The development of an evolutionary algorithm to predict outcomes in meteorological trends

Title from first page of PDF file (viewed September 9, 2010)Includes bibliographical references (p. 59-60)Evolutionary algorithms are techniques based on patterns found in nature and are useful in analyzing trends in data and in determining estimations of solutions to complex\ud problems where finding an exact solution may be impossible, as is true in the study of weather. Evolution is an optimization process that finds solutions to problems by employing a set of mutation and reproduction rules over a series of generations. A research question is\ud posed. How effectively can an evolutionary algorithm be used to evolve a rule based system that accurately predicts the weather? An evolutionary algorithm programmed in the Java language was created by the investigator to evaluate its usefulness to accurately predict outcomes in meteorological trends. The evolutionary algorithm was implemented on a rule based system that uses nonmonotonic reasoning.\ud The algorithm was then validated on actual measurements of weather data archived by the National Oceanic Atmospheric Administration (NOAA) taken from diverse geographic areas. The algorithm was successfully implemented on the rule based system, and a randomly generated set of rules was evolved over a series of iterations to achieve as much as seventy-eight percent accuracy. The algorithm showed more usefulness when applied to a large set of rules (several hundred) than it did toward small sets of rules (less than 100). It was found that a high level of mutation in the algorithm did not necessarily lead to a higher level of accuracy in prediction. Future work could include integrating other Artificial Intelligence techniques, such as probabilistically determining weights, using fuzzy logic and deploying different fitness functions to fine-tune and increase accuracy. It was concluded that evolutionary algorithms could be potentially very effective in developing weather prediction and forecasting systems