Having an integrated plan for the water distribution system (WDS) is a precondition to guarantee the safety of supply. Therefore, the development of such a plan can support the needs of WDS. Both integrated plans require highly accurate estimated values since they should not only cover failures in the system but also meet customers’ demand. Relevant literature lists many mathematical and empirical models for modeling water consumption in the recent decade. Most of the methods in literature are on low-resolution monthly scales because of the limitations in data availability. Besides, increasing the resolution of data would decrease the errors of forecasting models. This study suggested techniques to model high resolution temporal scale and improvement of the models. To analyses the nature of data, average mutual information, phase space reconstruction (PSR), and correlation dimension; where all these methods provide identification of chaotic behavior. The results showed the existence of chaos in the data in different temporal scales with the value of 3.4. Then, non-linear local approximation method (NLA), artificial neural networks, genetic expression programming, and multiple linear regressions techniques are applied for short-term (daily) and mid-term (monthly) forecasting. NLA showed the best performance with the accuracy of 98% for the test period among all models. Then, the highly accurate model of each techniques, were selected to be combined with PSR and wavelet decomposition pre-processing to improve the accuracy of the models. The results presented 1.2% improvement by PSR techniques. For transition of different resolution, this study employed the continuous random cascade method to transfer the resolution. Random Cascade gave the daily value out of monthly scale by 90% accuracy. Finally, using expected consumption values resulted by forecasting models, an approach to make the leakage detection procedure resource-efficient is to estimate the timeline of leakage event within the target district. The analysis of the value of consumption in two scenarios that were simulated in the target district related to consumption was provided using a detection classifier to investigate anomalies in the consumption patterns. This method detected the approximate time for the defined artificial leakage within the test case.