Using the dragonfly algorithm to find the optimal location of quality sensor of Lee and Deininger and Second and Third networks of EPANET software to reduce the contaminated water values

Abstract

Accessing reliable quantitative and qualitative drinking water is one of the requirements of human society. Therefore, the water distribution network is designed to distribute reliable drinking water to consumers. In general, the entrance of any unintentional and intentional contamination reduces the reliability of this network. Therefore, it is necessary to monitor the water distribution network using quality sensors to minimize the amount of contaminated water. In this research, three benchmarks of the Lee and Denninger network and the second and third network examples of EPANET software are selected as case studies. Here, by defining different scenarios, the optimal locations of quality sensors are determined using the dragonfly algorithm (DA) and the results are compared with the genetic algorithm (GA). This algorithm is a binary (zero-one) algorithm that has not any hyperparameters leading to reliable results. In addition, this algorithm is not used in the fields of quality monitoring of water distribution networks and therefore it is used here. The results showed the efficiency of the DA to solve this problem. In other words, for the Lee and Deninger network, the amount of contaminated water is reduced by 4.98 %, considering one sensor and consumption pattern one, 26.69 %, considering at least two sensors and consumption pattern one, 27.52 %, considering one sensor and consumption pattern two, and 19.76 %, considering least two sensors and consumption pattern two, using the DA. In addition, for the second network example of EPANET, the amount of contaminated water is reduced by 32.09 %, considering the 24-h consumption pattern and one sensor, and 20.8 %, considering the 24-h consumption pattern and at least two sensors. Finally, for the third network example of EPANET, the amount of contaminated water is reduced by 0.122 %, considering the 24 h consumption pattern and one sensor, and 0.345 %, considering the 24-h consumption pattern and at least two sensors.