In the first part of the study we explored at which level of electric vehicles, problems on the distribution grid start to occur. We show that at a certain level of plugin hybrid electric vehicles (PHEV) and fully electric vehicles (EV), peak load exceeds the tranformator rating and undervoltage and unbalance exceed the norm. In the second part of the study we show how a market based approach to coordination is able to mitigate these problems. Even in the worst case scenario with 50% PHEV and EV, peak load stays well below the tranformer rating and undervoltage is within the norms.

I will be presenting a paper on this topic in April at the i-sup 2010 conference.

The system consists of two applications; the observer (Wizard of Oz) application running on a normal PC and the user application running on the HTC Dream phone. The observer application is a Java application which enables us to pretend the application on the mobile phone is quite intelligent while actually the observer is doing all the intelligent stuff. The user application was implemented using the Android development platform.

The main part of the application was an “indoor-GPS” system. The phone application shows a map of a part of an office building and displays the position of the user. Because actual GPS is not reliable indoors and not sufficiently accurate, camera’s were used to see the position of a participant. The wizard uses these video images to determine where the participant is and click on a map in the observer application to indicate the current position of the participant. The observer application will then send this position to the phone application using either a wireless or GPRS connection.

I talked to a few participants after they did the experiment and they didn’t realize it was a WOZ experiment. At least for them the application behaved convincingly enough to think that somehow GPS was working indoors.

This example is the result of a simulation in the RoboCupRescue simulation environment and it shows an evolving hierarchical organization over time. Initially, at t=0, the organization consists of 2 layers:


As the simulation continues, the organizational structure starts to shape itself to the workload distribution over the agents. At t=100 the organization is structured as follows:

More and more agents finish their initial task and are then assigned to work for other agents. This results in the following organization structure at t=200

Mutual Adjustment

This is a decentralized organization where agents use mutual adjustment to coordinate the search and rescue effort. When an agent finds a victim, it will send a request for help to the other agents (an agent cannot rescue a victim alone). When another agent receives such a request for cooperation, it can send an offer which includes a time window in which the agent will available to assist in the rescue of that victim. When enough agents have responded to a request, an agent will send a confirmation of the definite time at which the rescue should take place.

Centralized

This is a centralized organization in which one agent (the manager) coordinates all activities between the other agents (the operators). The manager agent will first send instructions to the operators where to search for victims. When an operator agent finds a victim, the location and health state of that victim is sent to the manager agent. The manager agent will then assign the rescue task of this victim to a group of 3 agents.

The examples below show different communication networks that all cover 75% of the grid.

The next step is to develop different types of organizations and start running some exiting experiments!