A Study of Wind Turbine Parameter Calibration Using Machine Learning Approaches

Abstract

The inertia and damping coefficients of a wind turbine define how the turbine reacts in a transient state. Unfortunately, the values of these coefficients are not always accurately known due to various reasons. This research uses machine learning techniques to determine these coefficients. By perturbing the values of inertia and damping coefficients, thousands of transient events were generated through simulations. The electrical measurements (real and reactive power) of the transient events were used to train a multilayer perceptron (MLP) network to learn the mapping between these time-series data with the corresponding coefficients. A support vector machine (SVM) based regression method was also used to predict the coefficient values using the same input data, and its performance was compared with the MLP approach. While both methods achieved acceptable results, the MLP method outperformed the SVM method by a large margin. A sensitivity analysis was also conducted to evaluate the impact of measurement noises and the size of the training data on the performance of machine learning based wind turbine parameter calibration.