基于改进D-S融合量子粒子群BP网络的水电机组振动故障诊断研究

Most of hydroelectric generator unit fault manifest in the form of vibration, its causes are very complex. For conventional hydroelectric unit vibration fault diagnosis technology prone to inconsistent results problem, combined with the practical of Yunnan Province developing hydroelectric energy industry, this project intends to carry out hydroelectric generator unit vibration fault diagnosis technology research based on the improved D-S evidence theory combines with quantum particle swarm optimized BP neural network. The main contents include: (1) to build an adaptive redundant lifting wavelet noise reduction model of raw vibration signal in order to accurately identify the weak fault information.to study the differential local mean decomposition ability of adaptive decomposition to noise reduction signal and to extract multi-domain characteristic parameter; (2) to acquire the key factors of unit vibration fault by rough set attribute reduction mechanism; (3) the use of quantum particle swarm optimized BP neural network algorithm, to build the vibration spectrum, the vibration amplitude, and other signs subspace and fault subspace primary fault diagnosis model; (4) to study improved algorithm of D-S evidence theory, and carry on the decision-making fusion of multi-source information. Through this study, is expected to deepen the awareness for vibration fault evolution mechanism of hydroelectric generator unit, enrich the theory and methods of unit fault diagnosis research, and provide theoretical and practical basis for the regular maintenance transition to predict state maintenance of hydroelectric generating.

水电机组故障大多以振动的形式表现出来，其成因非常复杂。针对水电机组常规振动故障诊断技术易出现结果不一致的问题，并结合云南省大力发展水电能源产业的实际，本项目拟开展基于改进D-S证据理论融合量子粒子群BP网络的水电机组振动故障诊断技术的研究。主要内容包含:（1）构建原始振动信号的自适应冗余提升小波降噪模型，以准确识别水电机组的微弱故障信息。研究微分局部均值分解对降噪信号的自适应分解能力，提取多域特征参数；（2）由粗糙集属性约简机制获取机组振动故障的关键影响因素；（3）利用量子粒子群BP网络算法，建立振动频谱与振动幅值等多个征兆子空间与故障子空间的初级故障诊断模型；（4）研究D-S证据理论的改进算法，并进行机组振动故障诊断多源信息的融合决策。通过本研究，可望深化对水电机组振动故障演化机理的认识，丰富机组故障诊断研究的理论与方法体系，并为水电机组的定期维护过渡到预知的状态维护提供理论与实践依据。

研究基于证据理论融合量子粒子群BP网络的水电机组振动故障诊断技术，其主要成果包括：开展面向布谷鸟搜索与量子粒子群等算法研究，并提出一系列的性能提升策略；根据所采集的水电机组以及风电齿轮箱等设备的故障样本数据，设计了基于布谷鸟搜索和量子粒子群算法的故障诊断模型；构建了证据理论分别融合改进的量子粒子群和布谷鸟搜索算法的故障识别模型，实现了面向水电机组以及风电齿轮箱的故障诊断。结果显示，所提出的5种改进布谷鸟搜索算法均表现出较强的竞争力。同时，所开发的4种水电机组故障诊断模型具有较高的分类精度，可实现对水电机组振动故障的有效诊断。此外，所设计的3种基于改进量子粒子群算法的风电齿轮箱故障识别模型，也具有良好的诊断性能。项目资助发表SCI检索论文7篇、EI期刊论文6篇、获授权发明专利2项。通过本研究，增强了布谷鸟搜索和量子粒子群算法的寻优能力，丰富了水电机组振动故障诊断研究的理论与方法体系。