Self-Optimizing Control: An Old Idea Towards Digital Twins and Parallel Intelligence (DTPI)

The concept of self-optimizing control (SOC) can be classified as a type of learning control (LC) where a system is designed to adjust itself automatically to control an arbitrary dynamic process. Classic feedback control techniques and traditional optimal control techniques are also capable of automatically adjusting a system to a desired target but only if all the a priori information about the controlled process (plant environment) is known and can be described deterministically. Hence, learning control techniques becomes essential for systems where the a priori knowledge is unknown or incompletely known. Given this challenge, researchers through out the years and across discipline have developed and refined two possible approaches to address this challenge. One approach which is more conservative involves using the available known information to design the controller with an acceptable margin of error based on the designer’s experience. The other approach which is akin to learning control involves designing a controller which is capable of estimating the unknown information during its operation and determining an optimal control action based on the estimated information. In this work, we present the different classes of learning control which inspired the concept of self-optimizing control and the trend towards digital twin and parallel intelligence to improve system knowledge acquisition for better control.