دانلود رایگان مقاله انگلیسی با استفاده از یک الگوریتم ژنتیک برای بهبود پیش بینی نشت نفت - الزویر 2018

ABSTRACT

The performance of oil spill models is strongly influenced by multiple parameters. In this study, we explored the ability of a genetic algorithm (GA) to determine optimal parameters without the need for time-consuming manual attempts. An evaluation function integrating the percentage of coincidence between the predicted polluted area and the observed spill area was proposed for measuring the performance of a Lagrangian oil particle model. To maximise the objective function, the oil spill was run numerous times with continuously optimised parameters. After many generations, the GA effectively reduced discrepancies between model results and observations of a real oil spill. Subsequent validation indicated that the oil spill model predicted oil slick patterns with reasonable accuracy when equipped with optimal parameters. Furthermore, multiple objective optimisation for observations at different times contributed to better model performance.

Conclusion

An approach combining an oil spill model and the GA technique was presented to enhance prediction accuracy for a real oil spill event. First, a simplified criterion for assessing oil spill model performance was proposed, which was then maximised through a GA. Compared with results from manual methods, oil spill forecasts were better simulated using parameters obtained by the GA. The major conclusions of this study are as follows: (1) The proposed formula related to polluted area overlap ratio remedies the limitations of conventional oil spill models, which lack quantitative evaluation measures. Equipped with versatile weights, the equation was fully applicable to diverse scenarios. (2) The GA could determine suitable parameters that resulted in good performance of the oil spill transport model, which was distributed within a reasonable range rather than simply subjectively determined. To deal with information from multiple moments, i.e. a multi-objective optimisation, we adopted the principle of proximity, whereby data obtained at a later time were given more weight. Recent improvements in oil spill data acquisition ensured that sequential data were available; therefore, this principle is highly practical. (3) Some parameters, such as the wind drift factor and turbulent diffusion coefficient, were sensitive, indicating that they should be calibrated for each specific oil spill event, thereby avoiding an over-dependence on rules representing average conditions. In this way, our method could determine the most suitable parameters according to each location or event of interest.