Teamwork without talking: distributed system estimates maintain approximately centralized control of smart urban...

Real-time control of urban drainage systems (UDS) can reduce flood risk while enhancing water quality. However, internet-connected valves and weirs may become a liability during communications outages. Given that severe storms often cause both flooding and communications failures, it is critical that smart systems cope well with communications outages during these events. We propose a method to maintain approximately centralized predictive control of a UDS under intermittent communication by hosting approximations of the total system's dynamics and states on each networked microcontroller. This allows individual microcontrollers to act from an imperfect understanding of the total system, even when they have not communicated with the central server for several days. We show robustness to communications outages; maintaining centralized control in this case study when reporting only once every 7 days on average. We also examine the role of relative position within the network in determining the accuracy of each microcontroller's estimates of depths in the other storage basins. This investigation supports the idea that agents within a distributed control problem can make model-based inferences of total system conditions using their local measurements.