基于神经网络技术的风力机叶片覆冰预测方法

Icing condition prediction of wind turbine blade based on neural network technology

安装在寒冷地区的风力机叶片不可避免地会发生覆冰现象,影响风电机组的正常工作。叶片的不同覆冰位置和覆冰质量,使得叶片固有频率发生不同改变。文章对2 MW风力机叶片进行有限元模态分析,构建了覆冰叶片的固有频率变化率和不同位置覆冰质量之间的关系,利用该数据集训练BP人工神经网络并对预测能力进行分析。研究表明,使用考虑叶片不同位置覆冰对全覆冰叶片的固有频率的影响程度构建的关系式训练人工神经网络,预测叶片全部覆冰平均相对误差率为13.21%,比Gantasala等人的预测精度提高1.56%。训练好的人工神经网络模型可实现叶片覆冰位置和质量的检测,可为后续加热除冰、超声波除冰等提供数据支撑,提高除冰效率,降低能源消耗。

Wind turbine blades installed in cold areas are inevitably covered with ice, which affects the normal operation of wind turbines. The natural frequencies of blades vary with the locations and mass of icing. In this paper, the finite element modal analysis of 2 MW wind turbine blade is carried out, and the relationships between the change rate of natural frequency of ice-covered blade and the mass of ice-covered blade at different positions are constructed. The BP artificial neural network is trained with this data set and the prediction ability is analyzed. The research shows that the average relative error rate of predicting all ice-covered blades is 13.21%, which is 1.56% higher than the prediction accuracy of Gantasala et al., by using the relational training artificial neural network which is constructed considering the influence degree of different ice-covered locations on the natural frequency of all ice-covered leaves. The trained artificial neural network model can predict the locations and mass of icing. The result can provide data support for subsequent heating or ultrasonic deicing and improve deicing efficiency and reduce energy consumption.