TODAY’S STUDY: GETTING MORE SOLAR ON THE GRID
Integrating Solar PV into Utility System Operations Mills, et. al., October 2013 (Argonne National Laboratory)
Deployment of solar photovoltaic (PV) power generation is growing rapidly in the United States. Utilities and system operators are increasingly conducting studies of the impact of PV on operations, including assessments of short-term variability and uncertainty. Consideration of the complex issues surrounding sub-hourly variability and forecasting of PV power output has still been somewhat limited because of the difficulty of creating realistic sub-hourly PV datasets and forecast errors for future scenarios with increased PV production. How utility operations should be changed to more economically integrate large amounts of solar PV power is an open question currently being considered by many utilities. This study develops a systematic framework for estimating the increase in operating costs due to uncertainty and variability in renewable resources, uses the framework to quantify the integration costs associated with sub-hourly solar power variability and uncertainty, and shows how changes in system operations may affect these costs. Toward this end, we present a statistical method for estimating the required balancing reserves to maintain system reliability along with a model for commitment and dispatch of the portfolio of thermal and renewable resources at different stages of system operations. We estimate the costs of sub-hourly solar variability, short-term forecast errors, and day-ahead (DA) forecast errors as the difference in production costs between a case with “realistic” PV (i.e., sub- hourly solar variability and uncertainty are fully included in the modeling) and a case with “well behaved” PV (i.e., PV is assumed to have no sub-hourly variability and can be perfectly forecasted). In addition, we highlight current practices that allow utilities to compensate for the issues encountered at the sub-hourly time frame with increased levels of PV penetration.
In this analysis we use the analytical framework to simulate utility operations with increasing deployment of PV in a case study of Arizona Public Service Company (APS), a utility in the southwestern United States. In our analysis, we focus on three processes that are important in understanding the management of PV variability and uncertainty in power system operations. First, we represent the decisions made the day before the operating day through a DA commitment model that relies on imperfect DA forecasts of load and wind as well as PV generation. Second, we represent the decisions made by schedulers in the operating day through hour-ahead (HA) scheduling. Peaking units can be committed or decommitted in the HA schedules and online units can be redispatched using forecasts that are improved relative to DA forecasts, but still imperfect. Finally, we represent decisions within the operating hour by schedulers and transmission system operators as real-time (RT) balancing. We simulate the DA and HA scheduling processes with a detailed unit-commitment (UC) and economic dispatch (ED) optimization model. This model creates a least-cost dispatch and commitment plan for the conventional generating units using forecasts and reserve requirements as inputs. We consider only the generation units and load of the utility in this analysis; we do not consider opportunities to trade power with neighboring utilities. We also do not consider provision of reserves from renewables or from demand-side options.
We estimate dynamic reserve requirements in order to meet reliability requirements in the RT operations, considering the uncertainty and variability in load, solar PV, and wind resources. Balancing reserve requirements are based on the 2.5th and 97.5th percentile of 1-min deviations from the HA schedule in a previous year. We then simulate RT deployment of balancing reserves using a separate minute-by-minute simulation of deviations from the HA schedules in the operating year. In the simulations we assume that balancing reserves can be fully deployed in 10 min. The minute-by-minute deviations account for HA forecasting errors and the actual variability of the load, wind, and solar generation. Using these minute-by-minute deviations and deployment of balancing reserves, we evaluate the impact of PV on system reliability through the calculation of the standard reliability metric called Control Performance Standard 2 (CPS2). Broadly speaking, the CPS2 score measures the percentage of 10-min periods in which a balancing area is able to balance supply and demand within a specific threshold. Compliance with the North American Electric Reliability Corporation (NERC) reliability standards requires that the CPS2 score must exceed 90% (i.e., the balancing area must maintain adequate balance for 90% of the 10-min periods). The combination of representing DA forecast errors in the DA commitments, using 1-min PV data to simulate RT balancing, and estimates of reliability performance through the CPS2 metric, all factors that are important to operating systems with increasing amounts of PV, makes this study unique in its scope.
We analyze the impact of distributed and utility-scale PV on the APS system based on projected conventional generation, load, and wind and PV resources in 2027. Two PV deployment levels are considered: low PV is based on the PV that APS includes in its 2012 Integrated Resource Plan (IRP) base case, and high PV is based on the PV penetration that APS includes in the expanded renewables case of the IRP. The low-PV case includes sufficient PV to meet 8.8% of the annual energy, and the high-PV case includes enough PV to meet 17.0% of the annual energy (prior to any curtailment of renewables). Both cases also consider wind penetration of 4.9% of annual energy. Based on existing practices at APS five of the eight coal plants are treated as must-run units that can dispatch between minimum and maximum generation, but they cannot be turned off. Similarly, nuclear units are always operated at full nameplate capacity. We find that the combination of must-run generation, inflexible nuclear operations, and large amounts of solar in the high-PV case leads to severe operational challenges during low-load and high solar periods under the assumption that the utility cannot trade power with neighboring utilities. For a high-PV case to be practical, some solution to these challenges will be necessary. We included a “flexible nuclear” case as one option for introducing flexibility during low-load and high solar periods. The impacts of this level of PV deployment under the assumption of constant nuclear operation in the low-PV and high-PV cases and the alternative flexible nuclear operation in the high-PV case are summarized in Table ES-1.
The assumption of flexible nuclear operation in the high-PV flexible nuclear case (where all the nuclear units can operate below maximum output and can provide reserves) decreases the integration cost and greatly reduces the need to curtail renewables from almost 18% down to 3.4% of available renewables.
The addition of PV increases the variability and uncertainty between HA scheduling and RT operations. Additional balancing reserves are added in both the up and down direction to manage this uncertainty and variability. The peak and average requirement for balancing reserves in the up direction without PV, with low PV, and with high PV are summarized in Table ES-1, along with the estimated integration costs for low PV, high PV with constant nuclear operation, and high PV with flexible nuclear operation. The total integration cost is primarily due to the cost of holding resources in reserve during the HA scheduling that can then be deployed in RT to manage remaining uncertainty and variability (balancing reserve cost). The remaining portion of the costs (DA forecast error cost) is from redispatch of online generation, changes in UC within the operating day for peaking units, and imperfect UC decisions for other units based on imperfect forecasts between the DA and HA scheduling.
On the basis of the RT simulations of minute-by-minute deviations from the HA schedule, we find that the balancing reserves based on the 2.5th to 97.5th percentile of deviations are sufficient to achieve a CPS2 score that exceeds NERC minimum standards of a CPS2 score of 90% (Table ES-1), though none of the cases achieve APS’s current practice of aiming to maintain a 99% CPS2 score. The decrease in the CPS2 score, particularly in the high-PV scenario, indicates there is some degradation of CPS2 performance when balancing reserves requirements are based on the 2.5th to 97.5th percentile of deviations, an issue we address through sensitivity studies.
We conduct an extensive sensitivity analysis of system cost and reliability, using the high-PV (Flex. Nucl.) scenario as a benchmark, under different assumptions about balancing reserves, system flexibility, fuel prices, and forecasting errors. For these sensitivities we find integration costs vary within the range of $1.0 to $4.4/MWh-PV (Figure ES-1). The majority of the integration cost is due to an increase in the cost of balancing reserves held during HA scheduling, whereas DA forecast errors continue to be a smaller contributor to integration costs. Figure ES-1 shows that changes in fuel prices and forecast assumptions for wind and load do have an effect on integration costs, but the impacts are less pronounced compared to those for the other sensitivities, which are discussed in more detail below.
In the sensitivities related to balancing reserves, we examine two options for increasing the CPS2 performance with high PV penetration: (1) increase the amount of balancing reserve held in the HA or (2) increase the maximum rate of deployment (change from our initial assumption of full deployment in 10 min to full deployment in 5 min). Either option increases the CPS2 score to more than 95%, but both also increase the integration costs. There is clearly a trade-off between integration costs and the utility’s reliability level (Figure ES-2); the proper balance between the two will depend on the priorities of the utility.
On the basis of the sensitivities related to flexibility, we find that system flexibility is essential for minimizing integration costs. Flexibility is particularly important in this analysis because we assume that the utility absorbs all generation within its service territory (i.e., we assume that there are no opportunities to trade with neighboring utilities). In addition to the comparison of constant and flexible nuclear operations with high PV mentioned earlier, we show the impact of reduced system flexibility by reducing the capabilities of thermal generators to ramp from one hour to the next and minimizing curtailment of renewable energy. Renewables curtailment can be reduced to less than 1% by artificially introducing a large penalty for renewable curtailment into the UC/ED model. However, minimizing this curtailment changes the dispatch of thermal units and also results in an increase in the integration cost (Figure ES-1). Lower ramp rates of thermal units also increase the integration costs.
To further highlight the importance of flexibility, we constructed a worst-case scenario in which limits to flexibility (including constant nuclear output, low ramp rates for other thermal generators, and penalties on renewables curtailment) and increased balancing reserve requirements were simultaneously assumed. In this worst case, the integration cost increases to $9.6/MWh and renewables curtailment exceeds 10% of available renewable generation (despite the penalties for renewable curtailment), and it also becomes challenging to meet the balancing reserve requirements with frequent occurrences of reserve shortfalls. This case is unrealistic given the combination of several conservative assumptions regarding system flexibility and the actual ability of the utility to trade with other utilities. However, the results do highlight the importance of finding buyers for excess power during times with high PV production or the need to increase flexibility from existing thermal power plants or other resources.
Limitations and Future Work
This case study does have a number of limitations that should be considered in the examination of these conclusions. Simulations of future deployments of variable generation are always challenging because of the need to synthesize high-time resolution production and forecasts that reflect geographic diversity. The methods used in this study to synthesize 1-min PV production data are based on existing methods applied for wind power, but not otherwise applied to power from solar PV. Anomalies in the data may overstate short-term PV variability. Furthermore, the thermal plant characteristics were developed from publicly available datasets and likely differ from true plant performance.
There are also limitations based on the simplifications used to model actual strategies and procedures used by utilities. For instance, the model does not fully reflect the flexibility to make recommitment decisions at any time between DA and HA scheduling. This may understate the flexibility available from the thermal generators. On the other hand, the revised dispatch and commitments in the HA were made with a single run of the optimization model over a 24-hr period based on the assumption that the HA forecasts are known. This modeling assumption may understate the costs of managing variability and uncertainty. We also did not include transmission constraints within the utility footprint.
Another important limitation of this case study is that we did not model exchange with the broader Western Electricity Coordinating Council (WECC) area outside of the utility boundaries, an assumption that is likely to overstate the challenges with low-net-load periods, particularly if neighboring utilities are adding less renewables to their systems. These impacts could also be mitigated with the inclusion of APS in any of the proposed Energy Imbalance Markets (EIM) currently being examined in the Western Interconnection. This new sub-hourly real time energy market would provide access to more flexibility that might prevent significant amounts of renewables curtailment, and also reduce the costs associated with sub-hourly solar variability. In addition, the NERC Board of Trustees recently approved a new balancing standard that would replace the CPS2 with an alternative reliability metric. This new balancing standard may also alter the balancing reserve requirements in scenarios with high PV. Last, we did not remove any thermal generation capacity from the low-PV portfolio when the amount of PV was increased in the high-PV scenario, although PV is likely to have some capacity value. If refined estimates of the impact of PV on system operations, cost, and reliability are required, these limitations should be addressed.
Overall, we only investigated a small subset of the potential future sources of system flexibility in this study. Additional sources for flexibility include demand side programs that shift load into low net-load periods or energy storage. Demand resources can also contribute to the provision of operating reserves. Another important source of flexibility not considered in this study is the potential provision of reserves from wind and utility-scale solar resources. As long as the utility can send a control signal to these resources they can provide operating reserves, particularly in the down direction. Contributions to system flexibility from demand resources and renewables may very well be easier to implement than flexible nuclear operations as investigated in this analysis. In future work, we therefore recommend exploring these additional sources of system flexibility.
In this study, we did not consider dynamic balancing reserves that change based on expected weather conditions. In particular, if probabilistic forecasts are available they could be used to ensure that there was low probability of clouds before reserves would be reduced. This would potentially decrease the balancing reserve requirements for PV and lower integration costs accordingly. Factoring weather forecasts into the dynamic estimation of balancing reserves represents an interesting direction for future work. Finally, the potential use of stochastic scheduling strategies that make direct use of probabilistic forecasts to commit and dispatch system resources is also an area of active research. In this case, the mathematical objective of the scheduling problem is to minimize operating cost over a range of forecast scenarios for renewable generation. Hence, balancing reserves are scheduled implicitly rather than imposed as explicit reserve requirements. The potential benefits of stochastic scheduling for system cost and reliability is also an area we want to explore in future work.