class: top, left, inverse, title-slide .title[ # Estimate the number of offshore wind turbines using machine learning ] .author[ ### Lampros Sp. Mouselimis ] .institute[ ### Monopteryx ] .date[ ### 2023-09-20
monopteryx-dashboard/
linkedin
EURACTIV, March 2023
] --- class:hide_logo
# Advantages of offshore wind energy <br> Offshore wind energy is a clean, renewable source of energy that has the potential to generate a significant amount of electricity. Offshore wind farms are typically located in shallow waters, where the wind is stronger and more consistent than on land. <br> Offshore wind energy has the following advantages compared to other renewable sources: * it has a very low environmental impact: Offshore wind farms do not produce any emissions, and they have a relatively small impact on marine life * it is a very scalable source of energy: as technology improves, it is possible to build larger and more powerful offshore wind farms * offshore wind energy is a cost-competitive source of energy: The cost of offshore wind energy has been declining in recent years, and it is now competitive with other forms of electricity generation --- # Proposed solution <br> An important question that comes to mind is: * *How many and where wind turbines (or wind farms) can be built?* As of September 2023 I'm not aware of a solution (either in business or academic sector) that uses machine learning to predict the number of offshore wind turbines within a fixed geospatial grid. There are existing papers which compute, * [*optimal wind farm locations using a multi-stage process*](https://www.mdpi.com/2071-1050/10/3/749) * [*suitable sites for wind farms based on multi-criteria decision making*](https://www.sciencedirect.com/science/article/pii/S235248472100768X) * [*Fuzzy Analytic Hierarchy Process using GIS for offshore wind farm development*](https://www.sciencedirect.com/science/article/pii/S2590174521000283) <br> My proposed solution, which uses open geospatial data is capable of estimating the number of offshore wind turbines for a pre-specified resolution (for instance for a 2.27 x 2.27 km^2 hexagonal area). --- # Methodology <br> The following pre-processing & data sources were used: * wind farm locations were extracted based on Sentinel-1 satellite imagery * geospatial distances between existing wind turbines and nearest coasts were computed * EMODnet Bathymetry data was included * wind speed, weibull distribution & air density was incorporated * the Global Shipping Traffic Density was taken into account Once the data was in analysis ready format, algorithms were utilized to: * create a fixed grid for the European Area of resolution 2.27 x 2.27 km^2 using geospatial R packages * match the number of offshore wind turbines per grid cell and use it as a response variable * create predictors based on distance, bathymetry, wind speed and ship routes * train an ensemble of tree-based machine learning models and predict the number of offshore wind turbines per grid cell for all European countries (30 in total) * evaluate the results with cross-validation by using the Root Mean Squared Error (RMSE) as the evaluation metric --- # Workflow The following diagram shows the workflow: <img src="images/diagram.png" width="90%" style="display: block; margin: auto;" /> The next images show the results in alphabetical order for a sample of 3 countries --- class:hide_logo # Results (Denmark) <img src="images/Denmark.png" width="100%" style="display: block; margin: auto;" /> .pull-left[ <br><br><br><br><br> The area on the right, which is colored red, has a higher potential for wind turbine installations. The yellow colored grid cells include high estimated values as the Legend on the top-right of the image shows ] .pull-right[ <br><br> <img src="images/dk_2.png" width="90%" style="display: block; margin: auto;" /> ] --- class:hide_logo # Results (Malta) <img src="images/malta.png" width="100%" style="display: block; margin: auto;" /> .pull-left[ <br><br><br><br><br> The popups include additional information related to the area of the grid cell and the estimated number of wind turbine installations ] .pull-right[ <br><br> <img src="images/ml_3.png" width="90%" style="display: block; margin: auto;" /> ] --- class:hide_logo # Results (Netherlands) <img src="images/nl.png" width="100%" style="display: block; margin: auto;" /> .pull-left[ <br><br><br><br><br> In case of the Netherlands we see the horizontal and vertical lines (in green color) where the grid cells take the lowest values and they might correspond to shipping routes. The red color circles show high values for the predicted number of wind turbines ] .pull-right[ <br><br> <img src="images/nl_2.png" width="90%" style="display: block; margin: auto;" /> ] --- class:hide_logo # Web Browser Application A web browser application prototype is implemented to interactively visualize the predictions of the available countries, where a user can switch between countries, algorithms and leaflet providers using drop-down menus <br> <img src="images/greece.png" width="100%" style="display: block; margin: auto;" /> --- # Professional Services If you are looking for a professional to process remote sensing data and extract information using the R and Python programming languages don't hesitate to contact me https://monopteryx.netlify.app/contact/ <br> **References:** * [A global offshore wind turbine data set derived with deep learning from Sentinel-1 data](https://essd.copernicus.org/articles/14/4251/2022/) * [EMODnet Bathymetry data](https://emodnet.ec.europa.eu/) * [globalwindatlas](https://globalwindatlas.info/en/) * IMF's World Seaborne Trade monitoring system, Cerdeiro, Komaromi, Liu and Saeed, 2020 * [R programming](https://www.r-project.org/) & [Rstudio](https://posit.co/download/rstudio-desktop/)