Human-Computer Optimisation for Water Systems Planning and Management (HOWS)

Human-Computer Optimisation for Water Systems Planning and Management (HOWS). It is widely acknowledged that the water and wastewater infrastructure assets, which communities rely upon for health, economy and environmental sustainability, are severely underfunded on a global scale. For example, a funding gap of nearly $55 billion has been identified by the US EPA (ASCE, 2011). In England and Wales, the total estimated capital value of water utility assets is £254.8 billion (Ofwat, 2015), but between 2010 and 2015 only £12.9 billion was allocated for maintaining and replacing assets. Combined with the drive to reduce customers' bills, there will be even more pressure on water companies to find ways to bridge the gap between the available and required finances. As a result of this it is not surprising that optimisation methods have been extensively researched and applied in this area (Maier et al., 2014).

The inability of those methods to include into optimisation 'unquantifiable' or difficult to quantify, yet important considerations, such as user subjective domain knowledge, has contributed to the limited adoption of optimisation in the water industry. Many cognitive and computational challenges accompany the design, planning and management involving complex engineered systems. Water industry infrastructure assets (i.e., water distribution and wastewater networks) are examples of systems that pose severe difficulties to completely automated optimisation methods due to their size, conceptual and computational complexity, non-linear behaviour and often discrete/combinatorial nature. These difficulties have first been articulated by Goulter (1992), who primarily attributed the lack of application of optimisation in water distribution network (WDN) design to the absence of suitable professional software. Although such software is now widely available (e.g., InfoWorks, WaterGems, EPANET, etc.), the lack of user under-standing of capabilities, assumptions and limitations still restricts the use of optimisation by practicing engineers (Walski, 20).

Automatic methods that require a purely quantitative mathematical representation do not leverage human expertise and can only find solutions that are optimal with regard to an invariably over-simplified problem formulation. The focus of the past research in this area has almost exclusively been on algorithmic issues. However, this approach neglects many important human-computer interaction issues that must be addressed to provide practitioners with engineering-intuitive, practical solutions to optimisation problems.

This project will develop new understanding of how engineering design, planning and management of complex water systems can be improved by creating a visual analytics optimisation approach that will integrate human expertise (through 'human in the loop' interactive optimisation), IT infrastructure (cloud/parallel computing) and state-of-the-art optimisation techniques to develop highly optimal, engineering intuitive solutions for the water industry.

The new approach will be extensively tested on problems provided by the UK water industry and will involve practicing engineers and experts in this important problem domain.

 

B. Research Aim and Objectives

B.1 Overall Aim

To develop a new approach for designing and managing improved, near-optimal and engineering-intuitive water systems by incorporating visual analytics, heuristic optimisation and feedback-informed learning.

B.2 Detailed Objectives

  • To develop a flexible application for the intuitive visualisation, simulation and optimisation of WDNs.
  • To develop the application to enable continual learning wherein practicing engineers come to understand the optimisation process through interaction with the software system.
  • In collaboration with key industry representatives from the technology sector (Virtalis, XP Solutions), water (Bristol Water, South West Water) and consulting (SEAMS Ltd) to: (i) create new visual analytics capability for water distribution network design; (ii) learn optimisation heuristics from human interaction and embed them within the approach; (iii) develop real-world performance metrics that characterise ‘engineering intuitive’ solutions.
  • To evaluate these methods on benchmark problems (e.g. Anytown, Exenet – Wang et al., 2015); to establish a relationship between ‘mathematical optimality’ and ‘engineering intuitive’ solutions with the aim to create more ‘engineering intuitive’ objective functions.
  • To use the system to engage water industry experts and to investigate the stages of the learning process, test user intervention strategies and the performance improvements thereof.
  • To experiment with the parameters of the intervention strategies to reduce fatigue and maximise useful human input to the optimisation process.
  • To use the developed system to optimise the design, rehabilitation and operation of real-world water systems derived from the industrial users, e.g., the Exeter network with over 21,000 pipes (see the letter of support from South West Water).

B.3 Timeliness and Novelty

The proposed work is timely for a number of reasons. Although optimisation algorithms for complex engineering problems have improved in many ways over the past three decades, this work has almost solely been focussed on algorithms and has largely ignored practical issues that prevent the uptake of these solutions in the water industry. Optimisation techniques will continue to improve, but the proposed work is necessary to bring more realism to the problem formulation, improved understanding of the optimisation process to the practitioner and therefore more confidence in the techniques to the water industry.

The main novelty of the proposed work is the development of a computational framework that considers water system design and management as an optimisation problem comprising of both quantitative and qualitative/subjective criteria. Further novelty is brought through: (i) the development of visual analytics to allow users to interact with engineering optimisation problems in an intuitive manner; (ii) the direct involvement of industry practitioners in the optimisation process (human-in-the-loop), and the increased understanding of which interaction mechanisms prove most beneficial in engineering design; and (iii) an increased understanding of the development of tools for learning from human interaction in engineering optimisation and the use of cloud/parallel techniques to discover engineering intuitive solutions.

C. Programme and Methodology

C.1 Programme of Work

  1. Create the HOWS framework as a backbone with interfaces for optimisation algorithms, visualisation approaches, heuristics and user intervention.
  2. Develop the new software system to visualise all network assets and information from the optimisation framework (e.g., constraint violations, penalties incurred, costs incurred, decisions made).
  3. Develop a user interface in consultation with industry experts to allow for user interaction with the optimisation algorithm and the WDN at the individual element level.
  4. Develop the interaction engine (see C.2.3 below) to capture and make use of human input in the optimisation.  Create additional functional capability (e.g., new heuristics, constraints, objectives) to optimisation algorithms so they can be embedded into the framework (single, multi and many-objective EA, e.g., NSGA-II, NSGA-III, SSHH hyperheuristic developed by Kheiri & Keedwell, 2015).
  5. Engage industrial experts in the design of heuristics, constraints and objectives to determine engineering intuitive solution characteristics through continual learning and experimentation.
  6. Investigate and develop new parallel computing modules of the HOWS framework.
  7. Experiment with and develop the HOWS approach in collaboration with industrial partners to determine parameter settings on benchmark networks with a view of transferring the methodology to other infrastructure systems, e.g., gas, electricity, transport, etc.
  8. Apply HOWS approach to real-world problems in the water industry [return to step 5, if necessary].

Back to Artificial Intelligence research and applications

Google+