Optimizing Systems with Conflicting Objectives for a Limited Resource - Resubmission - 1

Project: Research project

Description

It is proposed to address research on problems that have two inherent difficulties. On the one hand, in a multi-component system different components have conflicting objectives which cannot be optimized simultaneously. On the other hand, in addition these components utilize the same limited resource which further restricts optimizing the overall operation.

A specific such scenario is being investigated in our currently funded research, namely the joint operation of radar and communications (on the same frequency spectrum) where the waveforms to optimize either component are the opposite of each other. It is, however, proposed to use this case only as a starting point and to consider other, potentially even more challenging situations.

To stay in the context of communications, one such system is presented in more detail but others can easily be added. The problem is that of decentralized data fusion in a multi-sensor network. The network is to track a moving target and depending on the network structure only certain sensors can communicate with each other. In addition they tap into a limited resource, their local battery power. As in the first example, the limited frequency spectrum is another challenge.

Problems as described above will be called COLRO problems, Competing Objectives Limited Resource Optimization. These problems in general are nonlinear and not convex. While the exact global optimization even for such problems with discrete or integer constraints has become a reality, it is to be expected that strong heuristics will be competitive, balancing computational effort and distance to the global optimum. Both types of methods will be investigated. The problems will mostly have no integer variables or only very few.

In a static situation the conflicting objectives can only be weighted against each other. In a dynamic situation an overall optimization of the system can at least approximately be achieved when a look-ahead feature is integrated. This will be done based on prior related research by the investigators, adaptive dynamic programming methods.

For highly heterogeneous and large systems the centralized optimization may be prohibitively complex and expensive. A distributed COLRO framework will be developed and be applied to various scenarios such as a fleet of UAVs or of autonomous vehicles.

StatusActive
Effective start/end date1/1/1912/31/21

Funding

  • DOD-USAF-AFRL: Air Force Office of Scientific Research (AFOSR): $344,850.00

Fingerprint

Communication
Data fusion
Global optimization
Unmanned aerial vehicles (UAV)
Dynamic programming
Sensor networks
Radar
Sensors