הנדסת מים

תקצירי כנס הידרו-אינפורמטיקה במים AI יישומי דאטה ו- 50 143 | מגזין המים הישראלי הנדסת מים | FAIR DIVISION WITH STORAGE FOR WATER ALLOCATION We are interested in water allocation mechanisms. To this end, we utilize artificial intelligence and study a novel model of fair division. In our basic setting, there is a central scarce, time-varying source of water that can be stored (possibly partially); and several agents with different, time-varying demand of water. We consider several mechanisms for allocating water at each point in time, study their properties, and design efficient algorithms that achieve these goals for most cases, based on the maximization of utilitarian, egalitarian, or Nash social welfare. In order to show the practical feasibility of our approach, we run our algorithms on real data based on water supply and demand in the western Negev, as well as on artificially-generated data. The simulation results are promising in their quality and suggest (and perhaps confirm) that Nash social welfare is a sweet spot between efficiency and fairness. SPATIO-TEMPORAL MEASUREMENTS OF DIRECTIONAL WATER WAVES POLARIMETRIC SENSING AND MACHINE LEARNING Noam Ginio, Dan Liberzona, Barak Fishbaina and Michael Lindenbaum Accurate measuremnt of water surface waves complex spatio-temporal structure is of high importance for many engineering and environmental protection applications. These include monitoring oil and other contaminants spills and their dispersion on the water surface, dispersion of particle nutrients and pillution by waves induced currents, rivers and man-made discharge jets interaction with the open water bodies, stability of solar panels deployes on water surface of lakes and reservoirs, ating effective waves energy converters, and many others. However no existing technique can provide near-real-time spatio-temporal data of waves and slopes. To address this problem we present a noval accurate and cost-effective measurement methodology for obtaining spatio-temporal distribution of water surface elevation (water waves) and directionality (slopes). Here we utilize Deep Learning (Artificial Neural Nnetworks - ANNs), approach, and latest advances in polarimetric imaging technology, to develope a remote sensing methodology for laboratory implementation. Inferring surface elevation, slope maps and waves’ directional spectra with high accuracy, from polarimetric data of artificial light source reflections from the water surface. Figure 1: UP – Tarining and implementation of ANNs; DOWN – Expereimntal setup The methodology, based on previously published proof of concept1, was further developed to constitite applicable measuremnt tool by improving supervised data collection of larger variety of monochromatic wave trains serving as the treaining sets, achieving higher Signal to Noise Ratio (SNR) in larger spatial sampling area obtained by in house developed artificial light source. In addition, we utilized Bayesian optimization algorithm for hyperparameters tuning for the deep learning on the collected data. We demonstrate the ability of the on deep learning network, trained on the collected simple monochromatic wave trains data, to produce high-resolution and accuracy reconstructions of the 2D water surface slopes of irregular waves fields propagating