6th International Conference Energy & Meteorology: Abstract Submission

Solar irradiance and PV forecasting for large PV power plants using numerical weather models, satellite data, and ground measurements (730)

Wiebke Herzberg 1 , Nicolas Holland 1 , Steffen Karalus 1 , Jefferson Bor 1 , Elke Lorenz 1
  1. Fraunhofer Institute for Solar Energy Systems ISE, Freiburg, Germany

Objective & Background:

In order to integrate variable photovoltaic (PV) power into the grid, reliable forecasts are of great importance. Naturally, the impact of forecasts grows with increasing power plant size. We are developing PV power forecast systems for time scales from 15 minutes up to several days ahead with a focus on large plants (>500 MW). Large plants can consist of multiple sub-plants having a diverse range of configurations (e.g. mono and bifacial modules, different tracking systems), which in itself represents an interesting challenge with regard to modelling the power output.

 

Method:

State of the art forecasting systems typically integrate numerical weather predictions (NWP) and PV power measurements. Many systems also include satellite based forecasts [1]. Our system additionally uses on-site measurements of global horizontal irradiance (GHI), which provides us with the advantage of being able to detect effects not dependent on irradiance conditions, e.g. curtailment or snow cover on the modules. At the current state, we combine persistence forecasts obtained from GHI measurements with numerical weather predictions to gain reliable forecasts for sub-hour horizons up to 48 hours ahead. For the combination we apply a linear model. The resulting combined GHI forecast is then fed into a simulation of the PV system.

 

Principal findings:

We present an evaluation of the performance of the forecasting system by means of an example power plant with a size of 49kW peak in Germany. The evaluation considers the accuracy of GHI forecasts and PV power forecasts (Figure 1). For sub-hour horizons, the persistence forecast clearly shows much lower root mean square errors (RMSE) than the NWP forecast, while for horizons of several hours, the NWP forecast shows smaller errors. Combining the two forecasts results in an improvement of accuracy for horizons up to 5 hours ahead, particularly evident at horizons around 1-2 hours.

 

Discussion & Outlook:

With RMSE values of 7.8% with respect to the installed PV power for a 15 minute horizon, 12.3% for a 3 hour horizon, and 14.7% for the day ahead, our results are currently in good agreement with similar studies [1]. For the final presentation we plan to include also an evaluation for a large plant, depending on data availability. To further improve the accuracy of the combined forecasts we plan to add satellite based forecasts as well as different NWP models to the system in the future.

 

 5c4afe1f5d1fb-pvsim_eval_combi_rel_total.png

Figure 1: Relative RMSE of persistence, NWP, and combined PV forecasts with respect to power measurements with increasing forecast horizon. The analyzed data ranges from January 2018 to June 2018.

 

  1. Wolff, B., et al.: „Comparing support vector regression for PV power forecasting to a physical modeling approach using measurement, numerical weather prediction, and cloud motion data", Solar Energy, Volume 135, October 2016, Pages 197-208