KnowledgeScape is an object-oriented real-time expert control system for the process industries that has built in adaptive modeling and optimization capabilities. The primary use of KnowledgeScape is the on-line continuous monitoring of plant performance and the calculation of new process set points that maintain and optimize performance as feed and operating conditions vary.

Conceptually, KnowledgeScape is designed around the idea of intelligent software objects that represent real-world plants and processing equipment. These software objects can be connected together representing flow of material from one object to the next or they can be placed within each other representing the concept of groups of equipment forming a circuit. These concepts are so powerful and unique that they have been patented (United States Patent Number 6,112,120)

One element of intelligence is the ability to predict one’s future. This concept is embodied within KnowledgeScape via adaptive on-line neural networks. Reasoning about the past to determine process set points that will better drive the plant to more desired levels is accomplished by a real-time expert system that allows knowledge to be recorded as both crisp and fuzzy rules. Rule based control is centered around the concept of heuristics which are rules of thumb that have been developed over time and represent generalized experiences that can be relied upon in the future.

A subtle and infrequently discussed fact is that the rules themselves represent an inverse model of the process. This leads to the question of what other types of models can be created for the processes and what could be done with them if they are more robust that the heuristic rules. KnowledgeScape uses on-line neural network models to adaptively model the process, which adds to the unique features of KnowledgeScape. The predicted future values of important process variables can be used in the rule system and additionally by a genetic algorithm component of KnowledgeScape to ask the models what would be the best new process set points to achieve the desired control objectives.

This model based optimization ability adds true artificial intelligence to KnowledgeScape in that it creates the basis for learning about the process. By enlarge, rule based expert control are quite static in nature. Knowledge about the process is contained within the rules and represents the knowledge of those who wrote the rules. Models that continually adapt to accurately represent the process as feed conditions and equipment conditions changes add learning to KnowledgeScape.

Crisp and fuzzy rules, adaptive models and genetic optimization all require substantial computing resources, especially if they are on-line and are being used simultaneously. KnowledgeScape uses a distributed hardware architecture that allows computers to be added at will to the control system to meet the computational needs as a systems grows.