Author(s): Haytham Awad; Akira Kawamura; Kenji Jinno
Linked Author(s):
Keywords: Water distribution networks; Motor valve control; Pressure regulation; Leakage; Self-Organizing Maps (SOM); Artificial Neural Networks (ANNs)
Abstract: This paper presents a simple and efficient technique for improving the existing optimal pressure regulation and leakage minimization algorithms for supervisory water distribution networks. With the assistance of Supervisory Control and Data Acquisition (SCADA) we have trained a Self-Organized Map (SOM), an unsupervised Artificial Neural Network (ANN), to classify well regulated pressure cases for the water distribution network based on its actual values of flow meter readings which reflect the real network water demands or consumption. After training the SOM, a simulation step is used to classify the unregulated pressure cases into the different model classes. Based on these classifications the appropriate electrical motor valves setting of the well pressure regulation events are used for the unregulated ones. Regarding that available algorithms deal directly with the pressure regulation problem from an optimization point of view which required a computational time depending on the water network size and the used optimization method and in most cases requires also a network simplification method which is considered as another optimization problem. Using SOM as a pre-optimization method could prevent all errors resulting from applying optimization models, save its computational time and provides us with an on-line pressure regulation method. Computational results for Block 12 of Fukuoka City water distribution network using a short- term data set demonstrate the effectiveness of using SOM as a pre-optimization tool for regulating 74% of events within the target pressure range.
Year: 2005