Implementation of a Multiagent Prototype for WWTP Management

The control and management of a Wastewater Treatment Plant (WWTP) is a complex task that requires the supervision of human experts. To ease this task, Multiagent Systems provide a framework for a more efficiently control in environmental processes. A prototype of multiagent system has been designed for WWTP management. Two subgroups of agents are identified: the interface and the intelligent agents. The first subgroup is designed to interact with the plant and the final users. This group contains the monitoring agent, which is used to store and manage all the sensor data incomes, the actuator agent which acts on the plant regulators, and the user agents which allow users to interact with the multiagent system. There are two intelligent agents: the modeling and the predictive agents. The modeling agent induces the models of the plant that users need. There are two methods to make the models. The first one creates rule-based systems to predict abnormal situations, and the second one identifies states of the plant and creates state-transition diagrams to represent the possible evolutions. The user can choose whether the model is incorporated to the set of interesting models. The modeling agent selects the most accurate model and proposes control actions on the plant. The Predictive agent works with the active model to answer the questions about the plant that are asked by the user. A prototype of the multiagent system has been implemented in Jade, a Java-based platform. In the paper, the agent functionalities are identified as agent behaviors, the communication protocols between agents are displayed, and the performance of the whole system is analyzed with actual data from a WWTP located in Catalonia.