV developers, network owners/operators and policy makers).
Power production from PV power plants shows systematic patterns determined by apparent movement of the sun in the sky. This profile is disturbed by short-term variability driven by clouds. Performance of PV power plants in South Africa depends on diverse geographic conditions, given by atmospheric and ocean circulation and orography. This study shows quantitative evidence that integrated power output from multiple PV power plants, distributed over a wider area, results in smoother daily power production profiles, which are more stable and less fluctuating.
The occurrence and variability statistics of clouds and their effect on photovoltaic power generation for any location in South Africa was calculated. Three aspects were analyzed:
The nature and variability of cloud occurrence was studied at four spatial levels of aggregation: from site-specific, relevant to one PV power plant, to the level of a large province, assuming integrated effect of many power plants.
The integrated power output is smoother and less fluctuating for higher number and more dispersed PV power plants.
Data and methods
For the PV power generation analysis, the SolarGIS high-resolution database and simulation tools were used.
The primary database consists of time series of satellite-derived solar and meteorological data, which are calculated at 5 km resolution and 15-minute time step. Satellite data are used, as this is only source of information covering continuously South Africa over many years. The data cover a continuous period of year 2005 to 2012 (8 years) and they are developed from atmospheric, meteorological and satellite models.
SolarGIS is developed and operated by GeoModel Solar.
The PV power production was analyzed rather than theoretical effect of clouds. This means that changing air temperature and the quality of the atmosphere were also considered as they have an integral impact on typical and extreme power generation.
In the electricity simulation, a mainstream PV technology commonly used in South Africa was considered.
The calculations assumed a large-scale ground-mounted PV power plant with crystalline-silicon modules mounted at 27° North, and high-efficiency central inverters. PV power generation was expressed as relative to the installed DC nominal capacity – in percent of nominal power: numbers close to 0% refer to minimum power generation, while numbers close to 100% indicate maximum possible power generation for the given time step. Simulation was primarily run at 15-minute time step.
In the simulation, the uncertainties given by solar and meteorological data were considered, as well as those of PV technology and of underlying models.
Simulation models and tools developed by GeoModel Solar were used.
Deployment of PV capacity in multiple sites over a larger geographical area stabilises the power generation profile. The reason is that stochastic ramps due to clouds compensate each other. The study calculated the PV power output assuming four theoretical levels of aggregation. One grid integration point of 5-km size was assumed for each aggregation level:
Perez et al showed that size of module field for multi-megawatt PV power plants (25 MW and larger) absorbs short-term variability that is shorter than 1 minute. It also concludes that 15-minute variability becomes spatially uncorrelated at distances of approximately 10 km. This suggests that for full variability description of one power plant (aggregation level 0), use of 1-minute data would be optimal and statistics based on 15-minute data may slightly underestimate the magnitude and occurrence of extreme ramps.
For describing integrated effect of short-term variability for 9 and more power plants, at aggregation level 1, 2 and 3, 15-minute data provided representative results. Geometric sampling method and systematic spatial aggregation was applied for every spatial unit, to provide information over the whole country - South Africa.
PV power generation profiles
Occasional or regular occurrence of heavy cloudiness is observed everywhere in South Africa, and at the level of one PV power plant, the power production can be very low. Even at the noon time (12:00±60min), the minimum instantaneous power production at a single site was found in the range of 2% to 15% of nominal (DC) installed capacity and it depends on the region.
Such a low power production can occur almost in any month, with exception of the season from July to September in the Northern Cape Province, where power production stays minimally at 30% and up to 50% production levels.
For cloudless sky, the highest power production is reached at noon (the sun is at the highest point in the sky). At the level of one power plant, the maximum power production values occur in a relative large range of values (78% to 92%) in South Africa. Lower values are found in regions with higher occurrence of clouds and higher temperature (e.g. in Northeast). High maximum power values are found in mountains with lower air temperature and low occurrence of clouds. From a seasonal point of view, the highest values are seen in March and September. The prevailing power production values around noon are in the range of 60% to 82%, which is also determined by local geography, mainly air temperature and cloudiness.
Power production from PV power plants between 11:00 and 13:00 SAST
The PV power production profile becomes more deterministic, smoother and predicable when PV power plants are deployed over a larger area. Compared to one single power plant, more power plants significantly stabilize the power production profile: already a limited number of PV systems installed in a small area (50- km rectangle) narrows down the maximum to minimum extremes. With increasing aggregation level, the deterministic component (given by movement of the sun) of solar power production becomes prevailing over stochastic component (clouds).
Minimum integrated PV power production at noon time represented by annual percentile
If PV power plants are deployed in a larger area (with increasing aggregation level), a strong increase is observed mainly for the minimum power production. Power production remain relatively stable - only a slight decrease was observed.
Distribution of PV capacity over a larger area increases the minimum values of integrated power production. The effect of aggregation on the minimum production has clear geographical representation.
Overcast weather rarely occurs over large areas in desert regions, while in the South-eastern coastal zones cloudy skies over a larger area are more common. Typical and maximum power production shows changing geographic patterns as larger areas and more PV power plants are integrated into one grid connection point.
Monthly variability (minima and maxima) was the most pronounced at the level of one power plant (aggregation level 0). With increased aggregation (more PV power plants deployed over a larger area) the interval between minimum and maximum monthly PV power production narrows down, while minimum power production limit increases dramatically.
Short term (15-minute) power generation variability
Short-term PV power production changes (ramps) are determined by clouds. At the level of one PV power plant, the magnitude of changes depends on the size of the power plant and the size and the velocity of clouds. Occurrence of negative and positive power production changes is strongly determined by the time of the day. In the morning the positive changes dominate, while in the afternoon the opposite happens.
Power production changes (ramps) from PV power plants between 11:00 and 13:00 SAST assuming 98% occurrence of all changes recorded in 2005 to 2012 in South Africa. The highest magnitude of ramps is seen during intermittent cloudiness: at the level of one power plant, 15-minute ramps may occasionally reach values as high as ±80% of nominal DC power. However 98% of ramps did not exceed ±40% of installed nominal power.
The distribution of installed PV capacity over a larger region reduces the effect of individual ramps, because stochastic power production changes compensate each other. Assuming 98% occurrence, already for small aggregation area at level 1 (50-km square), power production change can be reduced to maximum ±24%.
Aggregation over larger area shows reduction to maximum ±10% and ±6%, for aggregation levels 2 and 3 respectively. The magnitude of the changes depends on geography. Irrespective of cloud situation, fluctuations of PV power are much smoother for the 250-km aggregation area and for larger areas.
Significant differences due to seasonality were not observed, fast changes are rather a result of actual cloud pattern. For clear-sky conditions the variability follows the same pattern regardless of the aggregation level:
Integration over larger areas shows dramatically reduced short-term variability of PV power.
Occurrence of daily weather types and persistence of power production
For each month, daily weather has been categorized into three categories. Daily sum of PV power production has been compared to monthly possible maximum to distinguish between days with:
Occurrence and persistence of these three categories has been analysed. Sunny weather dominates practically in all regions of South Africa. Yet, Eastern and Southeastern regions, especially the coastal zone, are significantly influenced by cloud dynamics. This region has less than 220 sunny days per year, and a higher share of overcast days (more than 25) and days with intermittent cloudiness (more than 120 days).
The rest of the country, including most of the interior land and the Atlantic coast, shows stable weather with many sunny days (more than 220 days a year) and high PV power production, low occurrence of overcast days (less than 25 days a year) and lower occurrence of days with intermittent cloudiness (less than 120 days a year.
With increased aggregation level the occurrence of overcast days with low PV power production strongly reduces: from maximum 71 days for aggregation level 0 to maximum 20 days for level 3. With increased aggregation level, the minimum number of days with high and intermediate PV power production increases.
The highest persistence (stability) is seen in sunny weather. The maximum continuous duration of sunny weather recorded in South Africa is between 10 and 58 days. Even in the cloudiest region, such as Durban, continuous sunny weather may typically last 7 days. In the majority of the land area in South Africa, sunny weather can last more than 30 continuous days.
At the level of one power plant, overcast weather can persist in the range of 1 to maximum 9 continuous days. While in Northern Cape and neighbouring regions cloudy weather and low power production does not last more than 3 to 4 days, in the Eastern provinces this maximum can be up to 7 days and around Cape Town and Durban, the maximum number of days with continuous overcast weather is 9. Even in the cloudiest regions, overcast weather does not typically last longer than 3 consecutive days. In the Northern Cape, overcast days typically do not last longer than 1 day. Weather with variable cloudiness has relative high occurrence in Eastern, Southeastern and Southern parts of the country.
However occasional but long episodes of persistently instable weather (10 days or more) may be experienced also in the Eastern half of the country.
Aggregation of the output from a fleet of PV power plants over larger areas has clear benefits in reducing the effect of continuous periods of overcast days. The maximum persistence of overcast days drops from 9 days for a level of one single power plant to 6 and 4 days for aggregated levels 2 and 3, respectively. In general, the increasing the area of distributed power production results in more stable and more persistent production patterns, unless this trend is disturbed by specific microclimatic conditions (example of Durban with local high occurrence of clouds and an orographic barrier). In regions with relatively homogeneous weather conditions over vast territory, the aggregation reduces low and intermediate production situations and increases high production situations and their persistence. But aggregation over larger territories increases probability that clouds occur somewhere.
The simulations confirm that, more power plants distributed over a larger area produce more stable and predictable integrated power output with two major effects:
- The minimum power production increases (typical and maximum power production remain relatively stable)
- The magnitude and steepness of short-term power fluctuation (15-minute ramps) decreases.
Sunny regions dominate in South Africa. Most of the interior land and the Atlantic coast shows stable weather with sunny days exceeding 220 per year, less than 25 overcast days, and less than 120 days with variable cloudiness. Eastern and Southeastern regions are more influenced by clouds.
With increased aggregation of PV power plants, occurrence of overcast days reduces: from 71 maximum recorded days, for aggregation level 0, to only 20 days for level 3. With increased aggregation, minimum number of days with high and intermediate PV power production increases.
At the level of one power plant overcast weather persists in the range of 1 to maximum 9 continuous days in South Africa. With higher aggregation persistence of cloudy weather with low power production reduces.
In summary, photovoltaic power production capacities that are equally distributed over a larger area, effectively mitigates cloud variability and deliver more stable, persistent and more predictable power production patterns.
The results of this study will be used by ESKOM for planning of transmission and distribution grid infrastructure and PV power generation capacities. They also serve as a base for forecasting and grid operation. The results can also be used by government policy makers to advocate for widely distributed PV plants, owing to fact that PV production is very good all over the country.
The Impact of Cloud Cover on Large Solar Plants
In late August 2011, the U.S. Department of Energy's National Renewable Energy Laboratory (NREL) released the initial data from an ongoing study tracking how clouds can affect large-scale PV power plants in the 30 MW size range. The data produced from the NREL project gives an indication of what happens in a second-by-second time frame when clouds pass over a solar power installation and temporarily block the sun’s incoming radiation. It is hoped that the data models from the long-term monitoring by NREL’s Measurement and Instrumentation Data Center (MIDC) can be used by large-scale PV developers and utilities to develop technical strategies to better manage any solar power fluctuations.
Like wind farms that only produce power when the wind blows, generating electricity from the sun’s radiation happens only when that radiation is available. When clouds block that direct sunlight, there will be a decrease in power output from PV panels. Knowing to what degree there will be a power loss from a large-scale solar power installation means understanding the physical characteristics and impact of a cloud’s shadow as it passes over a large PV system. The more detailed that information is known ahead of time, the better energy managers can plan and control the power plants output. The key is being able to adequately verify what those changes will be and before the NREL study, no such data had existed.
What the NREL study has found so far is that apparently large-scale PV installations are not as quickly or dramatically as impacted by cloudiness as are smaller PV systems, such as those that are 1-MW or less. In fact, the larger the plant, the less the variability in power output due to cloud cover, according to Ben Kroposki, an NREL principal engineer and a leader of the project.
With a small 1-MW PV plant, the power fluctuation can go up or down very quickly since PVs respond rapidly to changes to the available sunlight. And that change can happen in seconds or tens-of-seconds. But with a larger, 30-MW installation, those fluctuations do not occur as rabidly, according to the study. This is particularly true with smaller clouds that simply do not affect the entire large-scale PV array at once.
The Oahu Solar Energy Study
The NREL study established 17 measurement stations near Hawaii’s Honolulu International Airport on the island of Oahu. NREL’s Solar Radiation Research Laboratory measured the impact of clouds with these sensor stations that collected data at 1-second intervals over the course of one year, producing 31.6 million seconds worth of data. Those time-synch measurements were all taken at exactly the same time and measured the level of solar radiation in the sun’s visible spectrum that reached a horizontal surface at ground level. The researchers also designed the equipment to incorporate a global positioning system to include 1-second measurements concurrently for each of the 17 sensor stations.
The collected data allowed NREL to set up a monitoring network that measured exactly how clouds would impact a large PV system through measuring the dips and jumps in PV power output based on changes in cloud cover. That data can then be used to predict what PV power output changes may occur at 1-second intervals for medium and large-scale PV power plants.
One interesting general conclusion is that with very large-scale PV arrays, there is a smoothing-out of the power fluctuations caused by cloud cover compared to the very sharp spikes and drops that a single PV panel or small rooftop PV array experiences when clouds pass.
The Hawaiian study comes in anticipation of what is expected to be the largest PV system for the islands. On March 24, 2011, Hawaii Electric Company had a ribbon cutting ceremony for its new 5-MW SunPower solar facility on the island of Oahu. Axio Power also plans to build a 5-MW solar facility on the island. These are the first two utility-scale projects as part of the state’s aggressive renewable energy plan, with another 20 to 30 MWs of large-scale solar projects in the pipeline for approval. The state already boasts the most installed solar kilowatts-per-person in the U.S., mostly from private residential installations.
Not the First Cloud Study
In a study published in early September 2010, the US Department of Energy's Lawrence Berkeley National Laboratory released a study, Implications of Wide-Area Geographic Diversity for Short-Term Variability of Solar Power, that refuted findings of earlier studies which had concluded that PV power plants could be greatly affected by short-term cloud cover that would limit their practicality as large-scale electric power sources. The Berkeley lab study, which measured PV output in Oklahoma and Kansas, concluded that the geographic diversity of large solar generating sites “are not substantially different from the costs for managing the short-term variability of similarly sited wind in this region.” In other words, it showed that the effect of cloud cover didn’t have as great an effect on PV arrays as formerly thought. The newly released NREL study appears to confirm those results.
The significance of the current cloud cover study project is the potential to not only manage the output of large-scale PV farms, but to also contain those fluctuations so they do not adversely impact power flow to the grid system. The data will also be useful to those researchers at commercial labs and universities who are developing the next generation PVs to better understand how clouds impact large-scale solar plants.
Clouds can cause significant jumps or “ramps” over a very short period of time and those jumps can cause fluctuations in the grid with the potential to cause power fluctuations and even power surges. With such detailed statistical data about how certain cloud patterns impact PV output, researchers can design systems to minimize the impact of passing clouds and mitigate those fluctuations. Such systems may include storage of electricity generated by the PV array or infrastructure improvements and software to stabilize power fluctuations to the grid.
Sharing the Data
NREL says the data from Hawaii can also be used to predict the impact of clouds on solar projects in other areas that have a similar climate. For the U.S., this means areas such as found in Florida and the desert American Southwest, where large-scale solar projects are currently planned or being built. According to the NREL, another study is already planned in Florida using irradiance sensors to monitor the effects on a 300-MW solar power plant.
The DOE-funded study by NREL also included a partnership with General Electric, the Hawaiian Electric Company and the Hawaiian National Energy Institute. But NREL has been given the right to share all knowledge from the data set based on one-second intervals with other researchers, forecasters, utilities and solar developers globally.