Stormcast – a Distributed Artificial Intelligence Application in the Weath - Er Domain

at higher layers in our architecture. This means that a monitoring module is replicated at the lowest layer, but controlled at the synthesizing module level. If, for example, one replica of a monitoring module fails, the controlling function in the layer above might still obtain results from the other two replicas when a time-out occurs (Hartvigsen and Johansen, 1990a). A third motivation in our design is reusability, allowing modules of the application to be reused in other applications as in (Johansen, 1988). Other typical applications is found in similar monitoring activities as • registration of traffic in an area in order to avoid traffic jam. This can be done by monitoring of the traffic. • warning of pollution. Sensors in different areas can monitor the current state of pollution. • general monitoring of sensor networks. In StormCast, cooperation among independent nodes has been studied in order to evaluate the proposed architecture for a distributed artificial intelligence application. A modular design, together with the utilization of de facto industrial standards, has appeared to be sufficient for the task of weather forecasting in a distributed artificial intelligence application. In addition, through the use of our simplifying approach, we have avoided problems with global state detections, synchronization, etc., which in turn influences the transparent view the users have of the application (Hartvigsen and Johansen, 1989). This modular design also makes scaling, reusability of software, and fault-tolerance possible. Adding a new domain to StormCast is straightforward, existing code is mainly copied and downloaded to the new domain of interest (Hartvigsen and Johansen, 1990a). Some AI applications normally demanding high-performance mainframes are distributed in its nature born to true parallel processing. Weather forecasting is one computational domain where a complex task can be partitioned in subtasks, distributed on several nodes for parallel execution. Cooperation between these sub-tasks can be done to meet the functional requirements of the application. This cooperation increases network traffic, but as a whole, this multiagent architecture is well suited to a hardware base of interconnected computers such as workstations (Hartvigsen and Johansen, 1990b). As a result, complex AI applications can be constructed for a loosely coupled environment of i.e. workstations in stead of expensive mainframes. Storm-Cast is designed to effectively utilize the aspects of true parallel processing of distributed tasks. This is achieved through a simple, modular design. We believe that this kind of multiagent architecture can be regarded as desirable …