Batch scheme recursive parameter estimation of continuous-time systems using the modulating functions method (Q1362445)

This paper deals with the modulating functions method, which is a very effective method for parameter estimation of continuous-time nonlinear systems that are linear in their parameters and have the form \[ D^{k_0} \zeta(u,y) =\sum^q_{j=1} \theta_j D^{k_j} \psi_j(u,y) +\theta_{q+1}, \] where \(\theta_j\) are the parameters to be estimated, \(\zeta (u,y)\) and \(\psi_j(u,y)\) are nonlinear function of \(u\) and \(y\), \(D=d/dt\), and \(\theta_{q+1}\) represents a bias term. The modulation is performed as follows \[ \langle \mu_i,D^{k_0} \zeta \rangle =\sum^q_{j=1} \theta_j \langle\mu_i, D^{k_j}\psi \rangle +\theta_{q+1} \langle\mu_i,1 \rangle, \] where the inner product is defined by \(\langle f,g\rangle =\int^T_0 fgdt\), and the modulation functions \(\mu_i(t)\) satisfy \[ D^j\mu_i (0)=0= D^j\mu_i (T),\;j=0,1, \dots,n-1\quad \text{and} \quad n=\max \{k_0,k_1, \dots, k_q\}. \] The paper develops a batch recursive estimation of a continuous-time system using the Pearson-Lee formulation of the modulating functions method. Two algorithms, using rectangular and trapezoidal rules, have been presented in which the estimates are updated as a fixed window is shifted one sampling time forward. These formulas are based on the identities of discrete Fourier transforms along with some inherent commutative properties. The method has been applied to noisy data and knowledge of initial conditions is not required. The second algorithm, using trapezoidal approximation, can yield more accurate estimates of the parameters when the process contains very large jumps. A simulation example with a Van der Pol oscillator is included to investigate the performance of the proposed estimation method.