Some Aspects on the Accuracy of System Identification

Abstract This paper studies accuracy in the usual system identification where continuous-time system is identified in the form of the corresponding discrete-time system. First we introduce condition number as a measure of identification accuracy, or identification worthiness of input signal. We show through numerical experiments that proper sampling period, system order and scaling of output signal can be determined by examining the condition number. Next we adopt the identification accuracy referred to continuous-time system coefficients, compare two identification method based on output error and equation error, and illustrate that the output error method gives more accurate results than the equation error method.