Wind Turbine Control Using Machine Learning for Power Optimization

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Wind turbines face inefficiencies due to varying wind conditions throughout the year, as a single setting does not optimize annual energy production, leading to suboptimal power output and mechanical stress.

Innovation Solution

A method using machine learning to collect and analyze historical and real-time data from wind turbines, training models to estimate power output and differential, and adjusting settings such as pitch and tip speed ratio to optimize energy production based on current conditions, comparing estimated outputs to reference values and applying settings if they exceed a threshold.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Productivity

If a single turbine setting is used throughout the year, then the device complexity is reduced and ease of operation is improved, but the productivity (annual energy production) deteriorates due to inability to adapt to varying wind conditions

Engineering Contradiction:
Improveannual energy productionVSAvoidcontrol system complexity
Core Design Contradiction:
ProductivityVSDevice complexity

Solution Approach 1:

The wind turbine control system automatically adjusts turbine settings by applying a machine learning model to current operational data, eliminating the need for manual seasonal adjustments. The system self-optimizes by comparing estimated power output against reference values and automatically implementing setting changes when improvements exceed a threshold, making the complex adaptation process autonomous and无需人工干预

Inventive Principle:
Principle #25Self-service

Solution Approach 2:

The system changes operational parameters (pitch setpoint, tip speed ratio setpoint) based on varying wind conditions. The machine learning model determines optimal parameter values by analyzing historical data and current conditions, allowing the turbine to adapt to different wind regimes throughout the year without manual intervention

Inventive Principle:
Principle #35Parameter changes

2Productivity

If turbine settings are changed manually twice per year, then some adaptation to seasonal changes is achieved, but the productivity deteriorates because a single setting cannot optimize for all wind conditions throughout the year

Engineering Contradiction:
Improveannual energy productionVSAvoidadaptability to varying wind conditions
Core Design Contradiction:
ProductivityVSAdaptability or versatility

Solution Approach 1:

The control system transitions from static seasonal adjustments to dynamic continuous optimization. The machine learning model is applied repeatedly at predetermined time intervals throughout the day and year, allowing the turbine settings to dynamically adapt to changing wind conditions rather than remaining fixed for extended periods

Inventive Principle:
Principle #15Dynamics

Solution Approach 2:

The system implements feedback by continuously monitoring current operational conditions, applying the machine learning model to estimate power output, comparing against reference values, and automatically adjusting settings when improvements are detected. This closed-loop feedback mechanism enables continuous optimization rather than open-loop manual adjustments

Inventive Principle:
Principle #23Feedback

3Productivity

If machine learning models are applied frequently to adjust settings, then the productivity and adaptability are improved, but the device complexity and computational requirements increase

Engineering Contradiction:
Improvepower output optimizationVSAvoidcontrol system complexity
Core Design Contradiction:
ProductivityVSDevice complexity

Solution Approach 1:

The system applies partial action by implementing a threshold-based decision rule: the machine learning model is applied frequently, but settings are only changed when the estimated power improvement exceeds a predetermined threshold. This avoids unnecessary adjustments while capturing significant optimization opportunities, balancing computational effort with actual benefit

Inventive Principle:
Principle #16Partial or excessive action

Data Source

PatentEP3613982B1Method for controlling operation of a wind turbine
Publication Date: 2023.01.25 GENERAL ELECTRIC RENOVABLES ESPANA SL
  • EP3613982B1 patent drawingFigure 1
  • EP3613982B1 patent drawingFigure 2
  • EP3613982B1 patent drawingFigure 3

AI summary

A method (600) for controlling operation of a wind turbine (100) includes collecting (610) training data, training (620) a machine learning model, obtaining (640) recent data, and applying (650) the machine learning model the recent data to output a reference power or reference power differential corresponding to the recent data. The machine learning model is then applied to the recent data to output at least one of estimated power or estimated power differential corresponding to values of the pitch setpoints and the tip speed ratio setpoints which differ from the recent data. A turbine setting is determined by comparing (660) the estimated power or estimated power differential to the reference power or reference power differential, and then applying (670) the turbine setting to the wind turbine if the estimated power or estimated power differential is greater than or equal to a threshold amount above the reference power or reference power differential.