Thesis title: Technical and economic impact of the huge Internet of Things and artificial intelligence in the creation of the intelligent water network
Promoting health and prosperity in all aspects of daily life is a challenge that every
man and woman is called to embrace. As an example, pursuing water safety is essential
for maintaining municipal infrastructure and public health, and improving the quality
of people’s lives. This thesis proposes a computational risk assessment methodology
involving the use of different Artificial Intelligence (AI) techniques to guarantee, above
all, three main concerns: interpretability, data scarcity, and performance. To this aim,
a methodology based on the usage of AI techniques, Fuzzy Logic (FL), and Machine
Learning (ML) is proposed here. To make this methodology concrete, based on the
water domain. Water quality assessment is the first study domain, developed in both
understanding Water Distribution Networks risk assessment, as well as understanding
the main factors affecting the potability of water. The first aspect is dealt with a compu-
tational risk assessment methodology involving the use of Fuzzy Inference System (FIS)
with Monte Carlo Simulation (MCS) to quantify and prioritize operational, environ-
mental, and structural risks in Water Distribution Network (WDN). The methodology
enhances traditional Failures Modes and Effects Analysis (FMEA) by taking linguistic
imprecision in the judgment of the expert into account, replacing deterministic Risk
Priority Number (RPN) with fuzzy-based risk assessment. In addition, a ML pipeline
is built based on data to make predictions about water potability in terms of physic-
ochemical features. Model performances are evaluated by cross-validation, Receiver
Operating Characteristic Curve (ROC) curves, and interpretability metrics like permu-
tation importance and SHapley Additive exPlanations (SHAP) values to determine the
most significant factors affecting water quality and pH, sulfate, and Total Dissolved
Solids (TDS) were the most reliable parameters for water potability. In this thesis, we
also implemented ML models, including advanced ensemble methods and a stacking
meta-model, for the prediction of water stress and classification of water scarcity on a
global scale . This thesis evaluates the technical and economic performance of Intelli-
gent Water Networks (IWNs) by integrating Internet Of Things (IOT) sensors and AI
techniques. Various AI approaches, including ML models and rule-based methods such
as FIS, were applied to monitor water quality and network parameters. Performance
metrics such as accuracy, precision, recall, and F1 score were analyzed, and model in-
terpretability was employed by using tools like SHAP to provide more explanations for
decision-making and network optimization.
Keywords: Fuzzy Logic, Water Quality Assessment, Machine Learning, Internet
of Things, Artificial Intelligence