Papers

Click on a paper to see more information.

Bunn, P.T., Boeman, L.J., Lorenzo, A.T. and Raub, J., Frontiers in Energy Research, 12, p.1434019, The expected solar performance and ramp rate tool: a decision-making tool for planning prospective photovoltaic systems (2024). journal link

The Expected Solar Performance and Ramp Rate tool (ESPRR) is an open-source interactive web-based application that reliably calculates ramp rate (RR) statistics and an expected power generation time series for prospective photovoltaic (PV) systems. Users create PV systems by defining site parameters. ESPRR uses those parameters with irradiance data from the National Solar Radiation Database (NSRDB) to create a time series of power output from which RR statistics are calculated. This study rigorously evaluates ESPRR's performance using 5 years of measured power output from a fleet of utility-scale systems and finds that ESPRR calculates stress-case RRs within an error of 0.05 MW/min and 0.42 MW/min for the worst-case RRs. We evaluate the expected AC power output in clear-sky conditions and find an NRMSE of less than 10% and an NMBE of less than 6% for the fleet's largest system. The NRMSE is 10%-15% of system capacity for non-clear-sky conditions, and the NMBE is about zero. The evaluation shows that ESPRR can estimate PV output and RRs that are representative of operational systems, meaning users can use the results from ESPRR in the decision-making process for designing new systems or when adding systems to an existing fleet. Since only system parameters are required to site a proposed system anywhere on a map, users can site and reposition a fleet of PV systems in a way that reduces significant RRs. As the grid-tied PV capacity continues to increase, the mitigation of significant RRs grows in importance. ESPRR can help developers and utilities create geographically diverse fleets of PV systems that will promote grid reliability and avoid significant RRs. ESPRR source code is available at https://github.com/UARENForecasting/ESPRR.

Bunn et al. Journal of Renewable and Sustainable Energy 12 053702, Using GEOS-5 forecast products to represent aerosol optical depth in operational day-ahead solar irradiance forecasts for the southwest United States (2020). journal link

This study aims to improve operational day-ahead direct normal irradiance (DNI) forecasts in clear-sky conditions using the Weather and Research Forecasting model. To create three different forecasting methods targeting the direct effect of aerosols on radiation, we use three different types of aerosol optical depth (AOD) data: (1) the Tegen aerosol climatology, (2) the persistence of measured AERONET AOD, and (3) the Goddard Earth Observing System model version 5 (GEOS-5) gridded forecasts of AOD. We evaluate each method at the Solana Generating Station, a concentrating solar power plant near Gila Bend, Arizona, and the University of Arizona, Tucson. We perform a retrospective DNI forecast analysis and find that including GEOS-5 forecast AOD improved the DNI forecast compared to using an aerosol climatology at both locations. At Tucson, where AOD is measured, we find that the persistence of measured AOD gives the best DNI forecast. However, the accuracy of that measured AOD reduces when translating it 225 km to Solana to forecast DNI 48 hours later. We then include the GEOS-5 AOD forecasts in one member of an operational forecast system and evaluate it against the other ensemble members that use the aerosol climatology. In clear-sky conditions, including GEOS-5 forecast AOD instead of the Tegen aerosol climatology, reduces the DNI forecast root mean square error by 27% at Solana. We found no significant differences during all-sky conditions because the relatively poor performance during cloudy conditions outweighs the improvements made in clear-sky conditions.

Harty et al. Solar Energy 185 270-282, Intra-hour Cloud Index Forecasting with Data Assimilaiton (2019). journal link presentation

We introduce a computational framework to forecast cloud index (CI) fields for up to one hour on a spatial domain that covers a city. Such intra-hour CI forecasts are important to produce solar power forecasts of utility scale solar power and distributed rooftop solar. Our method combines a 2D advection model with cloud motion vectors (CMVs) derived from a mesoscale numerical weather prediction (NWP) model and sparse optical flow acting on successive, geostationary satellite images. We use ensemble data assimilation to combine these sources of cloud motion information based on the uncertainty of each data source. Our technique produces forecasts that have similar or lower root mean square error than reference techniques that use only optical flow, NWP CMV fields, or persistence. We describe how the method operates on three representative case studies and present results from 39 cloudy days.

Holmgren et al. 44th IEEE PVSC Proceedings, A Comparison of PV Power Forecasts Using PVLib-Python (2017). poster

In this paper, we use the PVLib-Python forecasting tool to create hourly average PV power forecasts for a fleet of utility scale power plants and we compare the forecasts to observed plant generation. As an example of the utility of PVLib-Python for creating benchmark forecasts, we compare the forecasts derived from NOAA weather models (GFS, NAM, and RAP) with forecasts derived from a model run by U. Arizona.

A. T. Lorenzo Doctoral Dissertation Short-Term Irradiance Forecasting Using an Irradiance Monitoring Network, Satellite Imagery, and Data Assimilation (2017). defense

This dissertation will explore techniques to forecast irradiance that make use of data from a network of sensors deployed throughout Tucson, AZ. The design and deployment of inexpensive sensors used in the network will be described. We will present a forecasting technique that uses data from the sensor network and outperforms a reference persistence forecast for one minute to two hours in the fu- ture. We will analyze the errors of this technique in depth and suggest ways to interpret these errors. Then, we will describe a data assimilation technique, optimal interpolation, that combines estimates of irradiance derived from satellite images with data from the sensor network to improve the satellite estimates. These im- proved satellite estimates form the base of future work that will explore generating forecasts while continuously assimilating new data.

Lorenzo et al. Solar Energy 144 466-474, Optimal Interpolation of Satellite and Ground Data for Irradiance Nowcasting at City Scales (2017). journal link poster

We use a Bayesian method, optimal interpolation, to improve satellite derived irradiance estimates at city-scales using ground sensor data. Optimal interpolation requires error covariances in the satellite estimates and ground data, which define how information from the sensor locations is distributed across a large area. We describe three methods to choose such covariances, including a covariance parameterization that depends on the relative cloudiness between locations. Results are computed with ground data from 22 sensors over a 75x80km area centered on Tucson, AZ, using two satellite derived irradiance models. The improvements in standard error metrics for both satellite models indicate that our approach is applicable to additional satellite derived irradiance models. We also show that optimal interpolation can nearly eliminate mean bias error and improve the root mean squared error by 50%.

Holmgren et al. 43th IEEE PVSC Proceedings, An Open Source Solar Power Forecasting Tool Using PVLIB-Python (2016). poster

We describe an open-source PV power forecasting tool based on the PVLIB-Python library. The tool allows users to easily retrieve standardized weather forecast data relevant to PV power modeling from NOAA/NCEP/NWS models including the GFS, NAM, RAP, HRRR, and the NDFD. A PV power forecast can then be obtained using the weather data as inputs to the comprehensive modeling capabilities of PVLIB-Python. Standardized, open source, reference implementations of forecast methods using publicly available data may help advance the state- of-the-art of solar power forecasting.

Lorenzo et al. 43th IEEE PVSC Proceedings, Optimal Interpolation of Satellite Derived Irradiance and Ground Data (2016).

We describe how Bayesian data assimilation can be used to improve nowcasts of irradiance over small, city-scale, spatial areas. Specifically, we use optimal interpolation (OI) to improve satellite derived estimates of global horizontal irradiance (GHI) using ground truth data that was collected sparsely over Tucson, AZ. Our results show that the local data indeed improves the satellite derived estimates of GHI. A key to success with OI in this context is to prescribe correlations based on cloudiness, rather than spatially. OI can be used with a variety of data, e.g., rooftop photovoltaic production data or irradiance data, as well as with several different satellite derived irradiance models.

Lorenzo et al. Solar Energy 122 1158-1169, Irradiance Forecasts based on an Irradiance Monitoring Network, Cloud Motion, and Spatial Averaging (2015). journal link

We describe and evaluate forecasts of solar irradiance using real-time measurements from a network of irradiance sensors. A forecast method using cloud motion vectors obtained from a numerical weather model shows significant skill over a standard persistence model for forecast horizons from 1 min to over 2 h, although the skill metric may be misleading. To explain this finding, we define and compare several different persistence methods, including persistence methods informed by an instantaneous spatial average of irradiance sensor output and persistence forecasts informed by a time-average of recent irradiance measurements. We show that spatial- or temporal-averaging reduces the forecast RMS errors primarily because these forecasts are smoother (have smaller variance). We use a Taylor diagram, which shows correlation, RMSE, and variance, to more accurately compare several different types of forecasts. Using this diagram, we show that forecasts using the network of sensors have meaningful skill up to 30 min time horizons after which the skill is primarily due to smoothing.

Holmgren et al. UA-SVERI Variability Analysis (2015).

In this study, We collected and analyzed 10-second data for load, wind generation, and solar photovoltaic (PV) generation. Based on this information, several techniques were used to study variability of load, renewables, and generation. This report presents the effects of solar PV and wind resources on net load ramps in the desert southwest from June through November 2014.

Cormode et al. 2014 EU PVSEC Proceedings, Observed Fluctuations in Output From a Regional Fleet of PV Power Plants Used to Compute Hourly Schedules of Spinning Reserve Requirements (2014).

We examine fluctuations in power from an 80 MW fleet of utility scale power plants deployed around Tucson, Arizona, and a 500MW fleet deployed throughout Arizona and New Mexico. We observe that individual plants exhibit frequent rapid changes in power, greater than 50% of nameplate capacity in less than one minute. The aggregate fleet generally has slower ramps. Local utilities face the challenge of addressing this variability in a cost effective manner. We present a method to use historical data to estimate appropriate spinning reserves for variability mitigation for each hour of the day for different seasons. We contrast the results for recommended reserves when calculated on a fleet wide basis or when summed from a plant by plant basis.

Holmgren et al. 40th IEEE PVSC Proceedings, An Operational, Real-Time Forecasting System for 250 MW of PV Power Using NWP, Satellite, and DG Production Data (2014).

We developed a real-time PV power forecasting system for Tucson Electric Power using a combination of high- resolution numerical weather prediction, satellite imagery, distributed generation (DG) production data, and irradiance sensors. The system provides forecasts with 10 second resolution for the first 30 minutes and 3 minute resolution out to 3 days. Forecasts out to 30 minutes are updated every 60 seconds based on new data from DG installations and irradiance sensors.

Lorenzo et al. 40th IEEE PVSC Proceedings, Short-Term PV Power Forecasts Based on a Real-Time Irradiance Monitoring Network (2014).

We built an irradiance sensor network that we are now using to make operational, real-time, intra-hour forecasts of solar power at key locations. We developed reliable irradiance sensor hardware platforms to enable these sensor network fore- casts. Using 19 of the 55 irradiance sensors we have throughout Tucson, we make retrospective forecasts of 26 days in April and evaluate their performance. We find that that our network forecasts outperform a persistence model for 1 to 28 minute time horizons as measured by the root mean squared error. The sensor hardware, our network forecasting method, error statistics, and future improvements to our forecasts are discussed.

Cormode et al. 40th IEEE PVSC Proceedings, The Economic Value of Forecasts for Optimal Curtailment Strategies to Comply with Ramp Rate Rules (2014).

We present a method to calculate the economic value of forecasts, based on the use of forecasts to optimize curtailment strategies in scenarios with a ramp rate rule. We consider how and when to limit PV power output in order to comply with a ramp rate rule to avoid penalties, but also calculate how curtailment will reduce revenue from energy yields. This framework provides a way to assess the value of forecasts.

Lonij et al. Solar Energy 97 58-66, Intra-hour forecasts of solar power production using measurements from a network of irradiance sensors (2013).

We report a new method to forecast power output from photovoltaic (PV) systems under cloudy skies that uses measurements from ground-based irradiance sensors as an input. This work describes an implementation of this forecasting method in the Tucson, AZ region where we use 80 residential rooftop PV systems distributed over a 50x50 km area as irradiance sensors. We report RMS and mean bias errors for a one year period of operation and compare our results to the persistence model as well as forecasts from other authors. We also present a general framework to model station-pair correlations of intermittency due to clouds that reproduces the observations in this work as well as those of other authors. Our framework is able to describe the RMS errors of velocimetry based forecasting meth- ods over three orders of magnitude in the forecast horizon (from 30 s to 6 h). Finally, we use this framework to recommend optimal locations of irradiance sensors in future implementations of our forecasting method.

Cormode et al. 39th IEEE PVSC Proceedings, Comparing Ramp Rates from Large and Small PV systems, and Selection of Batteries for Ramp Rate Control (2013).

We compare the AC power fluctuations from a 1.6 MW and a 2 kW photovoltaic (PV) system. Both of these PV generating stations exhibit fluctuations exceeding 50% of their rated capacity in under 10 seconds. The smaller system can fluctuate more rapidly, exhibiting 50% dropouts in 3 seconds. Although the MW-scale system covers 4000 times as much ground area, the bandwidth of the fluctuations is remarkably similar. We explore explanations for this observation, and we discuss the impact of this on battery sizing.

Jayadevan et al. ASES World Renewable Energy Forum proceedings, Forecasting Solar Power Intermittency using Ground-Based Cloud Imaging (2012).

 We report progress towards developing methods to forecast solar-power intermittency due to clouds using analysis of digital images taken with a ground-based, suntracking camera. We show preliminary results of blockmotion estimation analysis applied to a sequence of sky images recorded in Tucson, Arizona. In addition, we discuss statistics of ramp rates and duration of cloudinduced intermittency based on the analysis of one year of photovoltaic power output data measured at one-second intervals for a 2-kW system.

Our work was also featured in a 2014 SEPA report on solar power forecasting. See the free executive summary: