A decision support system for optimised industrial water management

TL;DR

This work tackles industrial water management under scarcity by introducing a generic Decision Support System (DSS) that couples a network data model with an optimization engine. The optimisation is formulated as a generic Mixed Integer Nonlinear Programming (MINLP) problem, which is linearised into a Mixed Integer Linear Programming (MILP) formulation (via a discretisation parameter $K$) and complemented by a Constraint Programming (CP) alternative; this enables fast, scalable design and operational decisions across diverse networks. The framework is validated on three real-world-inspired case studies (oil refinery and two chemical industries), showing substantial freshwater minimisation (e.g., $-17.6\%$) and high wastewater reuse (nearly $90\%$ in the refinery, with notable reuse gains in chemicals) and highlighting the method’s practical impact for plant management. The approach advances industrial water management by offering a transferable, generic tool that integrates DSS and optimisation, enabling rapid exploration of design alternatives and operational strategies with broad applicability beyond bespoke solutions.

Abstract

Water scarcity and the low quality of wastewater produced in industrial applications present significant challenges, particularly in managing fresh water intake and reusing residual quantities. These issues affect various industries, compelling plant owners and managers to optimise water resources within their process networks. To address this cross-sector business requirement, we propose a Decision Support System (DSS) designed to capture key network components, such as inlet streams, processes, and outlet streams. Data provided to the DSS are exploited by an optimisation module, which supports both network design and operational decisions. This module is coupled with a generic mixed-integer nonlinear programming (MINLP) model, which is linearised into a compact mixed-integer linear programming (MILP) formulation capable of delivering fast optimal solutions across various network designs and input parameterisations. Additionally, a Constraint Programming (CP) approach is incorporated to handle nonlinear expressions through straightforward modeling. This state-of-the-art generalised framework enables broad applicability across a wide range of real-world scenarios, setting it apart from the conventional reliance on customised solutions designed for specific use cases. The proposed framework was tested on 500 synthetic data instances inspired by historical data from three case studies. The obtained results confirm the validity, computational competence and practical impact of our approach both among their operational and network design phases, demonstrating significant improvements over current practices. Notably, the proposed approach achieved a 17.6% reduction in freshwater intake in a chemical industry case and facilitated the reuse of nearly 90% of wastewater in an oil refinery case.

Figures (13)

• Figure 1: Insertion of dummy nodes on the preprocessing stage
• Figure 1: User optimisation configuration under the operational phase
• Figure 2: Use Case Diagram
• Figure 2: User view of the results under the operational phase
• Figure 3: Domain Diagram