Similarity-Based Control: Bettering Operation of Iterative Learning Under Noisy I/O Data

This paper targets at establishing a similarity-based control framework between heterogeneous systems to better the operation of iterative learning control (ILC) by proposing a data-based design approach in the presence of noisy input-output (I/O) data. Owing to the absence of model information, appropriate I/O tests are designed to guarantee the data sufficiency, based on which the admissible behavior of the controlled system can be estimated. Moreover, the I/O data-based similarity and similarity indexes are presented to measure how close the admissible behaviors of heterogeneous systems are. Thanks to the similarity indexes and by employing an experience projection mechanism, a similarity-based control approach enabling the controlled system to learn from a heterogeneous ILC system is developed using only the I/O data. It is shown that the controlled system can accomplish the tracking tasks by learning from the successful control experience from a similar heterogeneous ILC system, thereby eliminating the necessity of trial-and-error processes. A rigorous performance analysis of the resulting learning error is performed. Simulations are also provided to verify the effectiveness of the similarity-based control.