发电技术 ›› 2026, Vol. 47 ›› Issue (1): 65-74.DOI: 10.12096/j.2096-4528.pgt.260106

• 新能源 • 上一篇    下一篇

基于多层感知器神经网络的风机叶片覆冰预测模型研究

韩斌1, 曾志祥2, 孔繁新1, 谢楠2, 刘志强2   

  1. 1.西安热工研究院有限公司,陕西省 西安市 710054
    2.中南大学能源院,湖南省 长沙市 410083
  • 收稿日期:2025-01-08 修回日期:2025-02-20 出版日期:2026-02-28 发布日期:2026-02-12
  • 作者简介:韩斌(1981),男,硕士,高级工程师,研究方向为新能源发电,hanbin@tpri.com.cn
    曾志祥(1998),男,硕士研究生,研究方向为风机覆冰,213912074@csu.edu.cn
    孔繁新(1986),男,工程师,研究方向为新能源发电,kongfanxin@tpri.com.cn
    谢楠(1991),男,博士,讲师,研究方向为能源系统开发与评价、气体水合物,xienan@csu.edu.cn
    刘志强(1970),男,博士,教授,研究方向为传热传质与系统节能等,liuzq@csu.edu.cn
  • 基金资助:
    国家自然科学基金项目(52206037);湖南省科技创新计划项目(2023GK1050);湖南省科技创新计划项目(2023JJ60154)

Research on Ice Accretion Prediction Model for Wind Turbine Blades Based on Multi-Layer Perceptron Neural Network

Bin HAN1, Zhixiang ZENG2, Fanxin KONG1, Nan XIE2, Zhiqiang LIU2   

  1. 1.Xi’an Thermal Power Research Institute Co. , Ltd. , Xi’an 710054, Shaanxi Province, China
    2.School of Energy Science and Engineering, Central South University, Changsha 410083, Hunan Province, China
  • Received:2025-01-08 Revised:2025-02-20 Published:2026-02-28 Online:2026-02-12
  • Supported by:
    National Natural Science Foundation of China(52206037);Science and Technology Innovation Program of Hunan Province(2023GK1050)

摘要:

目的 在寒冷地区,风力发电机叶片结冰问题会显著降低发电效率并增加安全隐患,因此精准的结冰预测技术至关重要。为了提高风力发电机叶片结冰预测的准确性,提出一种基于多层感知器神经网络的覆冰预测模型。 方法 采用正交试验与计算流体力学相结合的方法,收集了不同工况下风力发电机叶片的结冰特征数据,并基于这些数据构建了多元线性回归和多层感知器神经网络2种预测模型。 结果 通过平均相对误差和最大相对误差等评价指标进行性能评估,发现多层感知器神经网络的覆冰预测模型对于明冰的预测,其覆冰质量、覆冰最大厚度的平均相对误差均小于7%,最大相对误差均小于20%;对于霜冰的预测,其覆冰质量、覆冰最大厚度的平均相对误差均小于3%,最大相对误差均小于13%。经对比,多层感知器神经网络模型在相对误差等指标上优于多元线性回归模型。 结论 该研究为风电行业提供了一种新的、更精确的结冰预测方法,有助于提升风力发电的安全性和效率。

关键词: 风力发电, 神经网络, 多层感知器, 风机叶片覆冰, 霜冰, 明冰, 预测模型, 风电场

Abstract:

Objectives In cold regions, the icing problem on wind turbine blades significantly reduces power generation efficiency and increases safety risks, making accurate icing prediction technology crucial. To improve the accuracy of ice accretion prediction for wind turbine blades, this study proposes an ice accretion prediction model based on multi-layer perceptron neural network. Methods The study combines orthogonal experiments with computational fluid dynamics to collect ice accretion feature data for wind turbine blades under different operating conditions. Based on these data, two prediction models are developed: multiple linear regression and multi-layer perceptron neural network. Results Performance evaluations using metrics such as average relative error and maximum relative error reveal that the ice accretion prediction model based on multi-layer perceptron neural network achieves an average relative error of less than 7% and a maximum relative error of less than 20% in predicting both ice mass and maximum ice thickness for glaze ice. For rime ice, the model achieves an average relative error of less than 3% and a maximum relative error of less than 13%. By comparison, the multi-layer perceptron neural network model outperforms the multiple linear regression model in terms of relative error and other metrics. Conclusions This study provides a novel and more accurate method for ice accretion prediction in the wind power industry, contributing to improved safety and efficiency in wind power generation.

Key words: wind power generation, neural network, multi-layer perceptron, wind turbine blade icing, rime ice, glaze ice, prediction model, wind farm

中图分类号: