Metrics Development for Evaluating the Accuracy of Solar Power Forecasting

- Indicates paper has been withdrawn from meeting
- Indicates an Award Winner
Thursday, 6 February 2014: 4:30 PM
Room C114 (The Georgia World Congress Center )
Jie Zhang, National Renewable Energy Laboratory, Golden, CO; and B. M. Hodge, A. Florita, S. Lu, H. Hamman, and V. Banunarayanan

Solar power forecasting is a challenging task, since solar power generation is present in both the transmission and the distribution sides of the grid. Forecast inaccuracies can result in substantial economic losses, as electric grid operators must continuously balance supply and demand to maintain the reliability of the power grid. This work develops a suite of generally applicable and value-based metrics for deterministic solar forecasting for a comprehensive set of scenarios (different time horizons, geographic locations, applications, etc.).

The proposed solar forecasting metrics can be broadly divided into 5 categories, which are (i) statistical metrics, including Pearson's correlation coefficient, (normalized) root mean squared error, mean absolute error, mean absolute percentage error, mean, and Kolmogorov–Smirnov test Integral (KSI); (ii) variability estimation metrics, including different time and geographic scales, and distributions of forecast errors; (iii) uncertainty quantification and propagation metrics, including standard deviation and information entropy of forecast errors; (iv) ramping characterization metrics, including ramp hit rate, swinging door algorithm signal compassion, and heat maps; and (v) economic and reliability metrics, including regulating reserve requirement and flexibility reserve requirement represented by 95th and 70th percentiles of forecast errors, respectively.

The performance of the proposed metrics is evaluated using the actual and forecast solar power data from the Western Wind and Solar Integration Study. Different spatial and temporal scales of forecast errors are compared based on the metrics. In addition, a comprehensive framework is developed to analyze the sensitivity of the proposed metrics to three different types of solar forecasting improvements through a design of experiments methodology, in conjunction with response surface and sensitivity analysis methods. The results show that the developed metrics can efficiently evaluate the quality of solar forecasts, and assess the economic and reliability impact of improved solar forecasting.