(12b) A Novel Biomimetic Approach to Process Control By Exploiting Memory and Cognition

Typical learning algorithms used in the area of process control utilize the ‘experiences’ in the immediate past. However, for uncertain cyclic systems or for systems that ‘revisit’ the same or similar transients due to inherent evolution of the system, occurrence of certain disturbances, and/or desired control trajectory, experiences learnt in the ‘distant’ past can be exploited to achieve superior control performance. The main challenges in utilizing past experiences are in identifying the similarity of the current problem with one or more previous control problem(s), and then in calculating the optimal control moves based on the evaluation of ‘dissimilarity’ measure. In this work, we propose a novel biomimetic control system that mimics human memory and cognition system by drawing from the past experiences to achieve superior performance for the current control problem.

In this work, a dynamic sparse coding (DSC) algorithm is developed to identify the features of a non-linear time-varying system for the purpose of estimating the similarity to previous control scenarios. Each experience is stored in a compact cluster of “memories” by using a subtractive clustering technique for unsupervised learning of unique, or “core”, control problems. Cores are automatically updated to improve future performance based on the new ‘experience’ gathered while solving the current control problem. When there is an acceptable similarity to a past control problem, then the experience from the past controller is delivered in the form of updated controller parameters. Overall, the control system automatically evolves to an optimal system with continued improvement as new ‘experiences’ are gathered.

Recognizing that it is unlikely for a controller to experience all possible forms of control scenarios upfront, or even after long exposure, an algorithm is developed that mimics the human cognitive decision support system. By integrating the memory base with a fuzzy cognitive map (FCM), informed decisions can be made on how to optimally proceed against a large array of un-encountered control scenarios. Overall, it is desired that a seamless independent operation of the control system be achieved.

The algorithms and approaches are validated on an acid gas removal unit as part of an integrated gasification combined cycle plant. This unit consists of large number of equipment items, is highly nonlinear, and involves significant mass and heat integration along with considerable time delay. It is observed that even if the controller performance is poor to start with, repeated use of past experiences leads to superior control performance. Performance of the control system is also studied under various un-encountered control scenarios.