Predictive Control of Wind Turbines by Considering Wind Speed Forecasting Techniques Mahinsasa Narayana Northumbria University mahinsasa.narayana@unn.ac.uk Ghanim Putrus Northumbria University ghanim.putrus@unn.ac.uk Milutin Jovanovic Northumbria University milutin.jovanovic@unn.ac.uk Pak Sing Leung Northumbria University ps.leung@unn.ac.uk Abstract- Fixed speed wind turbines have low efficiency as compared to variable-speed, fixed-pitch wind turbines. The latter are required to optimize power output performance without the aerodynamic controls. A wind turbine system is operated such that the points of wind rotor curve and electrical generator curve coincide. In order to obtain maximum power output of a wind turbine generator system, it is necessary to drive the wind rotor at an optimal rotor speed for a particular wind speed. A Maximum Power Point Tracking (MPPT) controller is used for this purpose. In fixed-pitch variable-speed wind turbines, wind-rotor parameters are fixed and the restoring torque of the generator needs to be adjusted to maintain optimum rotor speed at a particular wind speed for optimum power output. In turbulent wind environment, control of variable-speed fixed-pitch wind turbine systems to continuously operate at the maximum power points becomes difficult due to fluctuation of wind speeds. Therefore, a special emphasis is given to operating at maximum aerodynamic power points of the wind rotor. In this study, wind speed forecasting techniques are considered for predictive optimum control system of wind turbines to reduce response time of the MPPT controller. Index Terms-« Maximum power point, Predictive control, Wind speed forecasting , and Wind turbine characteristics are the main factors that determine the optimum operating points. Rotational speed of the system cannot be instantly varied due to the wind rotor momentum of inertia. Therefore, it is difficult to track optimal rotational speeds with wind speed variations. Response time of the controller depends on the turbulence dynamics and affects the performance of the system. Optimum line o 0.. 1 Wind rotor 1+- curves CO, CO2 Rotational speed Fig. 1. Function of MPPT mechanism ofa wind energy convers ion sy stem I. INTRODUCTION II. WIND SPEED FORECASTING Wind speeds continuously varies and although wind rotor is required to drive at an optimal rotor speed for a particular wind speed, wind rotor speed can not be instantaneously changed. Therefore, the response of the wind rotor to wind speed variation affects the performance of the system. Wind speed time-series data typically exhibit autocorrelation, which can be defined as the degree of dependence on preceding values] l] , Autocorrelated time series models are usually used for wind speed prediction. In an autocorrelated wind speed- time series, the value of wind speed in anyone time step is strongly influenced by the values in previous time steps. Therefore, in this study wind speed prediction techniques are applied to improve the response of wind rotor speed variation and energy capture. Time series prediction takes an existing series of data and forecasts the future values. Linear statistical models are commonly used for time series prediction [2, 3]. The use of Neural-networks is also a promising technology in forecasting and can be used to predict time series wind data [4-9]. Neural networks can be classified into dynamic and static categories. Static (feed-forward) networks have no feedback elements and contain no delays; the output is calculated directly from the input through feed-forward connections. In dynamic networks, the output depends not only on the current input to the network, but also on the previous inputs, outputs, or states of the network. Dynamic networks can also be divided into two categories: those that have only feed-forward connections, and those that have feedback, or recurrent connections. Dynamic neural network For optimum operation of wind turbines, if the wind speed is method is more suitable for time series forecasting as it can varied from V} to V 2, the wind rotor rotational speed should be trained to learn sequential or time-varying pattern [10, 11]. be changed from OJ} to OJ], as shown in Figure 1. In systems For this study, a neural network incorporated with a tapped that employ wind speed sensors, the wind sensor provides the delay line with delay from 1 to 5 and five neurons in hidden turbine rotational speed reference to the MPPT controller layer is used for wind speed prediction. This is a straight according to the wind rotor characteristics. This reference is forward dynamic network, which consist of a feed-forward compared with the actual rotational speed of the rotor. Wind network trained by back-propagation with tapped delay line at speed, turbine rotational speed and wind rotor & generator an input. This is part of a dynamic network, called focused 978-0-947649-44-9/09/$26.00 ©2009 IEEE Authorized licensed use limited to: University of Northumbria. Downloaded on March 31,2010 at 11:09:06 EDT from IEEE Xplore. Restrictions apply.