New approach to model-order selection

摘要: A new, direct and practical scheme is proposed for determining the model orders of a system, its signal disturbance using key properties Kalman filter (KF). Unlike conventional methods, it enjoys unique property being both necessary sufficient. The system described by Box–Jenkins model, whose accessible input output are corrupted unknown zero-mean white Gaussian-distributed disturbances measurement noise. outputs asymptotically-stable linear time-invariant systems driven an inaccessible Gaussian noise process, respectively. Predictive analytics used to estimate exploiting smoothness randomness noisy input. signal, models their associated KFs identified various selected minimising KF residuals so that these become processes. model-order corresponds minimum-variance residual. Equivalently, minimum order when number poles or estimates all identical equal to, exceeding minimal order. successfully evaluated shown outperform commonly-used but only sufficient Akaike Information Criterion.