International Conference on Computer Systems and Technologies - CompSysTech’2006 COMPARISON OF GENETIC ALGORITHMS AND PARTICLE SWARM OPTIMISATION FOR FERMENTATION FEED PROFILE DETERMINATION Karl O. Jones Abstract: In recent years the area of Evolutionary Computation has come into its own. Two of the popular developed approaches are Genetic Algorithms and Particle Swarm Optimisation, both of which are used in optimisation problems. Since the two approaches are supposed to find a solution to a given objective function but employ different strategies and computational effort, it is appropriate to compare their implementation. A study is presented illustrating the performance of both genetic algorithms and particle swarm optimisation, demonstrating their ability to generate a fermentation process feed profile based on a number of objective functions. Results demonstrate how the learning mechanism developed an optimal feed profile which meets the defined criteria. Keywords: Feed profile, fermentation, genetic algorithm, particle swarm optimisation. INTRODUCTION Fermentation processes are associated with the production of yeast, pharmaceuticals, foods and beverages, chemicals, and bulk enzymes. These processes amount to over £1 billion per annum to the UK economy alone, hence there are significant cost incentives for improving the profitability and/or efficiency of the processes by employing modern approaches, such as artificial intelligence techniques. The fermentation of Saccharomyces cerevisiae is especially problematical to adequately model and/or control accurately, owing to its intrinsic time varying and non-linear dynamics. The process should be controlled such that the maximum biomass is produced in the shortest time using the minimum raw materials, such as substrate and oxygen. The cost of the various components of a growth medium can have a significant effect on the overall cost of fermentation processes since they can account for between 38% and 73% of total production costs. The organic carbon source is often the most expensive component. Ratledge [1] has made a detailed analysis of annual price and availability of major carbon substrates. In fed-batch fermentations, the embedded exponential growth pattern requires a corresponding substrate supply. This substrate demand can pose a considerable problem for a non-optimised substrate feed profile. Any excess substrate is a waste of resources, while substrate lack is growth limiting factor. In theory the optimal feed profile is an exponential increase matching the demand from the increasing cell numbers. This type of profile can be difficult to achieve since many industrial fermentation processes do not have sophisticated feed pumps: instead they use pumps which deliver at a fixed rate for a set length of time (that is, quantized levels). It is possible to adapt the feed profile so that a high biomass yield is obtained whilst minimising the total substrate supplied. One approach to formulating a more optimised feed profile is to utilise Artificial Intelligence techniques such as Particle Swarm Optimisation (PSO) or Genetic Algorithms (GA). In the 1950s computer scientists studied evolutionary systems as optimisation tools, introducing the basics of evolutionary computing. Until the 1960s, the field of evolutionary systems was working in parallel with GA research. When they started to interact, a new field of evolutionary programming appeared by introducing new concepts of evolution, selection and mutation. Holland [2] defined the concept of the GA as a metaphor of the Darwinian theory of evolution applied to biology. Implementation of a GA begins with a population of random chromosomes. The algorithm then evaluates these structures and allocates reproductive opportunities such that chromosomes which represent a better solution to the problem are given more chance to “reproduce”. In selecting the best candidates, new fitter offspring are produced and reinserted, and the less fit removed. Like - IIIB.8-1 -