# XAI for transparent wind turbine power curve models

**Authors:** Simon Letzgus  
**arXiv:** [2210.12104](https://arxiv.org/abs/2210.12104)  
**Date:** 2022-10-21  
**Summary page:** https://www.summarizepaper.com/en/paper/2210.12104v2/

## Key points

- Importance of accurate wind turbine power curve models in advancing wind energy for global energy transition
- Machine learning methods have shown superiority but lack transparency, hindering practical application
- Use of Shapley values as an explainable artificial intelligence method to uncover strategies learned by ML models from operational wind turbine data
- Concerning trend towards complex model architectures driven by test set performance focus leading to development of physically implausible strategies
- Call for greater integration of XAI methods in model selection processes for transparency and reliability in wind turbine power curve modeling
- Proposal to utilize explanations provided by XAI for root cause analysis in wind turbine performance monitoring to reduce downtime and optimize utilization of turbines

## Summary

In the study "XAI for transparent wind turbine power curve models" by Simon Letzgus, the importance of accurate wind turbine power curve models is emphasized in advancing wind energy as a solution for global energy transition. While machine learning (ML) methods have shown superiority over other approaches, their lack of transparency has hindered practical application. To address this issue, Shapley values - an explainable artificial intelligence (XAI) method - are applied along with latest insights to uncover strategies learned by ML models from operational wind turbine data. The findings highlight a concerning trend towards complex model architectures driven by test set performance focus, leading to development of physically implausible strategies. Thus, there is a call for greater integration of XAI methods in model selection processes to ensure transparency and reliability in wind turbine power curve modeling. Additionally, a practical approach is proposed to utilize explanations provided by XAI for root cause analysis in wind turbine performance monitoring. This can reduce downtime and optimize utilization of turbines in the field. The research emphasizes the significance of incorporating transparent and interpretable AI methodologies in developing accurate and applicable wind turbine power curve models for driving forward sustainable energy solutions.