Multi-Agent Control System for a Municipal Water System RADHIKA KOTINA 1 , FRANCISCO P. MATURANA 2 , AND DAN CARNAHAN 2 1 Department of Electrical and Computer Engineering, Cleveland State University, Cleveland – OH, USA 2 Advanced Technology, Rockwell Automation, Mayfield Heights – OH, USA 1 http://www.csuohio.edu, 2 http://www.ra.rockwell.com Abstract. The control system for a municipal water system needs to meet the criteria of maintaining continuity and reliability in the water supply for satisfying the consumer demand while saving the energy costs and maintaining the quality of water. In this paper, we propose an intelligent agent system for a municipal water system. This work shows the benefits of using the agent-based approach in handling different scenarios including the uncertain behavior of the system. A distributed control strategy is implemented and promising results are evaluated in a simulation of a water distribution system. Key-Words: - Agents, distributed control, intelligence, quality, planning, scheduling, demand, cost 1 Introduction Water distribution systems play a vitally important role in preserving and providing a desirable life quality to the public. Water supply, operation cost, and water quality are herein studied with greater attention due to their importance and complexity. There are three key issues in this area: (1) Demand, (2) Energy costs, and (3) Water quality. In the past, much of the effort in the design of water distribution systems had emphasis on the aspect of least cost. Today, there has been a growing awareness that it is equally important to have a public water distribution system possessing high service reliability and also water quality. Demand is a critical aspect affecting the control of these systems. Knowledge of the current and future demand will determine how much water is needed in the tanks and at what time. This in turn provides a time-based control strategy that meets the predicted demand while achieving the cost and quality objectives. There are many ways to predict the demand [1]. A method is to predict the demand using the historical data for the specific period [2]. Energy costs are important aspects affecting the operation of a water system. Optimizing the operation of a pump system in a municipal water system can reduce energy costs and also realize other economic and operational benefits. Theoretical and empirical studies of pump scheduling in various water supply systems suggest that 10% of the annual energy and related costs may be saved by optimizing pump operation [3]. Out of the several techniques deployed for optimizing this scheduling process, the most commonly seen from literature are the genetic algorithms [1][4][5][6] with single objective for minimizing the cost of pumping and with multiple objectives for minimizing the number of pump switches. The latter technique reduces the maintenance cost along with the cost of pumping. Use of dynamic programming for optimizing the pump scheduling claims to reduce the energy costs by 12.5% when compared to a base-level control design [7]. Several other techniques like simulated annealing [8] and fuzzy logic [9] can also be found in literature. In all these techniques, the pump operation is pre-scheduled ahead of time and any unpredicted change and/or perturbation make the network prone to a non-reconfigurable damage. Water quality is affected by the time a parcel of water is retained in a storage tank. New water entering a tank from a reservoir is assumed to have age zero. The cumulative age of the water is a factor that helps define the quality of the water. The aging of water in a tank is primarily a function of water demand, system operating strategy, and the system topology. The average retention time is found to be 1.3 days and the maximum is 3 days [10]. Mixing (or turnover) can be used to decrease the water age. From the survey, all the key issues mentioned above are found to be managed by a single centralized controller. However, relying on a single intelligent controller causes a survivability problem in case of damage to the controller itself. This poses the need for a more efficient, safe and reliable technique for controlling the system while enabling reconfiguration to respond to unpredicted changes. Classical control systems based on feedback techniques generally cannot manage computational complexity, nonlinearity and uncertainty. Complex problems like this can be resolved by using distributed agents, as they can handle combinatorial complexity in real time [11]. Agents can schedule Proceedings of the 5th WSEAS Int. Conf. on Artificial Intelligence, Knowledge Engineering and Data Bases, Madrid, Spain, February 15-17, 2006 (pp464-469)