Continuous-time system identification with fractional models from noisy input/output measurements using third-order cumulants

This paper considers the problem of identifying continuous-time fractional systems from noisy input/output measurements. Firstly, the differentiation orders are fixed and the differential equation coefficients are estimated using and estimator based on Higher-Order Statistics: fractional fourth-order cumulants based least squares. Then, the commensurate order is estimated along with the differential equation coefficients. Under some assumptions on the distributional properties of additive noises and the noise-free input signals, the developed estimator gives consistent results. Hence, the noise-free input signal is assumed to be non Gaussian, whereas the additive noises are assumed to be Gaussian. The performances of the developed algorithm are assessed through a numerical example.