基于高斯过程的风电机组部件建模与监测研究

发布时间：2018-01-09 00:12

本文关键词：基于高斯过程的风电机组部件建模与监测研究　出处：《华北电力大学》2015年硕士论文　论文类型：学位论文

【摘要】：风力发电是新能源发电的新兴力量。经过近几年的迅猛发展,我国风电产业正处在由粗放型发展向精密型发展的阶段。在发展速度放缓的过程中,解决发展初期遗留下来的技术问题成为风电制造企业关注的焦点。其中,风电机组的状态监测是亟需解决的关键点之一。本文采用高斯过程(Gaussian process,GP)进行建模分析,由于高斯过程建模既能提取运行数据的随机分布规律,又能有效的分离测量噪声,适合风电机组大数据样本的建模工作。同时风电机组部件监测是通过建模和分析残差方式实现的,因此提高建模精度对监测分析的意义重大。论文的主要研究内容如下：1、由于风电机组建模数据集较大,协方差矩阵维数较高,直接求解高斯过程协方差矩阵逆存在一定的困难。为此采用Cholesky分解法避免矩阵求逆可能存在的矩阵病态,同时采用缓存矩阵解决矩阵求逆重复计算的问题,从而保证高斯过程建模的快速性和准确性。2、风电机组具有强随机性和间歇性工作的特点,对象工况复杂多变,高斯过程优化最优解可能不是全局最优解,为此提出信赖域高斯过程回归方法进行监测研究。同时信赖域的优化算法中包括二阶导数信息,为避免运算量较大造成计算量过大的问题,简化海森矩阵的计算,提高建模效率,加速二阶优化过程。3、将以上高斯过程改进方法应用于两个监测对象,即齿轮箱温度和塔架振动。通过研究监测对象的运行特征,提取与监测对象相关的变量集,构建相应的高斯模型。将残差结果与其他建模方法进行对比,验证了高斯过程建模的高效稳健。同时通过塔架振动的状态监测分析,监测出塔架故障所在,表明高斯过程建模分析能够实时监测塔架故障。
[Abstract]:Wind power is a new force of new energy power generation. With the rapid development in recent years, China's wind power industry is in from extensive development to precision development stage. In the process of slowing down the speed of development, to solve the technical problems of the legacy of the early development of the wind power manufacturing enterprises has become the focus of attention. Among them, the wind turbine state monitoring is one of the key points that need to be solved. This paper uses the Gauss process (Gaussian process, GP) are modeled and analyzed by Gauss process modeling can extract the random distribution of data, and can separate the effective measurement noise, modeling of wind turbine for large data samples. At the same time, wind power the monitoring unit components is achieved through the modeling and analysis of residual method, thus to improve the modeling accuracy of monitoring and analysis of great significance. The main contents of this thesis are as follows: 1, the wind turbine modeling The data set is large, the dimension of the covariance matrix is higher, there are some difficulties in solving inverse Gauss process covariance matrix. Using Cholesky decomposition method to avoid inverse matrix ill conditioned matrix may exist, and the use of cache matrix to solve the inverse matrix problem of repetitive calculation, so as to ensure the accuracy and speed of.2 Gauss process modeling, characteristics of wind power the unit has strong randomness and intermittent work, object condition is complicated, the optimal solution may not be the optimal solution of Gauss process, the trust region Gauss process regression monitoring research. At the same time trust includes two derivative information optimization domain, to avoid a large amount of computation caused by the large amount of calculation. Simplified calculation of Hessian matrix, improve the modeling efficiency, accelerate the optimization process of order two.3, the Gauss process improvement method is applied to the two monitoring The object that the gear box temperature and vibration of the tower. By running the characteristics of monitoring objects, monitoring object extraction and related variables, construct the Gauss model. The residual results and other modeling methods are compared and verified in an efficient and robust modeling of the Gauss process. At the same time through monitoring the tower vibration analysis, monitoring tower the fault that Gauss, the process modeling and real-time monitoring of tower failure.

【学位授予单位】：华北电力大学
【学位级别】：硕士
【学位授予年份】：2015
【分类号】：TM315

【参考文献】