Performance of Typical Optimization Algorithms on Inversing Multi-scene Source Parameters

Determination of release source parameters of hazardous pollutants is an important basis for adequate disposal on sudden air pollution accidents. The parameters (source strength, location, etc.) needed to be estimated for different accidents (such as leakage and fire) are different. It is of great significance to study and compare the estimation performance of different optimization algorithms for fast and accurate estimation of source parameters in actual accidents. Based on the SO₂ release experimental data of 1956 Prairie Grass field experiments, a comparative evaluation study on the application of typical optimization algorithms, including genetic algorithm (GA), particle swarm optimization (PSO) and particle swarm optimization-simplex coupling algorithm (PSO-NM), in source parameter inversion under different scenarios of unknown parameters was carried out in this paper. The inversion accuracy, stability and time cost of different optimization algorithms were analyzed and discussed. Results show that the inversion performances of various optimization algorithms are significantly distinct when inversing different emission parameters. In the case of single-parameter inversion (only the source strength is unknown), GA and PSO-NM have similar accuracy with absolute value of relative deviation of 24.0%, better than that of PSO algorithm (37.6%); and all the three algorithms have better inversion stability with coefficient of variation of less than 0.004. In the case of multi-parameter inversion (source strength Q and position x, y and z), PSO-NM has the best and GA has the worst inversion accuracy, however, the performance of robustness is opposite. Additionally, the increase of the number of inversion parameters obviously affects the inversion robustness of different algorithms. The coefficients of variation for source strength inversion when inversing the four parameters are higher than those of single-parameter inversion by 0.50, 0.12 and 0.29 for PSO, GA and PSO-NM, respectively. For the time cost, the PSO algorithm performs best; PSO-NM algorithm shows a significant increase in calculation time cost under stable atmospheric conditions.