Day-Ahead Wind Power Forecasting in Poland Based on Numerical Weather Prediction

Bogdan Bochenek,Adam Jaczewski,Gabriel Stachura,Mariusz Figurski,Piotr Sekuła,Marcin Wdowikowski,Jakub Jurasz,Tomasz Strzyżewski

doi:10.3390/en14082164

Abstract

The role of renewable energy sources in the Polish power system is growing. The highest share of installed capacity goes to wind and solar energy. Both sources are characterized by high variability of their power output and very low dispatchability. Taking into account the nature of the power system, it is, therefore, imperative to predict their future energy generation to economically schedule the use of conventional generators. Considering the above, this paper examines the possibility to predict day-ahead wind power based on different machine learning methods not for a specific wind farm but at national level. A numerical weather prediction model used operationally in the Institute of Meteorology and Water Management–National Research Institute in Poland and hourly data of recorded wind power generation in Poland were used for forecasting models creation and testing. With the best method, the Extreme Gradient Boosting, and two years of training (2018–2019), the day-ahead, hourly wind power generation in Poland in 2020 was predicted with 26.7% mean absolute percentage error and 4.5% root mean square error accuracy. Seasonal and daily differences in predicted error were found, showing high mean absolute percentage error in summer and during daytime.

Highlights

Solar and wind energy are slowly becoming significant players in the Polish energy system
We proposed a day-ahead wind power forecasting system based on multiple machine learning methods, an accurate limited area numerical weather prediction (NWP) model and hourly data time series of power generated from wind turbines that can be applied on the country level in Poland and possibly in other countries with similar climatological conditions
We have shown that it is possible to predict a day-ahead wind energy production with high accuracy not necessarily by applying a physical or statistical model for every wind farm but by building machine learning based model with wind speed forecasts over the whole country

Summary

Introduction

Solar and wind energy are slowly becoming significant players in the Polish energy system This is visible from a perspective of their rated capacity, which, as of the end of 2020, was 3420 MW [1] and 6347 MW [2], respectively, for solar and wind. Such capacities correspond to 19.8% of the total installed capacity [3], the share of solar and wind energy in satisfying the national electrical energy demand remains relatively low, at 1.4% and 9.2%, respectively. The complexity of the problem lies in an intrinsically variable and nondispatchable nature of solar and wind generators and in limited flexibility of the available infrastructure

Objectives

Methods

Results

Discussion

Conclusion