Computational Optimisation in the Real World (CORW) Session 1

Time and Date: 10:15 - 11:55 on 3rd June 2015

Room: M110

Chair: Timoleon Kipouros

377 A Solution for a Real-time Stochastic Capacitated Vehicle Routing Problem with Time Windows [abstract]
Abstract: Real-time distribution planning presents major difficulties when applied to large problems. Commonly, this planning is associated to the capacitated vehicle routing problem with time windows (CVRPTW), deeply studied in the literature. In this paper we propose an optimization system developed to be integrated with an existing Enterprise Resource Planning (ERP) without causing major disruption to the current distribution process of a company. The proposed system includes: a route optimization module, a module implementing the communications within and to the outside of the system, a non-relational database to provide local storage of information relevant to the optimization procedure, and a cartographic subsystem. The proposed architecture is able to deal with dynamic problems included in the specification of the project, namely: arrival of new orders while already optimizing as well as locking and closing of routes by the system administrator. A back-office graphical interface was also implemented and some results are presented.
Pedro Cardoso, Gabriela Schütz, Andriy Mazayev, Emanuel Ey, Tiago Corrêa
25 Fast Multi-Objective Optimisation of a Micro-Fluidic Device by using Graphics Accelerators [abstract]
Abstract: The development of technology that uses widely available and inexpensive hardware for real-world cases is presented in this work. This is part of a long-term approach to minimise the impact of aviation on the environment and aims to enable the users both from industrial and academic background to design more optimal mixing devices. Here, a Multi-Objective Tabu Search is combined with a flow solver based on the Lattice Boltzmann Method (LBM) so as to optimise and simulate the shape and the flow of a micro-reactor, respectively. Several geometrical arrangements of a micro-reactor are proposed so as to increase the mixing capability of the device while minimising the pressure losses and to investigate related flow features. The computational engineering design process is accelerated by harnessing the high computational power of Graphic Processor Units (GPUs). The ultimate aim is to effectively harvest and harness computing cycles while performing design optimisation studies that can deliver higher quality designs of improved performance within shorter time intervals.
Christos Tsotskas, Timoleon Kipouros, Mark Savill
294 Multi-objective Optimisation of Marine Propellers [abstract]
Abstract: Real world problems have usually multiple objectives. These objective functions are often in conflict, making them highly challenging in terms of determining optimal solutions and analysing solutions obtained. In this work Multi-objective Particle Swarm Optimisation (MOPSO) is employed to optimise the shape of marine propellers for the first time. The two objectives identified are maximising efficiency and minimising cavitation. Several experiments are undertaken to observe and analyse the impacts of structural parameters (shape and number of blades) and operating conditions (RPM) on both objective. The paper also investigates the negative effects of uncertainties in parameters and operating conditions on efficiency and cavitation. Firstly, the results showed that MOPSO is able to find a very accurate and uniformly distributed approximation of the true Pareto optimal front. The analysis of the results also shows that a propeller with 5 or 6 blades operating between 180 and 190 RPM results in the best trade-offs for efficiency and cavitation. Secondly, the simulation results show the significant negative impacts of uncertainties on both objectives.
Seyedali Mirjalili, Andrew Lewis, Seyed Ali Mohammad Mirjalili
502 Distributing Fibre Boards: A Practical Application of Heterogeneous Fleet Vehicle Routing Problem with Time Windows [abstract]
Abstract: The Heterogeneous Fleet Capacitated Vehicle Routing Problem with Time Windows and Three-Dimensional Loading Constraints (3L-HFCVRPTW) combines the aspects of 3D loading, heterogeneous transport with capacity constraints and time windows for deliveries. It is the first formulation that comprises all these aspects and takes its inspiration from a practical problem of distributing daily fibre board deliveries faced by our industry partner. Given the shape of the goods to transport, the delivery vehicles are customised and their loading constraints take a specialised form. This study introduces the problem and its constraints as well as a specialised procedure for loading the boards. The loading module can be called during or after the route optimisation. In this initial work, we apply simple local search procedures to the routing problem to two data sets obtained from our industry partner and subsequently employ the loading module to place the deliveries on the vehicles. Simulated Annealing outperforms Iterated Local Search, suggesting that the routing problem is multimodal, and operators that shift deliveries between routes appear most beneficial.
S Pace, A Turky, I. Moser, A Aleti
679 Performance Comparison of Evolutionary Algorithms for Airfoil Design [abstract]
Abstract: Different evolutionary algorithms, by their very nature, will have different search trajectory characteristics. Understanding these particularly for real world problems gives researchers and practitioners valuable insights into potential problem domains for the various algorithms, as well as an understanding for potential hybridisation. In this study, we examine three evolutionary techniques, namely, multi-objective particle swarm optimisation, extremal optimisation and tabu search. A problem that is to design optimal cross sectional areas of airfoils that maximise lift and minimise drag, is used. The comparison analyses actual parameter values, rather than just objective function values and computational costs. It reveals that the three algorithms favoured various extents of explorations on the different parameters.
Marcus Randall, Tim Rawlins, Andrew Lewis, Timos Kipouros

Computational Optimisation in the Real World (CORW) Session 2

Time and Date: 14:10 - 15:50 on 3rd June 2015

Room: M110

Chair: Andrew Lewis

235 Public service system design by radial formulation with dividing points [abstract]
Abstract: In this paper, we introduce an approximate approach to public service system design making use of a universal IP-solver. The solved problem consists in minimization of the total discomfort of system users, which is usually proportional to the sum of demand-weighted distances between users and the nearest source of provided service. Presented approach is based on radial formulation. The disutility values are estimated by some upper and lower bounds given by so-called dividing points. Deployment of dividing points in uences the solution accuracy. The process of the dividing point deployment is based on the idea that some disutility values can be considered relevant and are expected to obtain in the optimal solution. Hereby, we study various approaches to the relevance with their impact on the accuracy and computational time.
Jaroslav Janacek, Marek Kvet
439 An Improved Cellular Automata Algorithm for Wildfire Spread [abstract]
Abstract: Despite being computationally more efficient than vector based approaches, the use of raster-based techniques for simulating wildfire spread has been limited by the distortions that affect the fire shapes. This work presents a Cellular Automata (CA) approach that is able to mitigate this problem with a redefinition of the spread velocity, where the equations generally used in vector-based approaches are modified by mean of a number of correction factors. A numerical optimization approach is used to find the optimal values for the correction factors. The results are compared to the ones given by two well-known Cellular Automata simulators. According to this work, the proposed approach provides better results, in terms of accuracy, at a comparable computational cost.
Tiziano Ghisu, Bachisio Arca, Grazia Pellizzaro, Pierpaolo Duce
537 I-DCOP: Train Classification Based on an Iterative Process Using Distributed Constraint Optimization [abstract]
Abstract: This paper presents an Iterative process based on Distributed Constraint Optimization (I-DCOP), to solve train classification problems. The input of the I-DCOP is the train classification problem modelled as a DCOP, named Optimization Model for Train Classification (OMTC). The OMTC generates a feasible schedule for a train classification problem defined by the inbound trains, the total of outbound trains and the cars assigned to them. The expected result, named feasible schedule, leads to the correct formation of the outbound trains, based on the order criteria defined. The OMTC minimizes the schedule execution time and the total number of roll-ins (operation executed on cars, sometimes charged by the yards). I-DCOP extends the OMTC including the constraints of limited amount of classification tracks ant their capacity. However, these constraints are included iteratively by adding domain restrictions on the OMTC. Both OMTC and I-DCOP have been measured using scenarios based on real yard data. OMTC has generated optimal and feasible schedules to the scenarios, optimizing the total number of roll-ins. I-DCOP solved more complex scenarios, providing sub-optimal solutions. The experiments have shown that distributed constraint optimization problems can include additional constraints based on interactively defined domain.
Denise Maria Vecino Sato, André Pinz Borges, Peter Márton, Edson E. Scalabrin
622 An Investigation of the Performance Limits of Small, Planar Antennas Using Optimisation [abstract]
Abstract: This paper presents a generalised parametrisation as well as an approach to computational optimisation for small, planar antennas. A history of previous, more limited antenna optimisation techniques is discussed and a new parametrisation introduced in this context. Validation of this new approach against previously developed structures is provided and preliminary results of the optimisation are demonstrated and discussed. For the optimisation, a binary Multi-Objective Particle Swarm Optimisation (MOPSO) is used and several methods for generating a viable initial population are introduced and discussed in the context of practical limitations computational simulations.
Jan Hettenhausen, Andrew Lewis, David Thiel, Morteza Shahpari