水电机组综合质差异挖掘和差异化模式表征的仿生强智能故障诊断

The fault detection and diagnosis of hydropower units not only is the guarantee of its own safe operation, and is the essential demand of the national energy sustainable development. The complexity, strong coupling, mapping fuzziness and uncertainty of the hydropower unit faults, especially the increasing of the hydropower unit automation level, all give a new challenge to the intellectuality of fault diagnosis. Therefore, that carrying out intelligent exploration in different levels of fault diagnosis and making further study of comprehensive intelligent fault diagnosis based on the vertical integration of intelligent methods are meaningful to the hydropower units. This research is proposed to simulate the human brain thinking way, carry out intelligence researches in feature selection, fault characterization and failure recognition, and then establish strong intelligent fault diagnosis system for hydropower units. First of all, through simulating the feature selection mechanism of human brain, an instructive tree model is built with fuzzy logic to realize the intelligent optimization of the feature subset and explore the coupling and diffuse qualitative difference relations among hydropower unit faults. Then, the fuzzy associated feature vector (FAFV) is designed through simulating the object representation mechanism of human brain, and at the same time, an invalid-item is also defined for FAFV. The quality differences and quantity differences of the multi-dimensional fault are comprehensively integrated in FAFV to achieve the intelligent discrepant fault representation. Finally, on the basis of the intelligent optimization of feature subset and the discrepant intelligent fault characterization, the machine learning method is introduced to establish the biomimetic strong intelligent fault diagnosis method for hydropower units.

水电机组的故障检测与诊断不仅是其安全运行的保障，而且是国家能源可持续发展的需求。机组故障具有复杂、强耦合、映射模糊、不确定等特点，尤其是机组自动化水平的不断提高，是其故障诊断智能化的新挑战。因此，在故障诊断各层面开展智能化探索，研究纵向融合的全智能诊断问题，对水电机组有重要意义。本研究项目拟模拟人脑思维，从特征选择、故障表征、故障识别三个方面开展智能化研究，建立水电机组仿生强智能诊断系统。首先，模拟人脑分类的特征选择机制，构建启发树模型，同时引入模糊逻辑，在实现特征子集智能优化的同时，探明水电机组故障耦合和扩散的质差异关系；然后，模拟人脑分类的对象表征机制，设计模糊关联特征向量，并定义其无效项，以其独特的结构和无效项全面融合多维故障的质差异和量差异，实现机组故障的差异化智能表征；最后，在模拟人脑智能特征子集优化和智能故障表征的基础上，结合机器学习方法，实现水电机组故障的仿生强智能诊断。

近年来，水电机组更多地服务于间歇性负荷使水电机组所面临的运行环境将更加复杂，机组的大型化和复杂化使得其安全问题日益凸显，成为制约我国大型水电装备技术的瓶颈。对水电机组开展有效的故障检测与诊断策略研究，探寻水电机组故障的机理及应对策略不仅是保证水电机组安全运行的必要措施，更是国家缓解能源危机、改善环境污染、高效开发利用水电能源的必然需求。水利科学与海洋工程学科十三五发展战略规划（2016-2020）特别建议水力装备学科开展故障诊断的智能化理论与新技术、状态监测与故障诊断的新理论与新技术的研究。.本项目组严格执行项目计划，主要从以下三个方面进行了研究：.基于质差异挖掘的自适应特征选择研究：模拟人脑树形分类的启发式策略和自适应特征选择机制，设计了树启发式的自适应差异化特征选择方法，同时提出并构建了靶向特征选择方法，成功获取了每一种故障特有关键特征子集，实现了对水电机组轴系故障关键特征集差异性的深层次挖掘。.基于综合质差异挖掘的差异化故障表征研究：模拟人脑树形分类独特的特征组织机制，提出并构建了关联特征向量结构，综合挖掘水电机组故障间的复杂差异性，在此基础上，引入模糊理论和粗糙集分别构建了模糊关联特征向量和粗糙关联特征向量，进一步提升了关联特征向量的容错性，实现了水电机组轴系故障的有效差异化模式表征。.建立了基于综合质差异挖掘和差异模式表征的仿生强智能故障诊断系统：保留特征选择的树启发机制，结合模糊理论和粗糙集设计了启发式特征提取方法；将智能化的特征选择、模式表征、特征提取和机器学习方法纵向融合，建立基于综合质差异挖掘和差异化模式表征的仿生强智能诊断系统，实现水电机组故障的仿生智能识别。将滚动轴承故障诊断准确率提高到了99.83%，将水电机组故障集诊断准确率提升到了100%。.项目组发表了9篇学术论文，其中SCI检索2篇，EI检索4篇；申请了发明专利2项；参加了学术会议4次；培养了硕士研究生6名，完成了项目的成果预期。