DNVGL.com

Breadcrumbs

Winterwind 2019

A conference that focuses on the challenges of generating wind power in cold climates. Winterwind offers seminars, debates, poster exhibition, networking, social events and technical visits. Parallel with the conference there is a fair.

Contact us:

Chris Gowen Chris Gowen
Marketing Communication Advisor - UK, Ireland, Scandinavia, Africa and Middle East

For more information about this event

Contact us
SHARE:
PRINT:
Winterwinds 2016 event

Event Information

Proud to support
DNV GL is proud to be a KiloWatt sponsor at Winterwind 2019.

We will be located on stand 13.

DNV GL Experts
Our experts will be sharing their expertise during the event:

Poster presentation: Wind-farm-scale blockage in stable regime associated with cold climates
Presenter: Stefan Söderberg, Principal Engineer, Renewable Energy Analytics
Date/time: Tuesday 5th February/12:30

Presentation: Using 1Hz data to monitor turbine integrity
Presenter: Carla Ribeiro, Head of Department, Renewable Energy Analytics
Date/time: Tuesday 5 February/13:00
Session: De-/anti-icing including new technologies, ice detection & control incl. Standards
Location: Room 2
Abstract: DNV GL will present improvements to its icing forecasting model through machine learning and the use of liquid water content forecasts.

Presentation: Improving short-term forecasting of turbine icing using machine learning
Presenter: Till Beckford, Team Lead, Renewable Energy Analytics
Date/time: Tuesday 5 February/13:00
Session: Forecasting cloud physics and aerodynamics
Location: Room 3
Abstract: DNV GL will present improvements to its icing forecasting model through machine learning and the use of liquid water content forecasts.

In 2015, DNV GL designed a model which estimates icing losses as part of a short-term wind energy forecast, using an ensemble of meteorological condition forecasts to predict the presence of icing conditions. The model uses an adapted Makonnen model to calculate ice load on turbine blades, and a power reduction model to make icing corrections to the forecast energy production. Analysis shows that while the adapted Makonnen model of ice accretion is very accurate, the ability to predict the onset of icing conditions is more challenging. A new model has been developed that takes advantage of the ability of machine learning classification models to find patterns in multivariate data sets, and introduces liquid water content into the ensemble of meteorological forecasts.

The original model was shown to improve forecast accuracy, reducing annual Mean Absolute Percentage Error (MAPE) by up to 1%, and reducing MAPE by up to 5% for icy months. In this new validation, the improved model has been benchmarked against the original using 2 years of SCADA data from 3 wind farms in Sweden. It shows significant improvement in the accuracy of short-term forecasting of turbine icing, and will provide a financial benefit to end-users of icing forecasts used within DNV GL’s wind power forecasting system. DNV GL has also carried out a “market value analysis”, showing the approximate increase in revenue that can be generated by trading the resulting forecasts on the Elspot day-ahead energy market.

Icing has been shown to cause significant power losses for wind turbines in cold climates, such as Scandinavia. Studies have shown that over the course of a year wind farms can lose up to 13% of power due to icing, with monthly losses up to 50%. Individual icing events can lead to full power loss for a wind farm for over a week at a time. Quantitative fore-warning of such events is therefore a necessary requirement in incorporating the generation from these wind farms on the grid system. DNV GL has been providing short-term wind power forecasting services since 2003 globally for TSOs, utilities and asset operators, and currently forecasting for over 50GW of total installed capacity. We have recognised that forecasting icing events accurately is crucial for both AO&M purposes and for energy trading in some regions, and therefore have continuously investigated methods to better capture these events.

Presentation: WICE 2.0 – The new generation of ice loss models
Presenter: Stefan Söderberg, Principal Engineer, Renewable Energy Analytics
Date/time:Tuesday 5 February/15:30
Session: Pre-construction site assessment, measurements, models and standards
Location: Room 2
Abstract: In September 2018, DNV GL joined forces with Swedish cold climate experts WeatherTech. For many years, both organizations have taken an active part in the development of the knowledge base and innovative services for areas affected by atmospheric icing. The combined expertise of both companies has enabled the development of state of the art models, allowing customers to better predict the performance of turbines in cold climates.

WeatherTech has been developing a combined atmospheric and machine learning model to predict production losses caused by ice accretion on turbine blades, the WICE model. DNV GL’s approach has been to rely on the considerable amount of both production data from wind farms in cold climate, as meteorological data from measurement masts, to develop and empirical method of estimating these production losses. By combining WeatherTech’s WICE model, with DNV GL’s unique database of production data and experience in analysing such data, an unparalleled tool to predict icing losses has been developed. This model will be applicable worldwide when fully developed, and will be able to predict the benefit of IPS (Ice protection Systems) in reducing such losses.

In the present work, the authors have undertaken an independent validation study of the current version of WICE, in which model predictions have been compared with SCADA data from several wind farms and winter seasons not previously known to WeatherTech or used in the training of the model. The analysis has been carried out on a wind farm level as well on a turbine by turbine level. Fundamentally, a unique long-term validation has also been undertaken. For sites with a long data record, an evaluation of long term correction methods has also been carried out.

In parallel with the validation study, the authors have also worked on improving the model. Improvements in many areas have been made, namely the machine learning setup used, the processing of the production data, the addition of more training sites, and the long-term correction method. The authors propose to present the outcome of the independent validation and quantify the model improvements, and how these can aid the cold climate industry in reducing uncertainties.

Pre-book meetings
To pre-book a meeting or for any further information, please email contact.energy@dnvgl.com.