Evolutionary Strategic Optimisation of a Dynamic Transportation Network by Ms Allina Williamson
Supervised by Dr T Kalganova.
Changing market conditions and rapidly increasing fuel costs have driven an increase in the cost of transportation and therefore, in the need to optimize transportation and supply chain networks. These networks have, however, become increasingly complex and difficult to model; therefore the commercial solutions are becoming increasingly ineffective. This project has developed an evolutionary strategic (ES) algorithm which addresses the shortcomings of the commercial solution, and when applied to a real-life transportation network has proven to outperform the commercial software.
Phase 1: Algorithm Development
The transportation network was optimised towards 3 different business objectives : producing the network with the shortest path, the largest amount of profit, and, one that is the most resilient (least likely to fail).
The results from the shortest path and profit were comparable (with only a difference of 0.0017% in total path length values and 0% in profit), but it was during the resiliency investigation that the limitations of the commercial software first became apparent. While the proposed algorithm was able to produce results for resiliency comparable with other optimization techniques being investigated, the commercial software had no way of modelling resiliency, and therefore was not able to produce a solution at all.
Phase 2: The Dynamic Transportation Network Optimisation
Once the function of the algorithm had been proven on a basic network, it was then tested against the shortcomings of the commercial solution on more complicated and variable transportation networks.
The network was scaled up to one with a cycle value 98 times larger than the original, and a beta value 239 times larger. The stability of the proposed algorithm was proven when this network was solved in just over 3 minutes; when the same network size was fed into the commercial solution, the software failed as the network was too large.
The “energy spike investigation” explored the differences in network layout and profit produced with a sharp increase in energy prices. The commercial software solved this by optimising each energy price individually. At the same time, the proposed algorithm completed the same experiments, the profit projections were of a higher value. The reliability of the proposed algorithm is demonstrated against commercially available solutions and the linear programming approach most commonly used in commercial software.
The “lane game” experiment optimised profits over a one-year period, and considered not only a monthly increase in energy prices, (and therefore transportation cost), but also the discount provided by shipping companies if consistent lanes were utilised over a set period. The proposed algorithm was able to model these real world problems and provide solutions, but the commercial software failed as it is unable to model dynamic and changing networks.
This investigation proposed a conceptually new algorithm capable of optimising transportation networks of a larger size and complexity than is currently published, providing optimal solutions with each run. The results from experiments on dynamic and constantly changing networks are a significant area of research, due to the fact that it is an area with increasing impact on supply chain management, but one that has been overlooked so far in commercial solutions.
Case study: Lego Bike under control by Mr Shing Thant Aung, an undergraduate Electronic Engineering student
Supervised by Dr R Powell
A proper physical model of a bicycle was adapted for the Lego Mindstorm (copyright) bicycle. It was simplified to a fifth order model (five interacting first order differential equations). Linear control theory was used despite it being a non-linear system. The assumption of small perturbations was enough.
Notice that the bike is programmed to lean to the right and as it does so the control system kicks in to maintain stability by steering to the right. This is how bicycles and motorbikes work. The gyroscopic affects are completely negligible owing to the slow speed and light wheels. This shows that bike stability is from the "lean-steer" reaction only. (There was no remote control of any sort in this experiment.)
The Jaguar Land Rover Prize was awarded to Mr Shing Thant Aung for his work on controlling a Bicycle at Brunel Engineers (BE) showcase.
Level three: Balloon mission to the end of space