Model identification using virtual compact mapping model

A new adaptive multi-phase model identification method is proposed. The first phase identification, which is described as a real neuro-identification, is designed for estimating a coarse model, while the second phase identification, described as a virtual neuro-identification, is utilized for determining a fine model. The proposed approach utilizes the well-known multi-layered feed-forward neural network. A remarkable characteristic is that virtual signals are artificially generated and are also added to the real input-outputs measurements. Using the newly generated input-outputs, mapping structure could be easily identified. The complement approach, based on real and virtual model identification, could be utilized as an efficient model identification. Experimental results are provided to show the successful application of the proposed approach.