This study presents a high-performance wind generation system with induction machine (IM), specifically devised with the target of maximising the efficiency of the electromechanical conversion, and contemporary minimising the number of the system sensors and their cost. To this aim, the control system has been integrated, from one side, with an intelligent maximum power point tracking (MPPT) technique, so to make the generator track the power available in the wind, from the other side with techniques for the minimisation of the electrical losses (ELMT). Particularly, the power converters' switching losses have been reduced adopting a discontinuous pulsewidth modulation, while the IM overall losses have been reduced by a suitable electric losses minimisation technique. Contemporary, to reduce costs and increase the reliability of the system, the system has been devised as a fully sensors-less generation unit, meaning that both the wind speed and the machine speed sensors are not present. The anemometer has been substituted by the wind speed estimator integrated in the MPPT, based on the growing neural gas (GNG) network. The encoder has been substituted with an intelligent IM speed estimator, the so called MCA EXIN + reduced order observer (ROO). The performance of the adopted technique has been verified experimentally on a suitably devised test set-up.
IntroductionNowadays, wind energy is becoming an ever more exploited renewable energy source [1,2]. Modern solutions range from permanent magnet synchronous machines, to doubly-fed induction machine or squirrel cage induction machine (IM), each supplied by a suitable power converter [3,4]. Whatever is the machine chosen, from the control point of view, the most challenging issues to be solved are described in the following. First, suitable maximum power point techniques (MPPT) should be developed, able to quickly track any variation of the wind speed, and correspondingly of its power, inside the variable speed range of the wind turbine, avoiding perturb and observer (P&O) [5][6][7] or hill and climb (H&C) algorithms, inevitably requiring very long times to converge in systems characterised by high inertias, as wind generators typically are, and causing persistent undesired oscillations around the MPPs. Even other methods, for example those exploiting proper wind turbine modelling [8][9][10][11] to compute the optimal power reference speed of the drive, can be problematic since they often require the knowledge of a wind speed sensor to work, whose presence can significantly increase the cost and reduce the reliability of the drive system, Some intelligent neural based MPPTs have been proposed in [12,13], respectively, adopting a two radial basis function NNs and a multi-layer perceptron trained by a backpropagation (BPN) algorithm as a virtual anemometer. Both these approaches present the following significant drawbacks: (1) with such approaches, no evidence of the correct learning of inverse characteristic of the wind turbine exists and, (2) a very complex and compu...