Dynamic Planning of System of Systems Architecture Evolution

Zhemei Fang, Daniel DeLaurentisCser 2014 · DOI

TL;DR

This paper tackles the complex challenge of dynamically planning the evolution of System of Systems (SoS) architectures. It introduces an Approximate Dynamic Programming (ADP) approach to make near-optimal, multi-stage decisions for adding, removing, or reconfiguring systems over time, balancing cost, performance, schedule, and risk.

What Problem Does This Paper Address?

Developing and evolving large collections of interconnected systems, known as Systems of Systems (SoS), presents significant programmatic challenges. These SoS are characterized by the operational and managerial independence of their constituent systems, yet their interactions create overarching capabilities. Decisions about how to evolve these architectures—adding, removing, or reconfiguring connections—can lead to cascading failures across operational and developmental dimensions. Traditional planning methods struggle with the inherent complexity, multi-scale decision-making, and vast uncertainties involved in SoS evolution. This paper aims to provide a framework for making these complex, sequential decisions more effectively.

Methods

The core of this paper's contribution lies in its application of Approximate Dynamic Programming (ADP) to the SoS architecture evolution problem. ADP is an umbrella term for techniques that address the "curse of dimensionality" in dynamic programming by approximating future value functions. Specifically, the authors employ:

  • Hierarchical Framework: A multi-time scale approach that separates high-level strategic decisions (e.g., system acquisition over years) from low-level operational decisions (e.g., system deployment over months).
  • Approximate Dynamic Programming (ADP): This is used to formulate the SoS evolution as a dynamic planning problem. ADP helps manage the computational complexity arising from the large number of systems, multi-scale decisions, and uncertainties.
  • Value Function Approximation: The paper mentions using linear value function approximation as a form of parametric model and also discusses the concept of post-decision state variables to simplify the Bellman equation.
  • Naval Warfare Scenario: A simplified Littoral Combat Ship (LCS) module is used as an illustrative example to demonstrate the method.

Datasets

Not discussed in the paper. The paper uses a notional Naval Warfare SoS example based on expert judgments and assumed parameters rather than real-world datasets.

Key Insights and Ideas

  • ADP as a Scalable Solution: The paper highlights ADP's potential to overcome the computational limitations of traditional dynamic programming when dealing with the vast state and decision spaces inherent in SoS evolution.
  • Hierarchical Planning is Crucial: Recognizing the different time scales involved in SoS (strategic vs. operational) is essential for effective planning. The proposed hierarchical framework addresses this by separating these decision levels.
  • Uncertainty is Pervasive: SoS evolution is subject to numerous uncertainties (e.g., technology, budget, market, threat). ADP's ability to handle these uncertainties is a key advantage.
  • Near-Optimal is Often Sufficient: For complex SoS problems, finding a near-optimal solution efficiently is often more practical than striving for a perfect but computationally intractable optimal solution.

Key Results

  • The ADP approach was applied to a simplified Naval Warfare SoS (LCS module) to demonstrate its applicability.
  • The ADP algorithm achieved a performance that stabilized after approximately 50 training iterations.
  • The ADP objective value was found to be about 6% lower than the optimal value obtained through a deterministic comparison. This sub-optimality is presented as a trade-off for computational efficiency.
  • The paper illustrates how ADP can provide sequential decisions for both strategic (architecture alternatives) and operational (system deployment) levels.

Limitations and Future Work

  • Deterministic Attack Assumption: For comparison purposes, incoming attacks were assumed to be deterministic. The authors acknowledge that future work should incorporate more realistic uncertainty in attack scenarios.
  • Simplified Example: The illustrative example is a simplified version of a real-world SoS. The authors suggest exploring more realistic, large-scale SoS concepts to fully demonstrate ADP's advantages.
  • Value Function Approximation: While linear approximation was used, more sophisticated value function approximations will be explored in future work.
  • Algorithmic Advances: Future work will focus on exploring algorithmic advances in ADP to enable efficient incorporation of forward state information, learning potential, and better quantification of uncertainties.

Summary updated

BibTeX
@inproceedings{fang2014dynamic,
  author = {Zhemei Fang and Daniel DeLaurentis},
  title = {{Dynamic Planning of System of Systems Architecture Evolution}},
  booktitle = {Cser},
  year = {2014},
  doi = {10.1016/j.procs.2014.03.055},
}

Loading related content...

Figure