Wind power generation forecast by coupling numerical weather prediction model and gradient boosting machines in Yahyalı wind power plant

Abstract

Forecasting of the wind speed and power generation for a wind farm has always been quite challenging and has importance in terms of balancing the electricity grid and preventing energy imbalance penalties. This study focuses on creating a hybrid model that uses both numerical weather prediction model and gradient boosting machines (GBM) for wind power generation forecast. Weather Research and Forecasting (WRF) model with a low spatial resolution is used to increase temporal resolutions of the computed new or existing variables whereas GBM is used for downscaling purposes. The results of the hybrid model have been compared with the outputs of a stand-alone WRF which is well configured in terms of physical schemes and has a high spatial resolution for Yahyalı wind farm over a complex terrain located in Turkey. Consequently, the superiority of the hybrid model in terms of both performance indicators and computational expense in detail is shown.

A comprehensive country-based day-ahead wind power generation forecast model by coupling numerical weather prediction data and CatBoost with feature selection methods for Turkey

“…Barque et al (2018), using the atmospheric variable forecast data obtained from the NWP model and the historical power generation data of the related power plant, improved the performance of WPGF made with gradient boosted trees by 17%. Özen et al (2021) proposed a hybrid WPG forecast by coupling the outputs of the Weather Research and Forecasting (WRF) model which uses FNL/GDAS data as input with a gradient boosting machine (GBM). In this study, the atmospheric variables extracted from the four grid points surrounding a wind farm were used to train the proposed model.…”

Short-term wind speed forecast for Urla wind power plant: A hybrid approach that couples weather research and forecasting model, weather patterns and SCADA data with comprehensive data preprocessing

“…Okumus and Dinler (2016) proposed a wind speed forecasting model for 1 hour ahead by using the combination of adaptive neuro-fuzzy inference system and feed-forward artificial neural network on three different sites and in four different months of the year 2014 and they got better results than these models’ individual forecasts. Also, Özen et al (2021), compared day-ahead wind power forecast made with a well-tuned WRF model that has the high spatial resolution for a wind power plant on a complex site with a hybrid system that was proposed in the study. By coupling low-resolution WRF model outputs with gradient boosting machine method, in a study in which it was suggested that the downscaling of physical models should be made with a machine learning model, this method outperformed the physical method both in terms of computational time and accuracy.…”

Analysis of Open-Source Data in Wind Power Prediction Modeling

“…Previous studies have found that combinations of datasets and forecasting methods can improve wind power and wind speed predictions. In [84] and [85], machine learning models were used to improve WRF wind power and wind speed forecasts. To the best of the authors' knowledge, however, prior work has not analyzed the effectiveness of combining all three aforementioned data sources using machine learning methods.…”

“…Pang et al explored the impact of other environmental variables using Shapley values and found that turbulence intensity, air density, and wind direction moderately impacted power output [86]. Özen et al included multiple WRF output variables as input to a power prediction model, however the importance of variables other than wind speeds was ambiguous due to the use of multiple wind speed variables at various heights [84]. This study further demonstrates that wind speed measurements alone are not sufficient for highly accurate wind power prediction.…”