Direct Representation of the Similarity Measure Based on Deterministic Learning Theory

In traditional rapid recognition of dynamical patterns based on the deterministic learning(DL) method, the approximation of the similarity between two dynamical patterns is usually derived by adding the approximation errors and inherent system dynamics of the two systems. In this paper, we attempt to seek the correlation between the system dynamics in the training phase and the recognition errors in the test phase by using inversion methods. A new dynamical pattern recognition method and error recognition system for discrete-time systems are proposed, which can directly represent the similarity between two dynamic patterns.