Regularized Hidden Markov Modeling with Applications to Wind Speed Predictions in Offshore Wind

22 Pages Posted: 16 Apr 2023

See all articles by Anna Haensch

Anna Haensch

Tufts University

Eleonora M. Tronci

Tufts University

Bridget Moynihan

Tufts University

Babak Moaveni

Tufts University

Abstract

Offshore wind power is rapidly growing as a vital component of the clean energy transition. As turbine design capacities continue to increase, there is a growing interest in monitoring both individual turbines and entire wind farms to ensure their performance while also reducing the levelized cost of energy. However, obtaining reliable and comprehensive data on these structures can be challenging, as it often requires costly and potentially dangerous installation procedures and significant computational resources. Therefore, it is critical to predict the needed information to properly assess the performance of offshore wind turbines when not available.In particular, the ability to accurately predict wind information can greatly assist in ensuring the safety and longevity of offshore wind turbines, while also reducing the costs. Being a quantity directly correlated to power generation and the wind input loads it can help better understanding the overall performance of the monitored system.To address these challenges, this paper introduces a modified Hidden Markov Model (HMM) frameworkbased strategy and the corresponding Python library, Hela. HMMs are unsupervised probabilistic models for sequential data capable of predicting time-based hidden states and information by looking at and learning from multiple observation types. The proposed HMM-based framework is designed for realworld applications, and it is used to predict wind speed ranges for an instrumented offshore wind turbine using the measured strain and acceleration data. The proposed modified HMM algorithm incorporates an alternative initialization technique and a regularization strategy to overcome some of the limitations of applying HMMs to experimental cases.The paper presents the modified HMM methodology in its theoretical structure, and it defines the different parameter choices and options for the user in the associated Python library, Hela. The code implementation is user-friendly and highly flexible in tailoring the algorithm to different applications.The strategy that was tested in this study exhibited exceptional classification accuracy in distinguishing between low and high wind speed categories, utilizing only the bending moments derived from the strain quantities. Additionally, the introduced regularization and initialization techniques demonstrated tangible improvements in the overall classification performance. Lastly, the Hidden Markov Model (HMM) method significantly outperformed established and common unsupervised data-driven strategies.The proposed framework and Hela library have the potential to significantly advance the field of offshore wind power by enabling more accurate and efficient monitoring and maintenance of wind turbines and wind farms. Furthermore, the optimal performance in predicting unmeasured quantities sets the proposed strategy to be a valid solution for several applications with unlabeled and missing information.

Keywords: Offshore Wind Turbines, Regularized Hidden Markov Modeling, Wind Speed Prediction, Vibration Monitoring

Suggested Citation

Haensch, Anna and Tronci, Eleonora M. and Moynihan, Bridget and Moaveni, Babak, Regularized Hidden Markov Modeling with Applications to Wind Speed Predictions in Offshore Wind. Available at SSRN: https://ssrn.com/abstract=4419807 or http://dx.doi.org/10.2139/ssrn.4419807

Anna Haensch

Tufts University ( email )

Medford, MA 02155
United States

Eleonora M. Tronci

Tufts University ( email )

Medford, MA 02155
United States

Bridget Moynihan

Tufts University

Medford, MA 02155
United States

Babak Moaveni (Contact Author)

Tufts University

Medford, MA 02155
United States

Do you have a job opening that you would like to promote on SSRN?

Paper statistics

Downloads
46
Abstract Views
206
PlumX Metrics