Wind gust detection and load mitigation using artificial neural networks assisted control

Abstract

The design of a wind turbine implies the simulation of definite conditions as specified in the standards. Among those operational conditions, rare events such as extreme gusts or external faults are included, which may cause high structural loads. Such extreme design load cases usually drive the design of some of the main components of the wind turbine: tower, blades and mainframe. Two different strategies are hence presented to mitigate the loads, deriving from extreme load cases, on the basis of the detection of wind gusts by means of ad hoc synthesized artificial neural networks. This tool is embedded into the main control algorithm and allows it to detect the gust in advance, to anticipate the control reaction, and by doing so reducing extreme loads. One of the strategies performs a controlled stop when wind gust is detected. The other rides through wind gusts without stopping, i.e., without affecting the wind turbine normal operation. Aeroelastic simulations of the Alstom Wind's wind turbines using these techniques have shown significant reductions in the extreme loads for all standard IEC 61400-1, edition 2 DLC 1.6 cases. In particular, the overall ultimate loads are largely reduced for blade root and tower base bending moments, with a direct impact on the structural design of those components.

Detection and compensation of anomalous conditions in a wind turbine

“…Although work on anomaly detection using a LIDAR is becoming more popular in the literature [4,5,6,7], work on this topic without the use of a LIDAR is still limited [8,9,10]. Moreover, while the work presented in [8,9,10] is dedicated to gust detection, the work presented here is more comprehensive dealing with various types of anomalies. It is important to note that this paper deals with "anomalies", which should be separated from "faults".…”

Design of Robust Controllers for Load Reduction in Wind Turbines

“…They are usually used to model complex relationships between inputs and outputs or to find patterns in data. Alstom and Ikerlan, in (Carcangiu, 2011), propose an original wind gust detection method based on ANN. In this method, 140 wind samples (input layer) are used to analyze 200 different options (hidden layer) to decide if exists a wind gust in the wind turbine.…”

Reinforcement Learning to Control Lift Coefficient Using Distributed Sensors on a Wind Tunnel Model

“…ANNs have been used in aviation for limited applications which include replacing components of conventional flight control systems [26] such as gains [27], or as look-up tables for aircraft collision avoidance algorithms [28]. Limited research has been conducted on the use and safe application of machine learning for flight control, but some studies have used ANNs in conjunction with adaptive control [29], for gust detection and load mitigation on wind turbines [30], for stabilisation and trajectory control of a hexacopter [31], a flapping wing [32], or a model-free simulation of a UAV [33], and to attain robust non-linear control for active guidance of a finless rocket [34]. Most of these studies trained the ANNs following a supervised learning approach with their performance tested in simulation, rather than with physical platforms.…”