Evolutionary Computation Research Team: Research Interests
The Research interests of the Evolutionary Computation Research Group include:
- Multi-Objective Genetic Algorithms,
- Genetic Algorithms for Scheduling, and
- Control and Applications of Evolutionary Algorithms.
1. Multiobjective Genetic Algorithms
In practical problems, we often want to optimise more than one measure of performance at once. The measures may conflict with each other, and it can be unsatisfactory to combine them into a single optimisation objective, or reduce them in some way so that only one is optimised. Examples of conflicting objective might include maximising speed and safety in a car, or keeping costs low and quality high in manufacturing.
Classical methods of dealing with these problems work by allocating weights to each of the objectives to indicate their importance in the problem, but this method is very subjective, may over-simplify the behaviour of the objectives, and it is often hard to find weights which can accurately reflect the real-life situation.
By using Evolutionary Computation, parallel optimisation techniques such as Genetic Algorithms mean we can find a set of Pareto-optimal solutions, in which each cost is found so that the whole set is no worse than any other set of solutions. The final decision between the sets of solutions can be made in an informed way, by a decision maker. The use of additional information from the evolutionary process can be incorporated to improve the choice offered to the decision maker.
Example in Manufacturing
Optimisation of a highly constrained factory scheduling problem provides two possible solutions: we can pick the one which produces goods to a high standard, but at a slightly higher cost, or we can choose the solution which keeps within budget, but means a slight reduction in quality. Schedules for both these can be found by the genetic algorithm.
Evolutionary Multiobjective Optimization Mailing List
Comments on Multiobjective Genetic Algorithms in Control Systems
Section 4.2, (Multiobjective Optimization) of the paper 'An Overview of Evolutionary Algorithms for Control Systems Engineering' contains further comments on Multiobjective Genetic Algorithms in Control Systems. This is available for download opposite.
2. Genetic Algorithms for Scheduling
The application of genetic algorithms to scheduling problems has interested many researchers. GAs seem to offer the ability to cope with the difficult search spaces involved in optimising schedules, and have been applied to other combinatorial problems such as the Travelling Salesman Problem with some success.
Research Issues on GAs for scheduling may include:
- GAs for job-shop scheduling theory,
- GAs for practical problem solving in industry,
- problem representation within the genetic algorithm,
- specialised operators for scheduling problems,
- results and effects of GA-based systems working in industry,
- techniques for improving performance.
Related problems in the same area are flow-shop scheduling, crew allocation, timetabling, maintenance scheduling and job batching.
The GA-Sched List
The GA-Sched List home page is a web page for researchers working on the application of genetic algorithms to scheduling problems. It contains people, research groups and other information contributed by the subscribers of the list.
3. Applications of Evolutionary Algorithms
Applications of Evolutionary Algorithms are descibed in 'An Overview of Evolutionary Algorithms for Control Systems Engineering' (available for download opposite).