Comparação entre métodos de estimação paramétrica aplicados ao controle adaptativo por modelo de referência

Abstract

The present work compared the performance of three parametric estimation algorithms, the Recursive Least Squares Method (RLSM), the Gradient Method (GM) and Recursive Least Squares Method with Forgetting Factor (RLSMFF), all applied to the Model Reference Adaptive Control (MRAC). An RLC circuit was used for the simulation and analysis of the three methods. The main contribution of the work was to reproduce and clarify the advantages of the methods discussed. Thus, the specific objectives were: (𝑖) Modeling the plant to be controlled; (𝑖𝑖) Discretize the plant; (𝑖𝑖𝑖) Build the control law; (𝑖𝑣) Implement the identification algorithm; and (𝑣)Validate the results through computer simulations. Based on numerical simulations, the performance of each algorithm and its characteristics were analyzed. The RLSM has an excellent transient regime, its covariance matrix is calculated at each new iteration, so it is a powerful method in the face of high reference signals. The GM has the slowest settling time when compared to the MMQR, it is an effective method when the controller gains are close to the true gains. The RLSMFF uses a forgetting factor to weight the most recent measures with greater weight. By taking into account the characteristics of each algorithm, having prior knowledge about the plant to be controlled, helps in choosing the method that best suits the needs of the designer, thus enhancing the performance of the control system.