• No results found

Organisers: Jonathan Thompson and Ahmed Kheiri

13/09/2018, 09:00, Room - Bowland SR20 Code: OR60A3401 KEYNOTE: Two Example Optimisation Problems from the World of Education

Dr Rhydian Lewis (Cardiff University)

This talk will consider two distinct combinatorial optimisation problems related to education, namely lecture timetabling and school bus scheduling. In both cases we will see that, by getting to the heart of these problems through the identification of their underlying sub-problems, we can design suitable algorithmic operators that are very useful in the production of high quality solutions. The first problem considered is the post enrolment-based course timetabling problem, which has crossovers with both graph colouring and bipartite matching. For this problem we focus particularly on the issue of solution space connectivity and demonstrate that when this is increased via specialised neighbourhood operators, the quality of the returned solutions is generally enhanced. We also make note of problem instances where our algorithm struggles in comparison to others and will offer some compelling evidence as to why this is so. The second part of this talk will look at the problem of designing real-world school transport schedules. Our problem model extends those previously used by considering some important but hitherto overlooked features such as the splitting and merging of routes, gauging vehicle dwell times, the selection of stopping points, and the minimisation of walking distances. This problem also contains a number of interacting combinatorial sub-problems; in this case, set covering, bin packing, and the vehicle routing problem. As a result, a number of new and necessary algorithmic operators will be discussed that can be used alongside other recognised heuristics. Primarily, the aim of this algorithm is to minimise the number of vehicles used by each school; however, secondary issues concerning journey lengths and walking distances can also be taken into account through the employment of suitable multiobjective techniques. The intention is for this talk to be accessible to all OR enthusiasts, not just those specialising in timetabling and scheduling.

What is the nature of your talk? A mix

Does your talk require prior knowledge of the subject area? A little Is your talk accessible and relevant to practitioners? Very

13/09/2018, 11:00, Room - Bowland SR20 Code: OR60A3587 Dynamic Scheduling Strategies for Continuous Bioprocesses

Mr Folarin Oyebolu, Prof Jürgen Branke (Warwick Business School) and Prof Suzanne Farid (University College London)

The biopharmaceutical industry has been moving from batch processes to semi-continuous manufacturing processes. These continuous bioprocesses are more failure-prone and process

183

failure is more consequential. In addition, the probability of failure is dependent on process run time which generally is determined independent of scheduling considerations. Prior scheduling or planning frameworks are static, deterministic, and almost exclusively consider batch processes. Those that model any continuous processes either do not account for stochasticity or do not intend on optimising facility schedules. This work presents a discrete-event simulation framework that models continuous bioprocesses in a scheduling environment. With this we can utilise dynamic scheduling policies to make operational decisions in a multi- product manufacturing facility and react to changes such as process failure events and uncertain demand. We first adapt different scheduling policies from the stochastic economic lot sizing literature and propose a novel look-ahead scheduling policy. Then, we apply an evolutionary algorithm to tune the policy parameters as well as the process duration. We demonstrate the benefit of parameter tuning and show that the tuned policies perform much better than a policy that estimates parameters based on service level considerations.

What is the nature of your talk? A mix

Does your talk require prior knowledge of the subject area? A little Is your talk accessible and relevant to practitioners? Relevant

13/09/2018, 11:30, Room - Bowland SR20 Code: OR60A3644 Exam Timetabling as a Grouping Problem: A Hyper-Heuristics Approach

Dr Anas Elhag (Lancaster University) and Dr Ender Ozcan (University of Nottingham)

Grouping problems are combinatorial optimization problems that require division of a set of objects into a minimum number of mutually disjoint subsets while simultaneously optimising another additional objective. Many real-world NP-hard problems are grouping problems, such as exam timetabling, graph colouring and data clustering. Each one of these problems has been tackled on its own, and there are many problem-specific solutions for each one of them in the scientific literature. This study presents a generic selection hyper-heuristic search approach, that deals with a single solution at any given decision point during the search process and employs a fixed set of standard reusable low level heuristics especially designed for the grouping problems. The application of standard low level heuristics enables the re- usability of the whole approach with different grouping problem domains with minimal development effort. The performance of different selection hyper-heuristics combining different components, implemented based on the proposed framework is investigated on a range of sample grouping problem domains, including exam timetabling, graph colouring and data clustering domains, and best result obtained in each domain are compared to the previously proposed problem-specific algorithms from the scientific literature. The empirical results show that the proposed approach is sufficiently generic and is able to find high quality solutions that are highly competitive to some previously proposed problem-specific approaches.

What is the nature of your talk? Practical

Does your talk require prior knowledge of the subject area? Some Is your talk accessible and relevant to practitioners? Somewhat

13/09/2018, 12:00, Room - Bowland SR20 Code: OR60A3549 Solving Urban Transit Route Design Problem Using Selection Hyper-Heuristics

Mrs Leena Ahmed andDr Christine Mumford (Cardiff University), Dr Yong Mao and Mr Philipp Heyken (Nottingham University) and Dr Ahmed Kheiri (Lancaster University)

184

The design of routes and schedules for a public transportation system is a hugely challenging problem that faces urban societies. With the increasing congestion and pollution resulting from dependency on private vehicles, it has become important to attract people into public transportations. In this work we address the urban transit network design problem (UTNDP) that deals with the design of efficient routes and schedules for public transit systems. We propose a new approach to solving the route design aspect of this problem based on hyper- heuristics and demonstrate that it is fast, flexible, efficient, and highly adaptable to real-world constraints. The UTNDP is considered a highly complex problem, in which exact methods failed. Therefore most of the recent published research on the UTNDP focused on heuristics and meta-heuristic techniques particularly GAs. However, population-based solutions such as GAs have the disadvantage of requiring long run-times when handling even the modest size networks. Thus, finding alternative methods that reduce run-times and scale to real size networks is a key concern. Another serious issue that has hampered research to date is the lack of public benchmarks with realistic dimensions and constraints. Fortunately, our team has developed techniques to extract realistic instances from publicly available UK data, on which we will be applying our methods. Hyper-heuristics are general search methodologies that work on the space of heuristics, controlling a set of low-level operators to improve a given solution. Previously on this project, hyper-heuristics have been applied on published benchmarks and provided better results than the current state-of-the art with much improved run times. This present work focuses on implementing a more realistic model of the problem, utilising real- world size instances and imposing real-world constraints such as restricted start and end points of routes.

What is the nature of your talk? A mix

Does your talk require prior knowledge of the subject area? A little Is your talk accessible and relevant to practitioners? Very

185