低样本学习下的浮式风机混合模型试验及尺度效应问题的研究

本项目围绕国家在深海风电能源开发技术的重大战略需求，聚焦浮式风机模型实验中的尺度效应问题。基于申请人在新能原设备新型模型实验方法领域的研究基础，为了对风机所受风载荷准确的预测，结合人工智能技术，拟利用机器学习里的低样本学习方法，利用对浮式风电场的实时监测来获得全尺度样机数据，从而用来训练BP神经网络结构，可以根据相关参数的输入实时预测对风载荷（包括转子推力，叶片激振力，陀螺效应力矩）。然后结合硬件在环实验技术对预测的风载荷数值进行实时模拟并施加在物理模型上，从而形成一套依赖于全尺度风电场监测数据的，基于低样本学习的浮式风机混合模型实验方法，并通过对单柱式风机模型实验验证实验方法，通过与先前基于仿真数据的人工智能混合模型试验的数据对比，深入分析在模型实验中引入全尺度监测数据的影响，为风机设计规范要求从全尺度数据中寻求风机设计优化和指导的规定提供技术支持。

This project focuses on the country's major strategic needs for deep-sea wind energy exploration and the scale effect in the floating wind turbine model test. In order to precisely predict the wind loa, the applicant intends to apply the low-sample learning method in training BP neural network structures using full-scale prototype data, which are obtained by real-time monitoring of floating wind farms. The prediction module thereby can be used to provide real-time values of wind load( such as rotor thrust, blade vibration force and gyro moment) based on relevant parameters. Combined with the hardware-in-the-loop experimental technology, the predicted values of the wind load is simulated in real-time and applied to the physical model, thereby forming a set of floating wind turbine hybrid model experimental methods that rely on full-scale wind farm monitoring data and based on low-sample learning. A spar-type floating platform model will be used in the experiment to verify the testing method. The in-depth analysis of the impact of introducing the full-scale monitoring data in the model experiment is addressed by comparing the testing results with previous ones. Hence to provide technical support in seeking wind turbine design optimization and guidance.