Machine learning to tap into the potential of offshore wind power
In 1887, inventor Charles Brush built a wind turbine in the backyard of his mansion in Cleveland Ohio. After generations of improvement by talented engineers, wind power is today one of the fastest-growing renewable energy technologies, encouraged by policy support, technology advances and financial incentives [1,2,3]. Offshore turbines specifically offer tremendous potential because of more consistent and higher wind speeds (improving yield) and fewer restrictions on size. Offshore wind capacity is expected to almost triple to 52 GW by 2023, with half the growth driven by the European Union and the other half by Asia . Under conservative scenarios, this capacity should reach 160 GW by 2030 and 350 GW by 2040 .
Iberdrola, global leader in power generation and one of the largest wind power producers, understands this potential. The group is investing heavily in renewables and in wind power specifically, to support the transition to low-carbon energy. In line with this strategy, its offshore wind capacity increased by 180% from 2016 to 2017 .
However, competitiveness compared to other technologies remains a key obstacle for offshore wind to fulfill this potential [8,9, figure 1]. Operations and Maintenance (O&M) costs account for 15% to 30 % of the Levelized Cost of Electricity (LCoE) and can be a strong lever of optimization [10 – 13]. Traditionally reactive, static and labor-intensive, offshore O&M includes costly unplanned maintenance and inspections – often in harsh conditions – and is expected to represent a $4.9 billion market by 2026 .
This is where machine learning can improve the future of the industry. Using information generated by sensors in the turbines and learning from historical data, maintenance can become predictive. Anomalies and failure modes can be proactively and precisely identified to be addressed before causing any damage or service interruption [15 – 17]. Specific actions can be suggested to operators, switching from a calendar-based maintenance schedule to a much cheaper asset condition-based maintenance.
Looking forward: Iberdrola’s strategy to decrease offshore O&M costs
Iberdrola is leading the ROMEO project (“Reliable OM decision tools and strategies for high LCoE reduction on offshore wind”), working with 12 industrial partners including IBM, OEMs Adwen and Siemens [18,19]. The objective is to develop a platform to manage and analyze data collected from offshore wind turbines to:
- Increase reliability and decrease downtime due to equipment failure,
- Increase the lifespan of key turbine components,
- Reduce O&M costs through a smarter use of resources and decreased need of foundations inspections.
The outcomes of the project will contribute to the improvement of health and safety, to the development of operational best practices in offshore wind, leading to increased competitivity and lower LCOE.
In the short-term (2020 horizon), four major offshore projects are underway in the U.K., Germany and France. East Anglia One (£2.5 billion CAPEX, 714MW) will be the largest offshore wind farm when it starts operations in 2020 and will serve as a pilot site for ROMEO .
In the medium term, Iberdrola plans to invest €9 billion to reach 3 GW of installed capacity in Europe and the U.S. by 2023 . The group will also dedicate €4.8 billion in digital transformation by 2022 and will focus on improving O&M by using data analytics and artificial intelligence and increasing the availability across its generation plants.
Leveraging Machine Learning and AI beyond offshore wind
I believe that the ROMEO outcomes could easily be transferable to other technologies, chiefly onshore wind and photovoltaic. Additionally, forecasting production could also be facilitated by machine learning technology, based on weather data, asset-specific data and site information [22,23]. Accurate predictions would then allow to anticipate and quantify curtailment , to further optimize asset management.
But Iberdrola’s digital strategy should extend beyond asset management. For example, Iberdrola Ventures has recently invested in InnoWatts, a start-up working on smart metering and demand forecasting using machine learning. I would argue that further such acquisitions would strengthen the group’s position as a “utility of the future” and build a stronger capability to lead its digital transformation (including in consumer-facing applications).
As they implement their digital strategy, I would advise Iberdrola to preserve integration and the overall unity of their I.T. landscape. Tremendous amounts of data are measured and generated: coherence and smart management of this data will be key to maintain alignment and efficiency. The development of overarching cloud-based IoT platforms such as GE’s Predix  or Siemens’s Mindsphere  illustrate this trend.
As Iberdrola moves along their digital transformation journey, I see further outstanding issues around:
- the role of engineering expertise and experience: how can analytics platforms complement technical competences and empower employees to improve performance?
- Industry-wide collaboration on data platforms (including OEMs): are consortia such as ROMEO sustainable in a competitive industry?
 Advancing the Growth of the U.S. Wind Industry: Federal Incentives, Funding, and Partnership Opportunities, US Department of Energy (DoE), February 2017
 Innovation Outlook, Offshore Wind, International Renewable Energy Agency, 2016
 Global Landscape of Renewable Energy Finance, International Renewable Energy Agency, 2018
 Annual Market Update 2017, Global Wind Report, General Wind Energy Council
 Renewables 2018, Market analysis and forecast from 2018 to 2023, International Energy Agency
 Offshore Energy Outlook, International Energy Agency 2018
 Iberdrola Integrated Report, February 2018
 Renewable Power Generation Costs, International Renewable Energy Agency, 2018
 Offshore Energy Outlook, International Energy Agency, 2018
 Sensitivity analysis of offshore wind farm operation and maintenance cost and availability, R. Martin et al., Renewable Energy, 2015
 Levelized Cost and Levelized Avoided Cost of New Generation Resources in the Annual Energy Outlook 2018, U.S. Energy Information Administration, March 2018
 Levelized cost of energy for offshore floating wind turbines in a life cycle perspective, A. Myhr et al., Renewable Energy, 2014
 Lazard’s Levelized Cost of Energy Analysis, version 11.0, November 2017
 Europe Wind Power Outlook 2017, MAKE consulting, 2017
 Machine learning methods for wind turbine condition monitoring: A review, A. Stetco et al., Renewable Energy, 2018
 Deep Learning for fault detection in wind turbines, J. Helbing et al., Renewable and Sustainable Energy Reviews, 2018
 Failure rate, repair time and unscheduled O&M cost analysis of offshore wind turbines, J. Carroll et al., Wind Energy, June 2016
 ROMEO Seeks to Improve Wind Farms with Machine Learning and IoT at the Edge, IBM research blog, last accessed November 12th 2018
 ROMEO project website (https://www.romeoproject.eu/), last accessed November 12th 2018
 Iberdrola website (www.iberdrola.com) last accessed November 12th 2018
 Iberdrola press release, October 24th 2018
 Deep learning based ensemble approach for probabilistic wind power forecasting, H. Wang et al., Applied Energy, 2017
 Ensemble methods for wind and solar power forecasting—A state-of-the-art review, Y. Ren et al., Renewable and Sustainable Energy Review, 2015
 Wind and Solar Energy Curtailment: Experience and Practices in the United States, National Renewable Energy Laboratory, 2014
 Digital Wind Asset Performance Management from GE Renewable Energy, 2017
 The Predix platform, GE website (https://www.ge.com/digital/iiot-platform) last accessed November 12th 2018
 Siemens Mindsphere, Siemens website (https://www.siemens.com/global/en/home/products/software/mindsphere-iot.html) last accessed November 12th 2018