Bench Talk

This semester I took Cooperative Control of Multi-Agent Systems (CC). Basically, the idea behind it is to make groups of robots behave like schools of fish. Each robot responds to its neighbor and environment to make the whole group act as one body. There are several advantages to this. One is that here is no “queen bee” to take out that would make all the other robots useless. The loss of any robot, whether through malicious attacks or just link failures, does not result in the loss of any other robots. Also, system expansion is easier. There are many applications where CC is very useful: groups of UAVs, large bus generator fault finding, cell phone networks, etc. Yet most problems do not deal with similar enough systems to make use of cooperative control, such as the control between the wall outlet (AC at 120V) and a laptop (DC at 12V).

For my end-of-semester paper, I summarized all the academic papers that used CC to tie multiple Electric Vehicles (EVs) together to optimize EV charging. If all EVs could be seen as one source by the grid then it would be easier to use them to lower the grid’s peak load thereby lower the cost of electricity. The challenge arises from the fact that all different models of EVs have different storage modules with different charge rates, energy reserves, and power reserves. Each paper described different approaches to solving this problem.

One paper had the car or charging station taken into account and all the individual vehicle variables, including what time the owner was going to need their car, and decided that vehicle’s selling and buying point for electricity. Then that number was published to the other agents, which included the power company, then the all agents adapted their selling and buying points until an a market price for electricity in the area was reached.

Another interesting approach was a cross between CC and Genetic Algorithms (GAs). Specifically, they use an Artificial Immune System algorithm. The Artificially Immune System algorithm allows each geographic area to act as one system while each agent takes on unique properties, just as individual cells in an immune system have unique properties. Each iteration of the GA brings the system closer to optimization while allowing each agent to define its properties and solution space. By continuously talking with similar neighbors, each individual finds their optimal solution faster, which takes into account current and expected neighbor states.

Cooperative Control of Multi-Agent Systems is fairly recently moving from theory to application. As it does so, it opens up a large solution space full of possibilities. Only time and academic papers will eventually show the use of this newborn technology. Personally, I am excited to be part of the learning process.

Sources:

[1] N. Matta, et al. "A cooperative aggregation-based architecture for vehicle-to-grid communications." Global Information Infrastructure Symposium (GIIS), 2011. IEEE, 2011.

[2] D. Q. Oliveira, A. C. Zambroni de Souza, and L. F. N. Delboni. "Optimal plug-in hybrid electric vehicles recharge in distribution power systems." Electric Power Systems Research 98 (2013): 77-85