An Advanced Computational Approach to System of Systems Analysis & Architecting Using Agent-Based Behavioral Model
Enterprises and System of Systems
Report Number: SERC-2012-TR-021-1
Publication Date: 2012-09-30
Project: Flexible Intelligent Learning Architectures for Systems of Systems (FILA-SoS)
Dr. Cihan Dagli
A major challenge to the successful planning and evolution of an acknowledged System of Systems (SoS) is the current lack of understanding of the impact that the presence or absence of a set of constituent systems has on the overall SoS capability. Since the candidate elements of an SoS are fully functioning, stand-alone Systems in their own right, they have goals and objectives of their own to satisfy, some of which may compete with those of the overarching SoS. These system-level concerns drive decisions to participate (or not) in the SoS. Individual systems typically must be requested to join the SoS construct, and persuaded to interface and cooperate with other Systems to create the “new” capability of the proposed SoS. Current SoS evolution strategies lack a means for modeling the impact of decisions concerning participation or non-participation of any given set of systems on the overall capability of the SoS construct. Without this capability, it is difficult to optimize the SoS design. The goal of this research is to model the evolution of the architecture of an acknowledged SoS that accounts for the ability and willingness of constituent systems to support the SoS capability development. Since DoD Systems of Systems (SoS) development efforts do not typically follow the normal program acquisition process described in DoDI 5000.02, the Wave Model proposed by Dahmann and Rebovich is used as the basis for this research on SoS capability evolution. The Wave Process Model provides a framework for an agent-based modeling methodology, which is used to abstract the non-utopian behavioral aspects of the constituent systems and their interactions with the SoS. In particular, the research focuses on the impact of individual system behavior on the SoS capability and architecture evolution processes. A proof of concept agent-based model (ABM) of the system interactions is developed and integrated with a genetic algorithm (GA) to explore the potential architectural design space, using a fuzzy associative memory (FAM) to evaluate candidate architectures for simulating SoS creation and evolution. The model evaluates the capability of the evolving SoS architecture with respect to four attributes: performance, affordability, flexibility and robustness. The method is applied to an Intelligence Surveillance Reconnaissance (ISR) “acknowledged” SoS as an example domain. The agent-based model represents a System Program Office (SPO) personnel’s interactions with the acknowledged SoS manager, and with the other Systems’ representatives. An agent models each SPO’s decision process and interactions. Since the SoS is comprised of a subset of the available Systems, the participation of each System is modeled as a binary choice in a “chromosome” genetic algorithm representing the possible subsets of participating systems and their interactions with other Systems within the SoS. The genetic algorithm approach allows a more thorough exploration of the architectural “space” composed of all the possible subsets of Systems and interactions than typical, biased, preconceived human selected subsets. Finally, the choice of achievable configurations from the various candidate architectures represented by the genetic algorithm generated chromosomes is made by a rule-based fuzzy associative memory. A proposed method is presented for developing 1) desired attribute membership functions, and 2) rules for combining sub-element values to achieve an overall architecture evaluation that can be ranked for selection.
The model elements are integrated into a toolset that can include the environment in which the agents operate, to select better architectures for successive waves of development. These “waves” of development may coincide with annual funding increments and/or major reviews. Initial integration of the GA and FAM was demonstrated in the ABM framework. Additional effort is planned to improve and modularize the agent models and the interactions among the agents, improve the interfaces among the GA, FAM and ABM components, improve the GA evolution algorithms, add stakeholder participation to creation of the fuzzy evaluations, and to create a better user interface for an executable tool set for future modeling in multiple domains.