Domain-informed CNN architectures for downscaling regional wind forecasts

High-resolution wind speed forecasts are of great importance to the wind energy industry, from short-term energy forecasting and trading to longer-term resource assessment and planning. Generating high-resolution regional wind forecasts currently requires compute-intensive numerical models to downsc...

Full description

Saved in:
Bibliographic Details
Main Authors: Alexander M. Campbell, Simon C. Warder, B. Bhaskaran, Matthew D. Piggott
Format: Article
Language:English
Published: Elsevier 2025-05-01
Series:Energy and AI
Subjects:
Online Access:http://www.sciencedirect.com/science/article/pii/S2666546825000175
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1849723227127939072
author Alexander M. Campbell
Simon C. Warder
B. Bhaskaran
Matthew D. Piggott
author_facet Alexander M. Campbell
Simon C. Warder
B. Bhaskaran
Matthew D. Piggott
author_sort Alexander M. Campbell
collection DOAJ
description High-resolution wind speed forecasts are of great importance to the wind energy industry, from short-term energy forecasting and trading to longer-term resource assessment and planning. Generating high-resolution regional wind forecasts currently requires compute-intensive numerical models to downscale from a global forecast. Black-box AI models, once trained, can produce results in a fraction of the time and cost; however, they tend to produce smoothed outputs, are not interpretable and generalise poorly. The domain-informed AI architecture presented in this work seeks to address these problems by incorporating prior static fields directly into the model architecture. Specifically, the proposed approach combines two sequential U-Nets – the first upsamples the input wind fields and expands the number of feature maps, a fusion layer then injects prior static data such as topography, and a second U-Net generates the final output wind field. This approach improves all performance metrics versus a baseline U-Net model and generalises better to out-of-sample scenarios. In addition, this study compares the performance of several loss functions, including standard pixel-wise measures such as mean-squared error, structural similarity and frequency-focused functions, and a function based on Wiener filter theory. All loss functions, with the exception of the Wiener loss, perform comparably and tend to attenuate higher-frequency detail. Although the Wiener loss encourages higher frequencies, it over-estimates amplitudes. A composite Wiener-L1 loss function balances generating high-frequency detail and correctly predicting amplitudes.
format Article
id doaj-art-2825cded1bc042a6afc32e3fb13fac94
institution DOAJ
issn 2666-5468
language English
publishDate 2025-05-01
publisher Elsevier
record_format Article
series Energy and AI
spelling doaj-art-2825cded1bc042a6afc32e3fb13fac942025-08-20T03:11:06ZengElsevierEnergy and AI2666-54682025-05-012010048510.1016/j.egyai.2025.100485Domain-informed CNN architectures for downscaling regional wind forecastsAlexander M. Campbell0Simon C. Warder1B. Bhaskaran2Matthew D. Piggott3Department of Earth Science and Engineering, Imperial College London, London, SW7 2AZ, UK; Corresponding author.Department of Earth Science and Engineering, Imperial College London, London, SW7 2AZ, UKShell Information Technology International Ltd, Shell Centre, London, SE1 7NA, UKDepartment of Earth Science and Engineering, Imperial College London, London, SW7 2AZ, UKHigh-resolution wind speed forecasts are of great importance to the wind energy industry, from short-term energy forecasting and trading to longer-term resource assessment and planning. Generating high-resolution regional wind forecasts currently requires compute-intensive numerical models to downscale from a global forecast. Black-box AI models, once trained, can produce results in a fraction of the time and cost; however, they tend to produce smoothed outputs, are not interpretable and generalise poorly. The domain-informed AI architecture presented in this work seeks to address these problems by incorporating prior static fields directly into the model architecture. Specifically, the proposed approach combines two sequential U-Nets – the first upsamples the input wind fields and expands the number of feature maps, a fusion layer then injects prior static data such as topography, and a second U-Net generates the final output wind field. This approach improves all performance metrics versus a baseline U-Net model and generalises better to out-of-sample scenarios. In addition, this study compares the performance of several loss functions, including standard pixel-wise measures such as mean-squared error, structural similarity and frequency-focused functions, and a function based on Wiener filter theory. All loss functions, with the exception of the Wiener loss, perform comparably and tend to attenuate higher-frequency detail. Although the Wiener loss encourages higher frequencies, it over-estimates amplitudes. A composite Wiener-L1 loss function balances generating high-frequency detail and correctly predicting amplitudes.http://www.sciencedirect.com/science/article/pii/S2666546825000175WeatherWind energyDownscalingCNN
spellingShingle Alexander M. Campbell
Simon C. Warder
B. Bhaskaran
Matthew D. Piggott
Domain-informed CNN architectures for downscaling regional wind forecasts
Energy and AI
Weather
Wind energy
Downscaling
CNN
title Domain-informed CNN architectures for downscaling regional wind forecasts
title_full Domain-informed CNN architectures for downscaling regional wind forecasts
title_fullStr Domain-informed CNN architectures for downscaling regional wind forecasts
title_full_unstemmed Domain-informed CNN architectures for downscaling regional wind forecasts
title_short Domain-informed CNN architectures for downscaling regional wind forecasts
title_sort domain informed cnn architectures for downscaling regional wind forecasts
topic Weather
Wind energy
Downscaling
CNN
url http://www.sciencedirect.com/science/article/pii/S2666546825000175
work_keys_str_mv AT alexandermcampbell domaininformedcnnarchitecturesfordownscalingregionalwindforecasts
AT simoncwarder domaininformedcnnarchitecturesfordownscalingregionalwindforecasts
AT bbhaskaran domaininformedcnnarchitecturesfordownscalingregionalwindforecasts
AT matthewdpiggott domaininformedcnnarchitecturesfordownscalingregionalwindforecasts