Intelligent Prediction Model for Non-Revenue and Demand of Urban Water: Case Study Beitunia City رسالة ماجستير

Abstract

Water is an incomparable rare and strategic natural resource. It is one of the key elements for life and social development as well. Some people lack access to drinking water as a result of considerable leakages in water networks. Water losses and water supply demands are viewed as one of the most important problems facing the water sector in Palestine. More specifically, the municipalities and water utilities suffer from this problem in a manner that causes disruptions and low-quality water service, in addition to significant financial losses. Therefore, accurate prediction of water losses and supply demands is considered as one of the essential remedies that offer efficient support for water resources. Applying a reliable prediction in urban areas could provide the basis for operational, tactical and strategic decisions for water utilities, which is crucial. The public utilities need to forecast water supply demands for the basic needs of people in addition to the requirements for manufacturing and agriculture, as well as for the development of new water sources. Prior knowledge real causes of water losses and proactive response to damages in water networks treatment could reduce losses, and, more importantly, it may save the financial resources in a manner that will strengthen the water sector. The large difference between the amount of water supplied and water consumed is one of the most important issues affecting water facilities, also known as "non-revenue water" [NRW]. Large amounts of water lost through leaks, non-invoicing to customers, illegal connections, poor water meter performance and inaccurate reading seriously affects the financial viability of water utilities. Thus, prediction of water losses and water demands have become important tools for managing and operating water supply systems. So, it is necessary to provide an approach that will help anticipate water losses and demand using artificial intelligence techniques to ensure a reliable water distribution system and solve the cause of water losses. Our research depends on historical data representing water supplies and consumptions in addition to the real water losses of Beitunia city. The main goal of this research is to explore, investigate and develop AI models that could be more efficiently used in predicting water losses as well as forecasting water demands in Palestine, and, more specifically, for Beitunia city. In this thesis, the work methodology consists of the evaluation of different aspects of the design of predictive neural networks, such as the inclusion of new learning algorithms in different neural networks VI architectures. Each neural network configuration is simulated and its predictions are compared with real data of NRW and demand for water. The obtained results show that the learning algorithm called Levenberg Marquardt which is used to optimize the MLPNNs-LM model has achieved the best scoring metrics when it is compared to another learning algorithm in different ANNs models like (RBFNN-Newrb and GAs-MLPNNs), while ARIMA model was less accurate than other NNs models. This is because the ARIMA model relies on linear data to be accurate. Hence, the municipality of Beitunia can employ an efficient system that will reduce cost as well as best utilize and manage water resources. More importantly, such success will help generalize our model for the municipalities and water utilities.