This article presents a real-time data analysis platform to forecast water consumption with Machine-Learning (ML) techniques. The strategy fully relies on a web-oriented architecture to ensure better management and optimized monitoring of water consumption. This monitoring is carried out through a communicating system for collecting data in the form of unevenly spaced time series. The platform is completed by learning capabilities to analyze and forecast water consumption. The analysis consists of checking the data integrity and inconsistency, in looking for missing data, and in detecting abnormal consumption. Forecasting is based on the Long Short-Term Memory (LSTM) and the Back-Propagation Neural Network (BPNN). After evaluation, results show that the ML approaches can predict water consumption without having prior knowledge about the data and the users. The LSTM approach, by being able to grab the long-term dependencies between time steps of water consumption, allows the prediction of the amount of consumed water in the next hour with an error of some liters and the instants of the 5 next consumed liters in some milliseconds.

Water leaks are considered as abnormal consumptions of water. It is a challenge to detect them instantaneously but recent wireless smart water meters offer new opportunities for remote monitoring and controlling of consumptions and leaks. This paper presents a complete solution that consists in collecting and analyzing water consumption data. This non-intrusive solution relies on information extracted from measurements taken on a single and centralized part of a distribution network. We propose a generic approach which is based on a formal representation of water consumption data. It is made of several models and features that are representative of extremely large datasets. We use daily water load curves and define a Minimum Night Flow (MNF) and a Period Without Null Consumption (PWNC). It is therefore easy for a procedure, i.e., an algorithm running in a loop, to deal with this representation in real time for the detection of abnormal consumptions and leaks. Our approach to detect water leaks has been implemented and validated with real consumption data from a smart meter installed in a tertiary building. The maximum daily consumption curve is used as a typical load curve and allows to detect big leaks. Extracting the MNF ad the PWNC from the instantaneous flow rate allows to detect smaller leaks. The discrimination between large and small leaks is based on whether or not the typical water load curve is exceeded.

In functional specification and multi-agents systems context, norms are essential to frame the behavior of agents in order to satisfy the objective of their organizations. These norms can be formulated with deontic logic. We opted for the MOISE+ normative organizational model for structural, functional and deontic modeling. We have shown that this model is not adequate for detecting normative conflicts and risks to be unable to detect them for complex multi-agent systems. In this context, we proposed a transformation towards the DisCSP (distributed constraints satisfaction problems) as an alternative to resolve normative conflicts. We have chosen the SDL normative language in order to express the norms in a formal way so that they will be exploited by the DisCSPs. These made it possible to check the consistency of normative multi-agents systems (NMAS) to satisfy the objectives of their organizations.

L’objectif de cette thèse consiste à étudier le comportement des utilisateurs en tant que consommateurs d’eau et d’électricité. Ceci est réalisé à partir de données collectées par des compteurs intelligents et disponibles sous la forme de séries temporelles d’événements irrégulièrement espacés dans le temps qui représentent les instants de consommation d’un seul point de mesure. Ces données sont également traduites sous la forme de Courbes de charge (Cdc) généralement échantillonnées. Après un pré-traitement, les données servent à la classification supervisée des CdC journalières d’eau avec différents algorithmes de Machine Learning (ML). Puis, une nouvelle approche a été proposée pour déterminer la fonction de densité de probabilité de la consommation d’eau journalière des utilisateurs afin de réduire les données et d’optimiser les ressources informatiques. Ceci a favorisé la mise au point d’un algorithme de détection des petites ainsi que des grandes fuites d’eau. De nombreuses approches de ML supervisées ont été appliquées et testées individuellement pour la classification des appareils électriques selon les caractéristiques harmoniques de courant qu’ils génèrent. Dans un second temps, nous avons proposé plusieurs approches hybrides qui combinent les méthodes précédentes afin d’obtenir des résultats de classification très performants. Enfin, dans le contexte de la prévision des consommations, des modèles hybrides qui combinent des approches déterministes et par apprentissage ont été implémentées et testées permettant de prédire les prochains instants de consommation ou les quantités de consommation prévues dans les futures heures avec des performances bien meilleures.