Developing optimal naval tactics is a complex problem domain by definition. A common practice is to simulate the process with possible input variations and use the outcome of its repetitive execution as a means for decision support. One of the challenges is that, for various reasons, stakeholders tend to agree on a fixed level of fidelity and scale, which is very difficult to alter at the later stages of the product life-cycle. Another challenge is the missing integrated effectiveness analysis tools, since the post-experimentation phase is usually presumed to follow a regular data-analysis process. This study proposes an agent based approach to simulation of naval tactics, with integrated effectiveness analysis features.
The agent based design incorporates multiple levels of modeling abstractions. The first level is the computation primitives of the models. Behavior Trees are used for the second level, with expressive power similar to DEVS and FSM. The agency abstraction is the third level, partitioning the problem space into two: the agents with bounded rationality, the environments with which the agents interact. Last level is the closed loop scenario definition required to run the simulation, by using the events fired and consumed by the agents.
The second challenge is attacked via a data centric approach to integrate effectiveness analysis features. The user is guided through analysis steps: selection of performance measures from the output logs of a specific simulation experiment and assigning their interpretation methods, construction of an effectiveness tree using a subset of the performance measures as leaf nodes, defining intermediate nodes to aggregate with predefined methods (including MCDM) and user defined methods, and evaluation of the result using successive queries on the database.
The paper explains how to integrate multiple levels in proposed agent based design, and discusses the integrated tree based effectiveness inference mechanism, for naval tactics simulation.