基于免疫算法的风电系统故障诊断技术研究

发布时间：2018-04-27 07:19

本文选题：故障诊断 + 双馈风力发电机　；参考：《南京航空航天大学》2014年博士论文

【摘要】：风电系统使用寿命通常达20年以上,常年经受酷暑严寒和极端温差的考验,受无规律变向、变负荷的风力作用以及强阵风的冲击等原因使得风电系统不可避免地会发生故障。因此,及早发现故障,减少故障所造成的损失是风电技术推广应用的重要研究内容,本文对风电系统故障诊断技术进行了深入的研究。本文首先介绍了双馈风力发电系统的工作原理,分别在3种不同的坐标系下建立了双馈风力发电机的数学模型,将双馈风力发电机模型采用小扰动分析法进行了线性化处理,又进一步建立了双馈风力发电机的空间矢量模型。对双馈风力发电机正常运行时的气隙磁密、定子并联支路环流特性、定子振动特性、转子振动特性进行了分析。在对风电系统故障原因、故障特征及故障机理进行分析的基础上,分别建立了双馈风力发电机定子短路故障模型、气隙偏心故障模型和转子绕组匝间短路故障模型。分析了生物免疫系统的基本原理及其在多条件、多变量情况下的数学模型。对人工免疫算法、单克隆免疫策略算法进行了理论分析并对其收敛性进行了理论证明,给出了免疫单克隆策略、多克隆策略算法的执行步骤。描述了风电系统故障免疫状态空间,建立了基于免疫算法的风电系统故障诊断响应模型。在此基础上借鉴生物免疫系统理论与运行机制将人工免疫系统的理论与风电系统故障诊断相融合,建立了适合风电系统故障诊断的改进的人工免疫系统。建立了风电系统网侧变流器、机侧变流器数学模型和控制策略,在克隆选择、神经网络算法的基础上,将自适应动态克隆选择算法良好的优化性能与BP神经网络相结合,提出了一种基于自适应动态克隆选择算法的神经网络系统,将优化的神经网络应用于风电系统变流器故障诊断。实验结果表明,该方法可以避开局部最小值,算法收敛速度快,具有较好的故障诊断性能。发电机是风电系统的关键组成部分,针对风力发电机的单一故障,本文对双馈风力发电机定子短路故障下气隙磁密、并联支路特性进行了分析,基于双馈风力发电机电压、电流和通量建立了风力发电机定子故障数学模型。提出了一种免疫记忆动态克隆策略算法,并将其应用于双馈风力发电机定子绕组单一故障诊断,将双馈风力发电机的4种故障特征量作为免疫记忆动态克隆策略系统的抗原。用双馈风力发电机工作状态参数对免疫记忆动态克隆策略系统进行训练,将训练阶段获得的记忆数据应用于发电机故障诊断。该方法使用记忆单元作为类别标签,记忆单元根据种群自适应性平均值不断更新,当种群自适应度的标准偏差为零时,存储器单元不改变,可确保算法早期的收敛性。实验结果表明所提出的基于免疫记忆动态克隆策略算法风电系统故障诊断系统具有比较好的分类效果,对于双馈风力发电机定子绕组单一故障诊断是适用和有效的。当风电系统运行时,双馈风力发电机定子电流中的故障信号比较微弱,故障分量的频率与基频分量的频率非常接近,故障分量的幅值也较小,易被泄漏的基频分量及噪声淹没。本文分别对转子故障、偏心故障、偏心与转子复合故障下的气隙磁密、定子并联支路环流特性进行了理论分析。在此基础上将小波分析与人工免疫系统相结合,针对风力发电机的复合故障,提出了一种基于小波-抗体记忆克隆算法的双馈风力发电机偏心和转子复合故障诊断方法。该方法首先对双馈风力发电机定子电流信号进行小波分析,经小波分析计算出小波系数,由小波系数计算出双馈风力发电机故障信号的能量,经过归一化处理形成故障特征量。将故障特征量隐喻为抗原,采用抗体克隆记忆算法生成抗体,经过选择、克隆、变异、抗体再选择、抗体记忆以及压缩等操作产生新的抗体,用产生的新抗体对风力发电机复合故障进行诊断。实验结果表明,本文提出的基于小波和抗体记忆克隆算法相结合的方法取得了比较好的风电系统故障诊断效果。针对风电系统综合故障,分别对风电系统偏心、定子、转子绕组短路故障以及复合故障下的振动特性进行了理论分析,提出了一种基于自适应多克隆策略算法的风电系统综合故障诊断方法。该方法将风电系统振动信号、电流信号作为故障的特征量,将其隐喻为免疫故障诊断系统的抗原。系统根据每个抗原、抗体种群亲合度的大小,自适应地调节抗体对抗原的适应性和抗体种群规模的大小,将抗原归并至某一组确定的抗体种群中。多克隆算法引入了交叉重组操作,使抗体的多样性在进化过程中得到了增加,这使得算法在避免陷入局部极小值的能力和局部搜索能力方面都得到了提高,将该方法应用于风电系统综合故障诊断取得了比较好的效果。基于上述研究,利用GE智能平台的集成开发环境和IFIX组态软件开发了基于免疫算法的风电系统综合故障诊断软件,完成了系统调试和算法验证工作。用故障诊断系统对风电系统综合故障进行诊断,通过实验比较了采用不同复合故障信号诊断方法对风电系统故障诊断结果的影响。实验结果表明,利用风电系统振动、风电系统电流复合信号对风电系统故障进行诊断可以取得比较好的诊断效果,系统具有良好的故障诊断能力。最后,本文对研究成果进行了全面总结,并对基于人工免疫系统的风电系统故障诊断技术进一步研究进行了展望。
[Abstract]:The service life of the wind power system is usually over 20 years, and is subjected to the test of severe heat and extreme temperature difference for a long time. The wind power system will inevitably be broken down by the irregular change of direction, the wind force of changing load and the impact of strong gust. Therefore, the early detection of the fault and the loss caused by the failure are the popularization of the wind power technology. In this paper, the fault diagnosis technology of wind power system is deeply studied in this paper. Firstly, the working principle of doubly fed wind power generation system is introduced. The mathematical model of doubly fed wind generator is established in 3 different coordinate systems, and the doubly fed wind generator model is linear with small disturbance analysis. The space vector model of the doubly fed wind generator is further established. The air gap magnetic density of the doubly fed wind generator in normal operation, the circulation characteristic of the stator parallel branch, the stator vibration characteristics and the rotor vibration characteristics are analyzed. On the basis of the analysis of the fault causes, the fault characteristics and the fault mechanism of the wind power system, the wind power generator is divided into two parts. The stator short fault model of the doubly fed wind generator, the air gap eccentricity fault model and the rotor winding short circuit fault model are not established. The basic principle of the biological immune system and the mathematical model under the condition of multi condition and multivariable are analyzed. The artificial immune algorithm and the monoclonal immunization strategy algorithm are analyzed and collected. The convergence is proved by the theory, the immune monoclonal strategy and the implementation step of the multi clone strategy algorithm are given. The fault immune state space of the wind power system is described, and the fault diagnosis response model of the wind power system based on the immune algorithm is established. On the basis of this, the theory and operation mechanism of the biological immune system are used for the theory of artificial immune system. Combined with fault diagnosis of wind power system, an improved artificial immune system suitable for fault diagnosis of wind power system is established. The wind power system network side converter, the mathematical model and control strategy of the side converter are set up. On the basis of the clonal selection and neural network algorithm, the optimal performance of the adaptive dynamic clonal selection algorithm and the BP God are good. Through the combination of network, a neural network system based on adaptive dynamic clonal selection algorithm is proposed. The optimized neural network is applied to the fault diagnosis of wind power system converter. The experimental results show that the method can avoid the local minimum value, the algorithm converges fast and has good fault diagnosis performance. The generator is a wind power system. In view of the single fault of the wind generator, this paper analyzes the air gap magnetic density and the parallel branch characteristics of the doubly fed wind generator stator short circuit fault. Based on the voltage, current and flux of the doubly fed wind generator, the mathematical model of the stator fault of the wind generator is established. A dynamic clonal strategy for immune memory is proposed. The algorithm is applied to the single fault diagnosis of the stator winding of a doubly fed wind generator, and the 4 fault characteristics of the doubly fed wind generator are used as the antigen of the dynamic cloning strategy system of the immune memory. The training phase of the immune memory dynamic clone system is trained with the working state parameters of the doubly fed wind generator, and the record of the training phase is obtained. Memory unit is used as class label, memory unit is constantly updated according to the adaptive average of the population. When the standard deviation of population adaptive degree is zero, the memory unit is not changed, and the convergence of the algorithm can be ensured. The experimental results show that the proposed immune memory is based on the immune memory. The dynamic cloning strategy algorithm has a good classification effect in wind power system fault diagnosis system. It is applicable and effective for the single fault diagnosis of the stator winding of the doubly fed wind generator. When the wind power system runs, the fault signals in the stator current of the doubly fed wind generator are relatively weak, the frequency of the fault components and the frequency of the fundamental frequency components. It is very close, the amplitude of the fault component is also small, and it is easy to be flooded with the fundamental frequency component and noise of the leakage. This paper analyses the rotor fault, eccentric fault, the air gap magnetic density under the compound fault of the eccentric and the rotor, and analyses the circulation characteristics of the stator parallel branch. On this basis, the wavelet analysis is combined with the artificial immune system, and the wind power is combined with the wind force. A hybrid fault diagnosis method of doubly fed wind generator is proposed based on the wavelet antibody memory cloning algorithm. The method of wavelet analysis is used to analyze the stator current signal of doubly fed wind generator, and the wavelet coefficients are calculated by wavelet analysis. The doubly fed wind generator is calculated by the wavelet coefficients. The energy of the fault signal is processed by normalization. The fault feature is metaphorical as an antigen, and the antibody cloned memory algorithm is used to generate antibodies. After selection, cloning, mutation, antibody selection, antibody memory and compression, new antibodies are produced, and a new antibody produced by the production of a new antibody is used to diagnose the complex fault of wind turbine generator. The experimental results show that a better fault diagnosis effect is obtained by combining the wavelet and the antibody memory algorithm, and the vibration characteristics of the wind power system eccentricity, the stator, the rotor winding short circuit and the complex fault are analyzed. A comprehensive fault diagnosis method for wind power system based on adaptive multi clone strategy algorithm is proposed. This method uses the vibration signal and current signal of the wind power system as the characteristic quantity of the fault, and metaphores it as the antigen of the immune fault diagnosis system. The original adaptability and the size of the antibody population size the antigen into a certain group of antibody populations. The polyclonal algorithm introduces the cross recombination operation to increase the diversity of the antibody in the evolutionary process, which makes the algorithm improve the ability to avoid the local minimum and the local search ability. This method has been applied to the comprehensive fault diagnosis of wind power system. Based on the above research, the integrated development environment of GE intelligent platform and the IFIX configuration software are used to develop the integrated fault diagnosis software of wind power system based on immune algorithm, and the system debugging and calculation verification are completed. The comprehensive fault diagnosis is carried out, and the effects of different fault signal diagnosis methods on the fault diagnosis results of wind power system are compared through experiments. The experimental results show that the wind power system vibration can be used to diagnose the fault of the wind power system, and the system has good results. In the end, this paper makes a comprehensive summary of the research results, and looks forward to the further research on the fault diagnosis technology based on the artificial immune system.

【学位授予单位】：南京航空航天大学
【学位级别】：博士
【学位授予年份】：2014
【分类号】：TM614

【参考文献】