Input signal reconstruction based on improved Moving Least Squares for nonlinear multiple-input multiple-output sensor

Meshless methods popularized in recent years are attractive choices for solving discontinuous and large deformation problems. As one of the most popular methods to form trial function, moving least squares (MLS) can accurately fulfill input signal reconstruction of nonlinear multiple-input multiple-output sensor. However, the parameter matrix obtained from MLS approximation sometimes is ill-conditioned even singular, which makes the signal estimation incorrect. By considering this problem, a novel method, the improved moving least squares (IMLS) is applied to data reconstruction in this paper. The algebra system based on IMLS method is not ill-conditioned with the weighted orthogonal functions replaced as the basis functions. Furthermore the estimation of sensor input signals can be obtained without calculating the inversions of any matrices, and the computing procedure is also faster than that of MLS method. At last the comparison of approximation accuracy between these two methods is presented and illustrates that IMLS is more superior in signals regression for nonlinear multiple-input multiple-output sensors.