Water distribution networks generate vast amounts of time series data every day through sensors measuring flow, pressure, and demand patterns. However, selecting the right analytical technique for a specific operational problem is often challenging due to the wide range of available methods and varying data characteristics.
Different hydraulic behaviours require fundamentally different approaches. A sudden pressure spike caused by valve closure should not be analysed using the same methods used for long-term demand forecasting or leakage detection. Selecting an inappropriate analytical method can lead to misleading conclusions, poor operational decisions, and ineffective interventions.
To address this challenge, as part of my 2026 company assignment, I developed a decision-making framework designed to help identify the most suitable time series analysis techniques for different water network scenarios. The goal was to bridge the gap between theoretical analytics and practical operational decision-making within the water industry.
Modern water companies rely on data-driven approaches to improve efficiency, detect anomalies, reduce leakage, and optimise network performance. Yet not every analytical method works well for every problem.
For example, forecasting water demand requires different techniques compared to detecting pressure anomalies. Without a structured approach, selecting an appropriate method can become inefficient and inconsistent.
Traditional analytical workflows often begin with choosing a model first and understanding the problem later. In practice, water network data behaves in many ways:
- Sudden transient spikes
- Gradual pressure drift
- Seasonal demand cycles
- Regime changes from operational switching
- Variance instability
- Multivariate hydraulic interactions
- Sensor noise and missing data
Each behaviour requires different diagnostic reasoning.
Therefore, to design the framework, I followed a simple principle:
Start with the operational problem, then diagnose the behaviour, and finally recommend the analytical approach.
The framework follows a hierarchical diagnostic structure:
Level 1: Define the Primary Objective
The user first selects the core analytical problem, such as:
- Forecasting
- Anomaly Detection
- Trend & Seasonality Analysis
- Volatility Modelling
- Structural Break / Regime Change
- Multivariate Relationship Analysis
- Data Quality & Preprocessing
- Transient Spike Detection
- Clustering & Pattern Discovery
Level 2: Behavioural Diagnostics
Each problem branch then asks structured diagnostic questions, for example:
Forecasting
- Is the series stationary?
- Is seasonality present?
- Are multiple seasonal cycles present?
- Is the system nonlinear?
Anomaly Detection
- Is the anomaly abrupt or gradual?
- Is the deviation persistent?
- Is it isolated or repeated?
Transient Detection
- What is the event duration?
- Is the spike high magnitude?
- What is the sampling resolution?
- Are events repeated or isolated?
Level 3: Recommended Analytical Techniques
Based on the diagnostics, the framework recommends suitable time series analysis techniques, including statistical, signal processing, and machine learning approaches, and assesses their suitability for different problem scenarios.
Figure 1 illustrates the framework structure through example operational problems, highlighting the associated data characteristics and diagnostic questions used to guide the recommendation of appropriate time series analysis techniques.
Figure 1 – Time Series Analysis Framework Structure
In line with my area of focus as part of the KTP with Edge Hill University, one particularly important section of the framework focuses on Transient Spikes & Short Events, which are highly relevant in water networks.
Hydraulic transients can occur due to valve operations, pump trips, PRV oscillations, rapid demand changes etc.
As shown in figure 1 the framework analyses:
- Event magnitude
- Duration
- Repetition behaviour
- Sampling resolution
For example, 15-minute pressure data may completely mask short hydraulic shocks that would otherwise be visible in 1-minute or sub-minute measurements.
The framework uses the response to the above diagnostic questions to guide users toward the most appropriate time series analysis techniques, including:
- Peak detection
- Wavelet transforms
- High-pass filtering
- Event frequency analysis
This creates a far more realistic operational diagnostic process compared to applying generic anomaly detection techniques.
To operationalise the framework, the logic was implemented as an interactive application using Streamlit.
The application:
- Guides users through diagnostic questions
- Dynamically adapts based on previous answers
- Recommends analytical techniques
- Provides engineering interpretation
- Structures the full analytical workflow
This interactive decision engine can be further developed into a comprehensive intelligent analytics platform based on operational requirements, data availability, and utility-specific objectives. Additional capabilities such as automatic ADF stationarity testing, automated seasonality detection, automated statistical diagnostics, uploading data directly into the framework, confidence scoring for recommended techniques, and automated reporting can be incorporated to enhance decision support.
Figure 2 presents a snippet of the Streamlit-based interactive application.
Figure 2: Streamlit-based interactive decision engine
Water companies are increasingly data rich. A structured diagnostic framework helps bridge the gap between raw data and actionable operational insight. As companies continue adopting smart network monitoring and higher-frequency sensing, structured decision-support systems like this will become increasingly important for extracting meaningful insight from complex water network data.
This project significantly strengthened my ability to connect data science concepts with practical challenges faced in real-world water distribution networks. Throughout the development of the framework, I gained deeper insight into how different time series characteristics such as seasonality, trends, transient behaviour, and data quality influence the selection of appropriate analytical techniques. One of the most rewarding aspects was designing a structured and intuitive framework capable of simplifying complex analytical decision-making into a clear, practical process that can be understood by both technical and non-technical stakeholders.
The project also enhanced my skills in time series analysis, problem structuring, critical thinking, and analytical method evaluation. In addition, it reinforced the importance of balancing technical accuracy with usability, clarity, and operational practicality when developing data-driven solutions for industry applications. Overall, the assignment provided valuable experience in combining engineering understanding with data analytics to support smarter and more efficient decision-making in the water sector.
