Memory-based Control of Nonlinear Dynamic Systems Part I - Design and Analysis

This paper presents a control design method based on short-term memory concepts. The main idea is to make use of certain memorized information such as current system response, previous system response and past control experience to build control action. Fundamentally, the desired control signal in the scheme is "learned" and generated from observing and processing the most recent experience stored in a memory. System performance can be continuously improved during system operation without requiring detail system dynamics. There is no need to repeatedly run the system with the same task. Another advantage of this approach is that the overall required memory space does not grow with time and is much smaller than most existing learning control methods. The first part of the paper addresses the detail design and analysis of the proposed method. The second part of the paper presents the application results of the method to numerous examples/systems