3.3 Predicted impacts of high penetration intermittent renewable generation

There have been a number of studies looking into the likely impacts of increased levels of intermittent renewable generation (IRG). They cover areas such as effects on fault response of systems, voltage and frequency regulation, fuel mix, generation flexibility and cost. These are all estimations of impacts, achieved through simulation and modelling. The predicted impacts of high penetration intermittent renewable generation include the displacement of conventional generation units, requirement of curtailing renewable generation output, load frequency control and associated costs.

A thorough survey of papers is given in [16] covering the impact of variable renewable generation power fluctuations on system performance. A general observation is that integrating renewable energy sources in power system grids will have impacts on optimum power flow, power quality, voltage and frequency control, system economics and load dispatch.

From [13], the load duration curve in Figure 16 shows original (without PV) net load and the predicted impact of incorporating 10, 30 and 50% PV penetration. Original net load data is from the California Independent System Operator (CAISO) for July 2007. Actual PV production data from CAISO during the period has been scaled to give penetration levels of 10, 30 and 50%. Some reduction in peak load is seen (far left), and for 30 and 50% PV, a large impact on minimum net load is observed. This will affect the generation portfolio, as it may be necessary to have conventional generation capable of running at a reduced minimum load. A reduction in minimum load due to increased PV penetration will therefore have impacts on the fuel mix. The graph in Figure 17 shows how the typical Californian fuel mix will meet net load demand with 30% solar penetration. In this case, only gas fired generation will be affected. Solar could be valued by the savings achieved through reduced gas generation.

Figure 16 Predicted load duration curve, CAISO July 2007, with 10, 30 and 50% solar [13]

Figure 17 Dispatch order for Californian fuel mix with 30% PV penetration (CAISO July 2007) [13]

The standard US fuel mix has a greater proportion of coal fired generation. The rate of power change of coal fired generation is largely limited by thermal inertia and unit specific fuel system limitations, making them ‘inflexible’. If the same load duration curve is used for the US fuel mix, an incursion into coal fired generation occurs, as seen in Figure 18 . This raises questions of generation flexibility and how coal fired generation will manage requirements to reduce output to allow integration of increased variable renewable generation. It is suggested in [13] that cheaper less flexible plants will need to be replaced with more flexible expensive plants.

Figure 18 Dispatch order for US fuel mix with 30% PV penetration (CAISO July 2007) [13]

Further cost impact studies are discussed in [17]. These studies looked into costs associated with increasing wind generation for systems: Xcel Energy North (Minnesota), Californian Independent System Operator (CAISO) and New York Independent System Operator (NYISO). The Minnesota Department of Commerce study (September 2004) assumed 15% (1500 MW) wind penetration, and the costs resulting from plans and procedures required to accommodate such wind penetration level came to no higher than $4.60/MWh. For CAISO, about 23-25% wind capacity is assumed. At the regulation timeframe (seconds) a maximum wind cost of $0.46/MWh was evaluated. For the load following time scale (ten minutes to a few hours), results focussed on the dispatch stack as a result of the variability introduced by wind. Due to the numerous conventional generators available no impact was measured. The (NYISO) study was the most extensive. Wind penetration of 10% (3300 MW in 34000MW system) is assumed and encompasses all timeframes. This comprehensive cost analysis looked into operating system costs, impact on customer payments, and reduction in emissions from conventional plants and impacts of wind forecasting. Wind generation was modelled using existing weather data and projected out to 2008 on expected demand. The conclusion from the NYISO study was that the New York power system can reliably accommodate at least 10% penetration of wind generation with only minor adjustments to its planning, operating, and reliability practices. 36MW of additional generation would, however, be required to maintain frequency at the no-wind level. In this study, it was reported that significant cost savings could be achieved mainly due to the displacement of fuel, primarily natural gas, by wind and by having accurate forecasts.

Conclusions from [13] are summarised as:

  • costs are moderate at load following and regulation level
  • costs are greater at scheduling level (if wind scheduled in and wind doesn’t blow then conventional generation required to ramp up) with accurate wind forecasting mitigating this impact
  • the greater the area of load balancing and robustness of the system the less the costs of wind integration.

The PÖYRY study [18] for Europe claims “there is a stark difference between the current output of intermittent renewables and what may be expected in 2030” and “wind and solar output will be highly variable and will not ‘average out”’. Figure 19 shows expected wind and solar generation profiles in 2035. This volatility directly impacts on cost, as shown in Figure 20. Prices are expected to become peakier and less predictable, representative of the nature of weather systems. There are likely to be short periods of very high cost when renewable generation is low and extended periods of low cost when renewable (and nuclear) generation is high. The significant difference in pricing volatility between 2010 and 2030 can be seen in Figure 20.

Figure 19 Aggregate hourly output for Northern Europe across four months [18]

Figure 20 Monthly wholesale prices (€/MWh) across different historical weather patterns [18]

The costs associated with required curtailment of wind generation are analysed in [19]. Figure 21 shows the relative cost (relative to cost if no wind curtailment in place) of higher levels of wind penetration for different levels of flexibility (minimum load as a percentage of annual peak load). As stated before, system flexibility can be described as the general characteristic of the ability of the aggregated set of generators to respond to the variation and uncertainty in net load. There is considerable difference between the average and the marginal costs, especially at higher penetration levels. For example, to achieve 50% wind penetration level in a 90% flexible system (i.e. thermal generators are able to cycle down to 90% below the annual peak load), the average cost of wind generation would be about 1.2 times the base cost. However, at the margin, the ‘last’ unit of wind generation installed to meet the 50% penetration level would cost about two times the base cost. For 80% penetration, higher flexibility is required and the average cost is just below 1.8 times the base cost but the marginal cost (not shown) is five times the base cost because of the high level of curtailment required.

Figure 21 Average and marginal relative cost of wind as a function of wind energy penetration due to varying levels of curtailment [19]

The hourly operation of the entire interconnected grid of the western United States with 25% solar penetration over a period 3 years was simulated in [6]. The dispatch profiles for a scenario with no solar power penetration and another with 25% solar penetration are shown in Figure 22 and Figure 23 respectively. A drastic change in dispatch is observed when the two scenarios are compared. Solar generation is seen to displace significant amounts of combined cycle generation with periods of zero production during peak periods. The generation from the coal units is also seen to be reduced during this time. The backing down of coal is a marked contrast to minimum load challenges associated with high wind generation scenarios that result in coal cycling during the night. Hydro generation is seen to shift from the load peak into subsequent load trough. Gas turbines are necessary on days with actual solar generation below forecast or with peak loads occurring after solar generation decreases [6]. It should be noted that the 1-hour resolution in this study is too slow for proper analysis of the impacts of rapid fluctuations in solar power generation, in the order of minutes or seconds, on the dispatch profile.

Figure 22 Dispatch for week in April - no solar [6]

Figure 23 Dispatch for same period as Figure 22 above - 25% solar [6]

Analysis in [6] found savings are to be had from introducing solar power, by displacing gas and coal fired generation and from a price on carbon. For the Western Electricity Coordinating Council (WECC), under the assumptions outlined below, $11 billion a year is saved in 2009 out of total operating costs of $43 billion, corresponding to a 25% reduction:

  • $2/MBTU coal, $9.50/MBTU gas, $30/tonne CO2 for 2017 with 2% annual escalation
  • extensive balancing area cooperation
  • all units economically dispatched while respecting transmission limitations
  • generation equivalent of 6% held in reserve, half spinning, half not spinning
  • 25% solar penetration.

The impact on spot price due to the integration of solar generation in Arizona was also analysed in [6]. It was found that adding new, zero marginal cost resources will generally decrease spot prices. However, large forecast errors with higher solar penetration cause expensive generation to be brought online to make up for these errors, leading to the observation of a diminishing benefit at high prices seen in Figure 24.

Figure 24 Spot price duration curve with 25% solar in Arizona [6]

A survey of papers on the likely factors limiting PV penetration is summarised in a report by Sandia National Laboratories [20] and presented in Table 4. The reported reasons restricting high PV penetration levels are broad and include:

  • ramp rates of conventional generation
  • reverse power swings
  • frequency control
  • voltage regulation.

It should be noted that the different papers looked at different types of grid and penetration contexts. For example some looked at distribution system penetration while others looked at central generation penetration. The maximum PV penetration level is highly situation-dependent and varies widely, as can be seen in Table 4 where the different papers looked at different scenarios.

Table 4 Expected factors limiting PV penetration level [20]

Reference # Maximum PV penetration level Cause of the upper limit
2 5% Ramp rates of mainline generators. PV in central-station mode.
4 15% Reverse power swings during cloud transients. PV in distributed mode.
5 No limit found Harmonics.
6 > 37% No problems caused by clouds, harmonics, or unacceptable responses to fast transients were found at 37% penetration. Experimental + theoretical study.
8 Varied from 1.3% to 36% Unacceptable unscheduled tie-line flows, The variation is caused by the geographical extent of the PV (1.3% for central-station PV). Results particular to the studied utility because of the specific mix of thermal generation technologies in use.
10 10% Frequency control versus break-even costs.
11 Equal to minimum load on feeder Voltage rise. Assumes of LTCs in the MV/LV transformer banks.
12, 13 < 40% Primarily voltage regulation, especially unacceptably low voltages during false trips, and malfunctions of SVRs.
16 5% This is the level at which minimum distribution system losses occurred. This level could be nearly doubled if inverters were equipped with voltage regulation capability.
18 33% or ≥ 50% Voltage rise. The lower penetration limit of 33% is imposed by a very strict reading of the voltage limits in the applicable standard, but the excursion beyond that voltage limit at 50% penetration was extremely small.

One of the challenges of integrating high penetration solar into the grid is the Load Frequency Control (LFC) required to manage power fluctuations from installed PV systems. A study was performed to investigate the extent of LFC required to manage power fluctuations from PV systems installed at five sites across the city of Nagoya in Japan [21]. The average insolation across the sites is used to calculate the combined power output assuming the PV capacity is 2% of total generation. Figure 25 illustrates the LFC capacity required to manage the fluctuations in power output from the PV. The total number of days in the study is 30, which corresponds to the sum of the magnitude of each of the columns in the plots. The bottom plot (LFC capacity of generators = 100% of PV capacity) shows 17 days where the LFC could not manage the fluctuations in load due to the PV. Only when the LFC is at 500% of installed PV are the fluctuations in load for all days managed. This result shows the extent of ancillary services or additional generation required to manage PV power output fluctuations.

The impact of intermittency on voltage regulation in distribution systems with high PV penetration levels was studied in [43], specifically voltage flicker and excessive transformer tap changes that may result from induced fluctuations in PV power. Actual load and solar irradiance data was used to simulate the impact of 20% PV penetration on a simple distribution system on partly cloudy days. Simulations were performed using 1-minute sampling rate and no flicker problems were encountered, but a significant increase in the number of transformer tap changes was observed, which would likely reduce the life expectancy of the load tap changing mechanism. The difference in the number of tap changes, with 1-minute and 5-minute time delays, when PV is integrated into the distribution system can be seen in Table 5. For 1-minute time delay, the number of tap changes was found to be four times higher with PV integrated compared to the base case without PV. The number of tap changes for the 5-minute delay scenario was 2.8 times higher than the corresponding base case. A higher rate of sampling (in the order of seconds) is needed to analyse voltage flicker more accurately as the 1-minute sampling interval is not sufficient to take into account the effects of faster changes in solar irradiance.

(a) capacity of generators available for LFC

Ka: 500% of PV capacity (10% of system capacity)

(b) capacity of generators available for LFC

Ka: 300% of PV capacity (6% of system capacity)

(c) capacity of generators available for LFC

Ka: 100% of PV capacity (2% of system capacity)

Figure 25 Required LFC as % of installed PV [21]

Table 5 Number of tap changes with and without PV (simulated) [43]

Date (d/m/y) w/o PV and 1 min. delay w/o PV and 5 min. delay with PV and 1 min. delay with PV and 5 min. delay
11/7/2010 20 14 92 34
12/7/2010 10 8 42 24
13/7/2010 10 6 26 20
Total 40 28 160 78

Various other studies have been conducted on the potential impacts of solar intermittency on different networks throughout the world, with varying levels of solar penetration and different generation and load profiles. Some general conclusions that can be drawn from these studies are:

  • The amount of solar generation that can be integrated into the utility power system without compromising grid stability and reliability varies widely. The determining factors are the amount of a utility’s load fluctuation and the regulating capability of existing conventional generating units. This observation indicates that the effect of solar generation intermittency on the power system is context specific and needs to be considered on a case-by-case basis. Therefore, a general cause-effect conclusion cannot be drawn.
  • Although high penetration levels of solar generation have the potential to cause adverse network impacts, corrective measures are available, such as assigning more generating units to regulating duty or installing fast-response generation such as combined-cycle generators. These measures can be effective if carefully planned. Simulations have shown high penetration limits are possible after adding fast-response, combined-cycle generating units to the existing generation mix or fast-acting energy storage. Another method is to control the solar generation output under intermittent cloud coverage during periods of peak system demand when the network has fewer generating units on standby and less on-line regulating capacity. However, these corrective measures may cause the system to deviate from its optimal operating condition, adversely affecting the economics of solar generation.
  • The worse-case cloud patterns turned out to be cumulus (fast-moving, well-defined clouds with clear sky between the clouds) and squall (a solid line of dark clouds moving across a clear sky). The squall line cloud pattern can cause complete loss of PV generation [44]. The speed of clouds and size of the dispersed PV system area determine how fast the complete loss occurs. For a 10km2 area, a squall line caused the loss of all PV generation within that area in 1.8 minutes [44]. The worst cumulus clouds, while causing less loss of PV generation than the squall line, might actually pose a more difficult problem for utilities, because their changes are random and their effect on the PV output is much less predictable than that of a squall line.