大型风电机组实时可靠性评估与预防维护策略研究

With the development of the wind power, the safety, reliability and stability of system operation is now becoming more and more important. Due to the complexity of the wind turbine system and their volatile operating environments, traditional methods have difficult to make reliability assessment and dynamic predictive maintenance policies. This research focuses on the reliability assessment and predictive maintenance of the wind turbine with the integration of multi-disciplinary theories and techniques. Some information fusion methods are adopted according to features of the wind turbine system. The research emphases are as follows: the reliability assessments of multi-source information based on the Bayes fusion for multi-system and multi-fault of wind turbine; a real-time reliability evaluation using D-S evidence theory for uncertainty of dynamic conditions; a hierarchy optimal predictive maintenance policy based on multi-attribute decision technique and fuzzy theory. Based on the above research, a hybrid system is proposed based on Bayes fusion, D-S evidence theory, multi-attribute decision and multi-agent system. An integration of real-time reliability assessment and dynamic predictive maintenance for wind turbines is developed using the hrbrid system. Thus research is expected to provide a theoretical and technical support for maintaining the operational reliability of the wind turbine system security and preventing failures.

随着风能利用的快速发展，对大型风电机组系统安全、可靠、稳定运行的要求也越来越高。大型风电机组由于系统复杂、运行环境多变，传统单一方法已很难满足系统运行可靠性评估和动态预防维护的要求。本项目集成多学科理论与方法，开展大型风电机组系统实时可靠性评估与预防维护策略研究。针对风电机组系统特点，拟采用信息融合方法，着重研究大型风电机组多部件与多失效模式贝叶斯融合的多源可靠性信息评估方法；基于D-S证据理论的大型风电机组动态环境下不确定状态的实时可靠性评估；基于模糊多目标决策支持系统的大型风电机组分层预防维护策略；在此基础上，提出贝叶斯融合理论、D-S证据理论和模糊多目标决策的多智能体混合体系，实现基于协同多智能体系统的大型风电机组实时可靠性评估与动态预防维护集成系统。本项目的研究有望降低风电机组的风险和维修成本，为最大限度保证系统的正常运行提供一定的理论与技术支撑。

随着风能利用的快速发展，对大型风电机组系统安全、可靠、稳定运行的要求也越来越高.。大型风电机组由于系统复杂、运行环境多变，传统单一方法已很难满足系统关键部件运行可靠性评估和动态预防维护的要求。本项目集成多学科理论与方法，开展大型风电机组系统关键部件实时可靠性评估与预防维护策略研究。在智能诊断领域，①提出了基于自适应特征提取和基于仿射传播的智能诊断方法。自适应特征选择新方法和自权重(Self-Weight)算法，该方法可自适应选择最优特征，提高了整个模型的智能化程度。不同故障类别与不同损伤程度的故障诊断结果表明，所提自适应特征选择方法大大提高了AP聚类效果和准确率。②VMD多源特征去噪与自适应密度峰值搜索算法(ADPS)的智能故障诊断方法。该方法采用将VMD算法用于特征去噪，可以根据样本自动确定聚类个数。在轴承故障和齿轮故障的定性诊断中，ADPS算法取得了满意的聚类效果；在对轴承全生命周期无标签数据的诊断中，ADPS聚类效果优良，进一步验证了该算法的优势。多源故障特征的故障预测领域，③研究了自适应谱峭度参数影响及在轴承全生命周期故障特征提取方法；④提出了基于加权约束稀疏编码的故障趋势与剩余寿命预测方法；⑤基于多域特征与改进D-S证据理论的齿轮箱故障智能诊断。.本项目的研究有望降低风电机组关键部件的故障风险和维修成本，为最大限度保证系统的正常运行提供一定的理论与技术支撑。