Wind speed estimation based control of Stand-Alone DOIG for wind energy conversion system

Abstract

A sensor less wind speed estimation scheme for variable-speed wind turbine generators has been analysed in this paper. Neural network principles are applied for sensor less wind speed estimation. Model of one pitch controlled horizontal axis wind turbine along with DOIG based generation system has been used for this study. The aerodynamic characteristics of the wind turbine are approximated by a radial basis function network based nonlinear input-output mapping. Based on this mapping, the wind speed is estimated from the measured turbine mechanical power, turbine angular speed and pitch angle. The resulting WTG system efficiently and reliably estimates the wind speed without any mechanical anemometers. Keywords-Wind energy conversion system (WECS), Sensor less control, Artificial neural network(ANN),Rradial basis function network (RBFN), Wind Turbine generator (WTG).

A radial basis function neural network approach to filtering stochastic wind speed data

“…Much of the development of this method has remained more or less the same at a basic level and the descriptions given in the following sessions are almost identical to those given in Powell (1987). When it comes to wind turbines (Kaur et al, 2014;Qiao et al, 2008;Tian et al, 2011), use an RBFNN for approximating highly non-linear dynamics of a turbine itself (Qiao et al, 2008). uses input-output mapping algorithms to train a RBFNN to obtain the wind speed by approximating the inverse of the function…”

“…In this way, bypassing the dynamics of the turbine to obtain the wind speed is a favorable outcome. Kaur et al (2014) utilizes the same approach and covers the entire operating range of the turbine, and obtains results with almost negligible error. However, again, bypassing the turbine dynamics is not present in this approach.…”

A Radial Basis Function Neural Network Approach to Filtering Stochastic Wind Speed Data

“…Much of the development of this method has remained more or less the same at a basic level and the descriptions given in the following sessions are almost identical to those given in [2]. When it comes to wind turbines, 207-2 [4] [5][6] use an RBFNN for approximating highly non-linear dynamics of a turbine itself. [4] uses input-output mapping algorithms to train a RBFNN to obtain the wind speed by approximating the inverse of the function…”

“…. [5] utilizes the same approach and covers the entire operating range of the turbine and obtains results with almost negligible error. [6] has identical development and error as [4].…”

“…With each element in the x vector propagating through out the entire network one at a time, the equation above can be rearranged to a matrix form as shown below, assuming c elements in the x vector. However, obtaining the weights means that the network output must be as close as possible to the training output: (5) As stated before, the left-most matrix here is known as the interpolation matrix. Reducing it to a compact form:…”