Pitch Control of Variable Speed Wind Turbine using Soft Computing Techniques

Abstract

Wind, solar, geothermal, and wave energy are examples of renewable energy sources that not only meet the electricity demand but also help to minimize global warming. In terms of cost-effectiveness, power requirements, operation, and performance, wind energy is today's most widely used renewable energy source. There are many challenges in wind turbines, such as wind fluctuation, unpredictable nature of wind, pitch control, generator speed control, location limitation, etc. Pitch control is the major challenge in wind turbines. The pitch angle controller adjusts the generator output power when the wind speed exceeds the rated wind speed. The pitch angle controller additionally stabilizes the rotor speed during transient disturbances. This project seeks to develop several soft computing techniques for variable speed wind turbine pitch control. Different techniques like Fuzzy Logic Controller (FLC), Neural Network (NN), Adaptive Neuro Fuzzy Inference System (ANFIS), and Recurrent Neural Network (RNN) are implemented on MATLAB and their results like mean square error and root mean square error are compared with each other. The goal is to use MATLAB/Simulink software to simulate controllers to manage the wind turbine blade pitch angle and keep the output power stable at the rated value. Performances of control are assessed and compared using the steady-state time of output power obtained from the simulation results as well as steady-state errors. In this study, the data is collected for National Renewable Energy Laboratory (NREL) 5 MW wind turbine. Wind speed, TSR, and power coefficient are the inputs and pitch angle is the output. There are 1361 samples of collected data in total. The overall sample size (951 samples) is 70% for the training data and 30% for the testing data (410 samples). The RNN is proposed to control the pitch angle of five MW VSWT .It is evaluated, that the RNN delivers the best results. The RNN models perform better in pitch controller because their dynamical memory integrates delay parameters. The RNN gave the better results other than three models. The MSE and RMSE given by RNN model having 15 neurons in one hidden layers for 1000 epochs are 3.28e-11 and 5.54e-06, respectively.