Computer Architectures for Incremental Learning in Water Management

Abstract

This paper presents an architecture and a platform for processing of water management data in real time. Stakeholders in the domain are faced with the challenge of handling large amounts of incoming sensor data from heterogeneous sources after the digitalization efforts within the sector. Our water management analytical platform (WMAP) is built upon the needs of domain experts (it provides capabilities for offline analysis) and is designed to solve real-world problems (it provides real-time data flow solutions and data-driven predictive analytics) for smart water management. WMAP is expected to contribute significantly to the water management domain, which has not yet acquired the competences to implement extensive data analysis and modeling capabilities in real-world scenarios. The proposed architecture extends existing big data architectures and presents an efficient way of dealing with data-driven modeling in the water management domain. The main improvement is in the speed (online analytics) layer of the architecture, where we introduce heterogeneous data fusion in a set of data streams that provide real-time data-driven modeling and prediction services. Using the proposed architecture, the results illustrate that models built with datasets with richer contextual information and multiple data sources are more accurate and thus more useful.