This report contains the written footprint of a Sandia-hosted workshop held in Albuquerque, New Mexico, June 22-23, 2016 on “Complex Systems Models and Their Applications: Towards a New Science of Verification, Validation and Uncertainty Quantification,” as well as of pre-work that fed into the workshop. The workshop’s intent was to explore and begin articulating research opportunities at the intersection between two important Sandia communities: the complex systems (CS) modeling community, and the verification, validation and uncertainty quantification (VVUQ) community The overarching research opportunity (and challenge) that we ultimately hope to address is: how can we quantify the credibility of knowledge gained from complex systems models, knowledge that is often incomplete and interim, but will nonetheless be used, sometimes in real-time, by decision makers?
From 2010 through the first half of 2015, the installed capacity of solar photovoltaics (PV) connected to the U.S. distribution system increased sixfold, from approximately 1.8 GW to more than 11 GW. This accounts for over half of the approximate total U.S. solar installations of 20 GW. Distributed generation from PV (DGPV) is expected to comprise 50%–60% of total U.S. PV capacity through at least 2020. The rapid deployment of high penetrations of DGPV into the distribution system has both highlighted challenges and demonstrated many successful examples of integrating higher penetration levels than previously thought possible. In this report, we analyze challenges, solutions, and research needs in the context of DGPV deployment to date and the much higher levels of integration that are expected with the achievement of the U.S. Department of Energy’s SunShot targets.
The research presented in this report compares several real - time control strategies for the power output of a large number of PV distributed throughout a large distribution feeder circuit. Both real and reactive power controls are considered with the goal of minimizing network over - voltage violations caused by large amounts of PV generation. Several control strategies are considered under various assumptions regarding the existence and latency of a communication network. The control parameters are adjusted to maximize the effectiveness of each control. The controls are then compared based on their ability to achieve multiple objectiv es. These objectives include minimizing the total number of voltage violations , minimizing the total amount of PV energy curtailed or reactive power generated, and maximizing the fairness of any control action among all PV systems . The controls are simulat ed on the OpenDSS platform using time series load and spatially - distributed irradiance data.
In this project, an integrated solution to measuring and collecting solar variability data called the solar variability datalogger (SVD) was developed, tested, and the value of its data to distribution grid integration studies was demonstrated. This work addressed the problem that high-frequency solar variability is rarely measured – due to the high cost and complex installation of existing solar irradiance measuring pyranometers – but is critical to the accurate determination of the impact of photovoltaics to electric grid operation. For example, up to a 300% difference in distribution grid voltage regulator tap change operations (a measure of the impact of PV) [1] has been observed due solely to different solar variability profiles.
To address the lack of knowledge of local solar variability, we have developed, deployed, and demonstrated the value of data collected from a low-cost solar variability sensor. While most currently used solar irradiance sensors are expensive pyranometers with high accuracy (relevant for annual energy estimates), low-cost sensors display similar precision (relevant for solar variability) as high-cost pyranometers, even if they are not as accurate. In this work, we list variability sensor requirements, describe testing of various low-cost sensor components, present a validation of an alpha prototype, and show how the variability sensor collected data can be used for grid integration studies. The variability sensor will enable a greater understanding of local solar variability, which will reduce developer and utility uncertainty about the impact of solar photovoltaic installations and thus will encourage greater penetrations of solar energy.
To address the lack of knowledge of local solar variability, we have developed, deployed, and demonstrated the value of data collected from a low-cost solar variability sensor. While most currently used solar irradiance sensors are expensive pyranometers with high accuracy (relevant for annual energy estimates), low-cost sensors display similar precision (relevant for solar variability) as high-cost pyranometers, even if they are not as accurate. In this work, we list variability sensor requirements, describe testing of various low-cost sensor components, present a validation of an alpha prototype, and show how the variability sensor collected data can be used for grid integration studies. The variability sensor will enable a greater understanding of local solar variability, which will reduce developer and utility uncertainty about the impact of solar photovoltaic installations and thus will encourage greater penetrations of solar energy.
Increasing the penetration of distributed renewable sources, including photovoltaic (PV) sources, poses technical challenges for grid management. The grid has been optimized over decades to rely upon large centralized power plants with well-established feedback controls, but now non-dispatchable, renewable sources are displacing these controllable generators. This one-year study was funded by the Department of Energy (DOE) SunShot program and is intended to better utilize those variable resources by providing electric utilities with the tools to implement frequency regulation and primary frequency reserves using aggregated renewable resources, known as a virtual power plant. The goal is to eventually enable the integration of 100s of Gigawatts into US power systems.
Increasing number s of PV on distribution systems are creating more grid impacts , but it also provides more opportunities for measurement, sensing, and control of the grid in a distributed fashion. This report demonstrates three software tools for characterizing and controlling distribution feeders by utilizing large numbers of highly distributed current, voltage , and irradiance sensors. Instructions and a user manual is presented for each tool. First, the tool for distribution system secondary circuit parameter estimation is presented. This tool allows studying distribution system parameter estimation accuracy with user-selected active power, reactive power, and voltage measurements and measurement error levels. Second, the tool for multi-objective inverter control is shown. Various PV inverter control strategies can be selected to objectively compare their impact on the feeder. Third, the tool for energy storage for PV ramp rate smoothing is presented. The tool allows the user to select different storage characteristics (power and energy ratings) and control types (local vs. centralized) to study the tradeoffs between state-of-charge (SOC) management and the amount of ramp rate smoothing.
Albedo and diffuse plane of array (DPOA) measurements are used in addition to standard global horizontal irradiance (GHI), direct normal irradiance (DNI), diffuse horizontal irradiance (DNI), and plane of array irradiance (POA) measurements to determine the impact of albedo on transposition model performance. Albedo measurements averaged 0.214. Daily albedo values ranged from 0.148 to 0.236 and were found to be correlated to daily relative humidity. DPOA measurements were compared to calculated DPOA values (from POA and DNI), and helped identify a suspected deviation from due south in the azimuth of the POA measurement. Since the measured albedo average was close to the typical fixed $^{albedo = 0.2}$ assumption, little difference was seen between using measured and fixed albedo (~0.15% differences in mean bias difference (MBD) and root mean squared difference (RMSD)). However, evaluation of transposition models at other fixed albedos showed an albedo change of 0.1 is found to increase total modeled insolation by approximately 1%. Thus, for locations with different ground surfaces (e.g., persistent snow cover of black surfaces), the impact of using measured albedo instead of the fixed $^{albedo = 0.2}$ assumption may be greater. Measurement deviations resulted in up to 2% changes in MBD and RIVISD when switching between interrelated measurements (e.g., GHI and DHI as inputs to transposition models versus DNI and DHI as inputs). Variation among transposition models was also up to 2% MBD and RMSD. Thus, for this data set, measurement deviation and transposition model selection are found to have more impact than using measured albedo instead fixed albedo.
We report an evaluation of the accuracy of combinations of models that estimate plane-of-array (POA) irradiance from measured global horizontal irradiance (GHI). This estimation involves two steps: 1) decomposition of GHI into direct and diffuse horizontal components and 2) transposition of direct and diffuse horizontal irradiance (DHI) to POA irradiance. Measured GHI and coincident measured POA irradiance from a variety of climates within the United States were used to evaluate combinations of decomposition and transposition models. A few locations also had DHI measurements, allowing for decoupled analysis of either the decomposition or the transposition models alone. Results suggest that decomposition models had mean bias differences (modeled versus measured) that vary with climate. Transposition model mean bias differences depended more on the model than the location. When only GHI measurements were available and combinations of decomposition and transposition models were considered, the smallest mean bias differences were typically found for combinations which included the Hay/Davies transposition model.
Accurately representing the local solar variability at timescales relevant to distribution grid operations (30-s and shorter) is essential to modeling the impact of solar photovoltaics (PV) on distribution feeders. Due to a lack of available high-frequency solar data, some distribution grid studies have used synthetically-created PV variability or measured PV variability from a different location than their study location. In this work, we show the importance of using accurate solar PV variability inputs in distribution studies. Using high-frequency solar irradiance data from 10 locations in the United States, we compare the ramp rate distributions at the different locations, use a quantitative metric to describe the solar variability at each location, and run distribution simulations using representative 1-week samples from each location to demonstrate the impact of locational solar variability on the number of voltage regulator tap change operations. Results show more than a factor of 3 difference in the number of tap change operations between different PV power variability samples based on irradiance from the different locations. Errors in simulated number of tap changes of up to -70% were found when using low-frequency (e.g., 15-min) solar variability.
Four PV power plant variability simulation methods - no-smoothing, time average, Marcos, and the wavelet variability model (WVM) - were compared to measured data from a 19MW PV power plant to test the relative accuracy of each method. Errors (simulated vs. measured) were quantified using five application-specific metrics: the largest down ramps, the largest up ramps, the mean absolute error in matching the cumulative distribution of large ramps, the total energy contained in down ramps over the entire period considered, total energy in down ramps on the worst day. These errors we evaluated over timescales ranging from 1-second to 1-hour and over plant sizes of 1 to 14MW and the total plant size of 19MWs to determine trends in model errors as a function of timescale and plant size. Overall, the WVM was found to most often have the smallest errors. The Marcos method also often had small errors, including having the smallest errors of all methods at small PV plant sizes (1 to 7MWs). The no-smoothing method had large errors and should not be used. The time average method was an improvement over the no-smoothing method, but generally has larger errors than the WVM and Marcos methods.