Condition Monitoring Of Wind Turbine Components Model Based On Gaussian Process

Wind power is a new emerging energy power generation. After several years of rapid development, wind power industry of China is changing from extensive to precision development stage. During the slow development pace, wind power manufacturers concern to solve the early remaining technical issues. The condition monitoring of wind turbine is one of the key points that needs to solute urgently. This paper applies Gaussian process modeling. It could extract random distribution law of operated data and separate measurement noises effectively. These two features make it suit for wind turbine modeling with large data samples. Wind turbine components monitoring applies modeling and residual analysis ways. Thus it is significant to improve modeling accuracy. The main contents are as follows:1. The modeling data set of wind turbine is large and dimension of covariance matrix is high. It is difficult to directly solve the covariance matrix inverse of Gaussian process. This paper uses Cholesky decomposition to avoid matrix ill, while uses cache matrix to solve the double counting problem. These methods ensure rapid and accurate of Gaussian process modeling.2. The work condition of wind turbine is strong random and intermittent. The optimal solution of Gaussian process may not be globally optimal solution. For this purpose, this paper proposes trust region Gaussian process for monitoring research. The trust region algorithm includes second derivative information. In order to avoid the large computation problems, this paper simplifies calculation of Hessian matrix to improve modeling efficiency and accelerate second-order optimal process.3. The above two improved methods are applied to monitor two objects including gearbox temperature and tower vibration. Gaussian process is constructed by studying operating characteristics of the object and extracting related variables.The residual results are compared with other modeling methods. Results verify the efficient and stable of Gaussian process modeling. Through condition monitoring of tower vibration, Gaussian process is proved to detect and monitor tower failures in real time.