典型耦合优化算法在源项反演中的对比研究

突发大气污染事故中，污染源的快速、准确确定是应急处置的基础.为研究有效的源项评估方法，本文基于美国草原外场SO₂释放实验，利用GA-PSO、GA-NM、PSO-NM 3种耦合算法，分别与高斯点源烟羽扩散模型结合，对源强和位置等污染源参数进行反演与对比，并从算法结构与大气扩散条件方面进行反演效果差异分析.结果表明，从源强反演角度看，PSO-NM反演结果的准确性最高、稳定性最强，平均误差（11.3%）与平均标准偏差（0.7g/s）明显低于GA-NM（16.4%、13.3g/s）与GA-PSO（29.0%、26.6g/s）.从位置反演角度看，PSO-NM的反演结果最为稳定，反演的平均标准偏差（0.29m）明显低于GA-NM（3.20m）与GA-PSO（3.03m）算法；在不稳定和中性扩散条件下，PSO-NM算法的位置反演准确性最高，误差为4.97m；但在稳定扩散条件下，GA-NM的位置反演误差（7.69m）最小.从反演效率角度看，PSO-NM与GA-NM反演时间最短，更适用于污染源的快速确定.

Comparative and study on the application of typical hybrid algorithms in source parameter inversions

Rapid and accurate estimation of source items was the basis for environment emergency disposal on sudden air pollution accidents. In order to search for effective methods for inversing source parameters, we conducted a comparison study on the performances of three hybrid algorithms (e.g., GA-PSO, GA-NM, PSO-NM) for estimating source parameters (strength and location). Three inversion models were developed by combining GA-PSO, GA-NM, PSO-NM with Gaussian dispersion model, respectively. The study was carried out based upon SO₂ leakage tests selected from 1956 Prairie Grass emission experiment. The impacts of algorithm structure and atmospheric diffusion conditions on source term inversion were analyzed. Results showed that for source strength, the PSO-NM algorithm performed more accurate and robust, the mean error and mean standard deviation were 11.3% and 0.7g/s, respectively, which were much lower than those of GA-NM (i.e., 16.4%, 13.3g/s) and GA-PSO (i.e., 29.0%, 26.6g/s). As for source location, the performance of PSO-NM was more robust, with average standard deviation of 0.29m, which was also much lower than that of GA-NM (3.20m) and GA-PSO (3.03m). Under the unstable and neutral atmospheric diffusion conditions, the accuracy of PSO-NM algorithm for estimating position parameter was the best, with an error of 4.97m; However, GA-NM method had the minimum error (7.69m) under the stable condition. As for computational efficiency, PSO-NM and GA-PSO spent less time in source item inversion, which were more suitable for inversing source parameters for sudden air pollution.