For all the reasons listed in the previous sections, MAS have been applied to solve several problems in power systems. The following four examples were published in the last decade. Based on the explanations from the previous sections, the reader is encouraged to analyze what choices the designers of the presented MAS have conducted. Other existing applications  also constitute interesting concepts.
15.5.1 Coordinated DC Bus Voltage Control
Lagorse et al. proposed a MAS-based coordinated DC bus voltage control scheme, described in . The objective of this algorithm is to automatically maintain the bus voltage at a constant value, which can be considered as the DC equivalent of frequency control in AC networks: as soon as there is a difference between supply and demand, the voltage level (resp. the frequency) moves away from its reference value and compromises system stability.
The DC bus can be the central part of an islanded network where various loads, sources, and storage units are connected. In the described test case, the microgrid includes a PV source, a battery, supercapacitors, a grid access point (to import/ export energy from/to the distribution system), and an active load, all connected to a common DC bus (see Fig. 15.23).
Each source and storage unit is connected to a power electronic converter controlled by an agent (Fig. 15.24). The agents cooperate to ensure that the bus voltage remains constant, using a token system to select which inverter controls
Communication bus DC bus
the bus voltage. A secondary objective is to minimize energy imported from the grid.
The inverter that has the token regulates the bus voltage while the other inverters are current controlled through PI controllers. The agents coordinate themselves to define two outputs: how each inverter should be controlled (voltage or current), and the corresponding voltage and current references.
For example, if a storage unit holds the token and its state-of-charge reaches its lower limit, the unit is not able to keep controlling the bus voltage anymore and needs to give the token to another agent. It then requests another agent to replace it. This request may or may not be accepted. If it is, then the other agent switches its inverter to voltage-controlled mode, while the agent that had the token switches it to current-controlled mode (Table 15.4). If the request is rejected, then the agent can ask another agent connected to the bus to replace it.
Table 15.4 An idealized discussion between two agents when transferring the token to control the voltage of the DC bus 
Each agent has a similar internal architecture, based on a finite-state machine which gives the agent a different behavior depending on whether it has the token or not. Each agent has its own characteristics derived from the type of hardware it is connected to: source, load, or storage unit. For example, as grid imports have to be minimized, the grid agent never asks to control the voltage, and can only receive requests. On the contrary, supercapacitors agents have the highest priority to regulate the bus voltage, because of their high voltage and very short response time compared to other sources.
The main interest of this voltage control algorithm resides in its decentralized nature, its flexibility and its fault tolerance. Other components can be added and/or deleted without requiring manual adaptation of the system, as long as the system is capable of knowing that these components are available and what are their main characteristics are. On the other hand, this scheme does not optimize how the resources used, which would require a different approach.