Groupe d’études et de recherche en analyse des décisions

Coding schemes for limited communication in decentralized resource allocation

Chinwendu Enyioha Chercheur postdoctoral en génie électrique, Harvard University, États-Unis

Efficiently coordinating power allocation in massive power systems, especially in a decentralized way, involves enormous communication between providers and users. Given communication constraints, a critical question to ask is - how can decentralized resource allocation strategies be implemented using limited communication? In this talk, I will present a decentralized power allocation algorithm using one-way communication in which suppliers broadcast a coordinating (price) signal using one bit per iteration and users consume power without relaying their anticipated demand to the supplier. In addition to providing conditions under which users' demand does not exceed suppliers’ capacity (primal feasibility), I will present practical coding schemes, called PF-codes (Primal-Feasible codes), for the one-bit communication that prevents a system overload. Further, I will show that the PF-codes asymptotically converge to the true primal/dual optimal solution, and present simulations demonstrating accuracy of our theoretical analyses.