Poster Abstract: Delivering Intelligent Home Energy Management with Autonomous Agents Conor Muldoon and Michael J. O’Grady and David Lillis and Tadhg O’Sullivan and Thomas Holz and Gregory M.P. O’Hare CLARITY: Centre for Sensor Web Technologies, School of Computer Science and Informatics, University College Dublin Abstract—This poster discusses the design and implementation of the decision making/reasoning infrastructure of an intelligent home energy management system that was developed as part of the Autonomic Home Area Network Infrastructure (AUTHEN- TIC) Project. Specifically, the poster focuses on the Agent Factory Micro Edition (AFME) functionality that enables the Home Area Network (HAN) to be managed for two home energy management scenarios representative of this space. The energy management system was tested and deployed in both laboratory and real home settings. I. I NTELLIGENT HOME ENERGY MANAGEMENT In this poster, the decision making infrastructure of an in- telligent home energy management system that was developed using Agent Factory Micro Edition (AFME) [2] is discussed. To drive the development of the home energy management system, two use cases were considered and tested in rooms within the Tyndall National Institute, University College Cork and a real home. The rooms were equipped with temperature, light, contact, and PIR sensors along with pressure mats, which were placed on seats. In the first use case, when the occupant enters the room, their presence is detected and light levels are adjusted to their preferred settings. If the temperature is below a point specified by the occupant, the air heater is switched on. If the temperature is too high, a message is displayed on the occupant’s tablet device to indicate that the room is being cooled. When the occupant sits down in front of the TV, chair occupancy is detected by a pressure mat. The light level is then adjusted and the TV is switched on through the use of a smart plug. When the occupant leaves the chair, light levels are switched back and a warning message is displayed on the TV for a manual switch off. When the occupant leaves the room, both the air heater and light are switched off. At any stage, if the window is opened, it is detected by a contact sensor and a message is displayed on the occupant’s tablet. In the second use case, a weather event is simulated through the use of a Waspmote that can detect water. When the event is detected, a message is displayed on the occupant’s tablet and the water heater is turned on, which is simulated by turning the kettle on through, again the use of a smart plug. The idea is that the system should adapt to occupant behaviour – heating water for a shower when the occupant is on their way home from work and it is raining for instance. When the water reaches the occupant’s preferred temperature, availability is shown to the occupant on the tablet. If the water is not used within a certain time period, an alert message is displayed. If water is repeatedly wasted, the occupant is asked to change their heating schedule. Subsequently, an energy and cost report is displayed on the occupant’s tablet. The reasoning/decision making module to realise the use cases has been implemented with AFME. AFME is a min- imized footprint intelligent agent platform for the rapid de- velopment of Multi-Agent Systems. It is based on the Agent Factory development framework [1], but designed for use with the Java Micro Edition (JME) Constrained Limited Device Configuration (CLDC). Although primarily intended for highly constrained devices, applications developed for JME CLDC can also be used on desktop and server machines. AFME is concerned with the development of computation- ally reflective agents. Computational reflection is a technique that enables a system to maintain meta-information about itself (an agent’s belief set) and to use this information to determine its behaviour. The behaviour of agents in AFME is represented using declarative antecedent-consequence rules, referred to as commitment rules, that determine the conditions under which commitments are adopted and actions are performed. To facilitate this, the conditions are matched against the agents’ belief sets at periodic intervals using resolution-based reason- ing. Resolution-based reasoning is the goal-based querying mechanism that is employed within Prolog interpreters. The reasoning process results in either failure or in a set of bindings being identified that cause commitment rules to be evaluated as true, leading to a number of commitments being adopted and actions being performed. The AUTHENTIC Reasoning Module comprises a set of agents, a set of software actuators, a set of perceptors, and the AUTHENTIC Service. The AUTHENTIC Service is a class that enables agents to interact with SIXTH, which is a Java- based middleware for the Sensor Web [3] that allows sensor- driven applications to be abstracted from the sensors they depend on. It provides a unified interface that enables a variety of sensor types to be integrated along with a standardised way for interacting with them. SIXTH allows the behaviour of sensors and physical actuators to be altered through the use of a re-tasking service. It is a modular framework and facilitates component updates in a distributed manner and without the need for a restart. For example, if the agent code were updated by the developer and the HAN was in operation in a number