Two-Dimentional LQG Benchmark Based Performance Assessment For Iterative Learning Control

Iterative learning control (ILC) is a kind of algorithm which aims to the system with repetitive control task. The main concept of ILC is to update the current control signal iteratively by using the previous batch control result, so that the system output will gradually track the given reference. For closed-loop control of batch processes, ILC is an effective strategy.As the control systems are widely applied in industry production and the control performance of the most systems will be degreed over time, monitoring and assessing the control performance become much more important. However, as for the batch processes, the control performance assessment (CPA) for ILC needs more attention.This study proposed a novel two-dimensional (2-D) linear quadratic Gaussian (LQG) benchmark for the CPA of the ILC. Based on the 2-D system theory, the ILC-controlled batch process can be converted to a 2-D system. To assess the performance of the converted 2-D system, the traditional LQG benchmark is extended to a 2-D LQG benchmark and a novel CPA tradeoff surface is further proposed.The system model is required when obtaining the LQG tradeoff surface. However, in the most industry process, the system model is nearly unknown. For the model-unknown system, a novel data-driven CPA method is further proposed. As for the converted 2-D model, a novel 2-D closed-loop subspace identification method is proposed. Based on the identified model, the CPA surface can be obtained and can be used to assess the system performance. Several simulation tests verified the proposed methods.