基于一种改进型遗传算法的源强反算

Source characterization inversion by a modified genetic algorithms

为突破遗传算法(GA)在危险化学品泄漏事故应用中存在早熟收敛等不足,有针对性地引入淘汰者基因库,提高种群多样性,避免算法过早陷入局部极值。同时,借鉴粒子群算法的跟随思想,引入启发信息,强化收敛域内的局部搜索力度,最终整理得到改进型遗传算法(MGA)。统计结果表明,MGA的计算结果更准确,误差适应性更强,可为泄漏事故现场的应急决策提供快速有效的数据支持。

Determination of emission sources characteristics is crucial to emergency decision involving hazardous chemical releases. Dominant methodology for characterizing emission sources is to optimize the cost function coupling receptor data to prior dispersion models with optimization method techniques. The literature suggests that GA is the most widely used direct optimization method for solving emission sources characteristics problems. However, the genetic algorithm has some drawbacks when it is used in engineering practice and may output premature convergence estimations, especially for the highly nonlinear problem. Loser-gene was introduced to increase population diversity and embedded search step takes current best point as searching starting point to deal with GA cataclysm problem. A modified genetic algorithm was worked out finally. A comparison analysis of data error sensitivity was made between the MGA and two confirmed hybrid genetic algorithms. The simulation results show that the MGA has better global extremum search performance and is more robust to data error to meet the needs of emergency rescue.