S.J. Iqbal,∗ M.D. Mufti,∗ S.A. Lone,∗ and I. Mushtaq∗∗
SMES, LFC, chopper, intelligent control, non-linear neural networks, ∗ Electrical Engineering Department National Institute of Technology, Hazratbal, Kashmir, India, 190006; e-mail: jvd@rediffmail.com, {muftimd, sadial_14}@yahoo.com ∗∗ ALSTOM Projects India Ltd, e-mail: [email protected] Recommended by Dr. Paulo F. Ribeiro S-function Nomenclature i, j area indicies (1, 2, 3) Δ the deviation f frequency Ptiei tie power out of area i ACE area control error NACE new area control erro
In multi-area power systems with steam reheat constraint and governor dead band nonlinearity, besides maintaining the normal frequency, the control engineer is faced with difficult problem of continuous electromechanical oscillations which the various tie-lines are subjected to whenever there is a change in customer load demand. The superconducting magnetic energy storage system (SMES) being a fast acting device can swallow well these oscillations and help in reducing the frequency and tie-power deviations. For better performance achievement, the use of nonlinear neural adaptive predictive control for active power modulation of SMES is proposed in this paper. Online nonlinear identification of each control area of the power system is performed using a two-layer nonlinear network with tapped delay line (TDL) inputs and then one-step ahead prediction of new area control error (NACE) is used for generating an optimal power command for SMES. The NACE, a newly introduced variable in this paper, comprises of area control error (ACE), a term proportional to derivative of ACE and a term proportional to SMES coil current deviation. The resulting control signal has an anticipatory character and intelligently meets the challenging control objectives. The power conditioning system (PCS) for the SMES comprises of IGBT-based voltage source converter (VSC) and two-quadrant DC chopper. Various components of the hybrid system are modeled and two special blocks – one for SMES unit and its PCS and another for adaptive neural identification, prediction and control, are built by developing S-function code in MATLAB. These blocks are used along with other standard blocks available in SIMULINK-Library for versatile SIMULINK implementation of the proposed scheme and simulation experiments are then carried out to demonstrate the effectiveness of the proposed scheme.
Important Links:
Go Back
