The Challenge
Her Majesty’s Government (HMG), under the Offshore Wind Sector Deal, plans to install 40GW of offshore wind electricity generation capacity (expected to represent ca 40% of the UK’s electricity need) from offshore windfarms in the UK by 2030. This deal has also set a target of supporting 60,000 jobs and £2.4bn per annum of exports by 2030. However, this potential expansion of windfarm locations presents a problem for air surveillance radar operators who are tasked with understanding the “air picture”.
Wind turbine sites consist of large structures which present significant Radar Cross Sections (RCS). In static positions, these returns can be mitigated using standard filtering techniques, however, a rotating turbine blade produces strong radar returns that can’t be mitigated through traditional clutter filtering.
We were tasked to investigate the use of the latest in machine learning techniques to better enable detection and tracking of potential risks in the vicinity of a wind farm.


The Approach
The doppler shift from turbine blades can be sufficiently large to appear in the side-lobes of a radar. Wind turbines will also tend to scatter the radar signal in many directions, and this can cause multi-path effects leading to false detection tracks.
Differentiating between spurious and real tracks can be a difficult task for an operator or a tracking algorithm but through a combination of modeling, simulation and scaled experiments, we planned to demonstrate that applying Deep Learning architectures can help to remove wind turbine returns whilst retaining the returns from tracks of interest.
The Approach
The doppler shift from turbine blades can be sufficiently large to appear in the side-lobes of a radar. Wind turbines will also tend to scatter the radar signal in many directions, and this can cause multi-path effects leading to false detection tracks.
Differentiating between spurious and real tracks can be a difficult task for an operator or a tracking algorithm but through a combination of modelling, simulation and scaled experiments, we planned to demonstrate that applying Deep Learning architectures can help to remove wind turbine returns whilst retaining the returns from tracks of interest.

The Outcome
We successfully developed simulation and scaled experiments which identified various deep learning approaches and determined which route was best suited to the task.
Our deep learning approach has been shown to remove background rotational clutter to better enable target detection and tracking and we are about to start a further phase with our customer to develop this technology further.