A New Evaluation Metric Reflecting the Lead-Lag Scenarios in Wind Power Forecasting

• Type: Article in International Conference Proceedings Book
• Date: 2021

Authors

Yuan, JL (Author) Other; Wang, F (Author) Other; Shafie khah, M (Author) Other; Zhen, Z (Author) Other; Catalao, JPS (Author) FEUP

Conference proceedings International

Title: IEEE Industry Applications Society Annual Meeting, IAS 2021, Vancouver, BC, Canada, October 10-14, 2021

Pages: 1-7

Indexing

ISI Web of Knowledge - 0 Citations

Scopus - 0 Citations

Other information

Authenticus ID: P-00W-30B

DOI: 10.1109/ias48185.2021.9677172

Abstract (EN): Wind power forecast evaluation matters greatly as wind power has an ever-increasing proportion in the power system. Generally speaking the forecasting result can be divided into lead-lag scenarios and common scenarios which depends on whether the wind process is predicted on time. During the lead-lag scenarios the errors usually change from large positive numbers to negative ones (or the opposite), especially in the both ends of the period. Compared with the common scenarios in the same value of root mean square error (RMSE), large changes in errors from positive to negative in a short time can cost nearly two times of spinning reserve but get the same assessment score. For power system the two scenarios should be evaluated differently, however, few metrics in the evaluation can indicate the lead-lag scenarios in that they dispose the errors ignoring the signs or time continuity of the errors, or analysis the errors in a macro-scale sight like 24 hours horizon scale. This paper proposes a new metric based on RMSE which detects the changes of signs of errors in a process of moving average. Except for normal advantages like objectivity, adaptability, unity, symmetry and stability, the new metric has the ability to reflect both the lead-lag scenarios and common scenarios. The new metric can be used in the evaluation of wind and solar power, load, price, demand response forecasting and the process of neural network parameter training.

Language: English

Type (Professor's evaluation): Scientific

No. of pages: 7