Masoud Tabesh; Arash Aghaei; Jalil Abrishami
Abstract
A water distribution network is one of the important parts of infrastructure systems. The efficient management and proactive planning of capital investment of these assets are fundamental for efficient and effective service delivered by water companies. The direct economic costs (i.e. rehabilitation ...
Read More
A water distribution network is one of the important parts of infrastructure systems. The efficient management and proactive planning of capital investment of these assets are fundamental for efficient and effective service delivered by water companies. The direct economic costs (i.e. rehabilitation investment, repair costs, water loss, etc.) as well as indirect costs (i.e. service and traffic interruptions, etc.) related to water pipe bursts are rapidly increasing. The ability to predict burst rate in pipes is an important strategic key in order to optimization of rehabilitation decision in water distribution systems. Most networks suffer from lack of enough and reliable data for bursts and failures. In this study basic variables which influence on pipes burst and burst statistical analysis have been identified and evaluated. Then common methods for burst predicting are discussed. In order to identify logical, useful and understandable patterns of breaks data, a data mining methodology named evolutionary polynomial regression (EPR) is described. Starting from a hybrid evolutionary strategy, EPR searches for patterns in data and returns symbolic expressions/models. This approach is demonstrated through a detailed case study. Required data were collected from the Mashhad Water Company which includes both asset and bursts data recorded for year 1384. The whole database was divided into 8 material/diameter classes (from 64 mm to 300 mm). The resulting models for burst prediction in different zones contain explicitly recognizable independent variables. The expression models confirm that pipe age, diameter and length are the most important variables leading to pipe bursts. Also the effects of pressure on pipe burst prediction were implicitly investigated. It was found that pressure is an important parameter which influences number of breaks in a pipe network.
Masoud Tabesh; Siamak Gousheh; Mohammad Javad Yazdan Panah
Abstract
Short-term water demand modeling plays a key role in urban water resources planning and management. The importance of demand prediction is even greater in countries like Iran with frequent periods of drought. Short-term water demand estimation is useful for planning and management of water and wastewater ...
Read More
Short-term water demand modeling plays a key role in urban water resources planning and management. The importance of demand prediction is even greater in countries like Iran with frequent periods of drought. Short-term water demand estimation is useful for planning and management of water and wastewater facilities such as pump scheduling, control of reservoirs and tanks volume, pressure management and water conservation programs. This helps the network managers to decrease vulnerability of the system and consumers and to increase network reliability.
Exact prediction of short-term water demand which is a function of different complex parameters is difficult, time consuming and even impossible. In this research Artificial Neural Networks method (ANN) with back propagation algorithm is used to develop a model for daily water demand prediction of the city of Tehran (center of Iran) using climate parameters and previous daily water consumption records. Several structures were considered with one and two hidden layers and 0- 50 neurons. Also different linear and nonlinear functions were tested for the layers and finally nonlinear function was chosen for the hidden layers and linear function for the output layer. Furthermore, each year was divided into two parts and different models were constructed for hot and cold months.
About four years data for Tehran daily water consumption was available. The model was calibrated by daily data for two years period. At the learning stage, in order to establish the optimal structure of the ANN model, the case of two hidden layers and seven neurons produced the best results and therefore, it was applied for water demand prediction in this city. Results indicate that ANN models can be used to predict Tehran short-term daily water demand, properly. Model evaluation showed that the correlation coefficients for demand prediction are more than 80% for both learning and testing data and the average of error is only 2%. Comparison of the results from ANN and Fuzzy models showed that the ANN outputs are more accurate. Furthermore, a simple formula was proposed to evaluate daily water demand of Tehran using results of a one layer one neuron ANN model with just two input parameters of the last day temperature and water consumption. The results showed good correlation with the measured values with mean error of less than 3%. It can be concluded that a useful model and a simple formula have been produced as the outcomes of this research which can help Tehran water company decision makers for better operating of the system.
