工业园区大气污染物排放总量核算模型优化算法反演性能对比

Comparative study on inversion performance of optimization algorithms for the total emission accounting model of air pollutants in industrial parks

  • 摘要: 基于工业园区污染物排放总量核算模型,以东南沿海某重点石化产业基地VOCs的排放总量核算为例,对比研究了Nelder-Mead单纯形法(NM)、双模退火优化算法(DA)、粒子群优化算法(PSO) 3种优化算法和通过偏差平方和(目标函数1)、对数变换(目标函数2)、双曲余弦变换(目标函数3)构建的3种优化目标函数在不同随机误差强度和失效站点数量下的反演性能。结果表明:PSO算法的反演性能与粒子数量有关,但整体表现不适用于当前条件下的反演优化,NM和DA优化算法具有较好的反演准确性(MARE<30%),但NM优化算法的反演计算效率约为DA优化算法的11~20倍,NM优化算法反演性能最好;3种优化目标函数均适用于当前条件下的反演优化(MARE<30%),其中目标函数1和目标函数3在随机误差强度、失效站点数量较小的情况下反演性能更好,而目标函数2在随机误差强度、失效站点数量相对较大的情况下反演性能更好。

     

    Abstract: Based on the total pollutant emission accounting model of industrial parks and taking the total VOCs emission accounting of a key petrochemical industry park on the southeast coast of China as an example, a comparative study of inversion performances was compared on three optimization algorithms, including Nelder-Mead simplex method (NM), dual annealing optimization algorithm (DA) and particle swarm optimization algorithm (PSO), under different random error intensity and number of failure sites. Besides, the inversion performance of three optimization objective functions constructed by deviation sum of squares (Objective Function 1), logarithm transformation (Objective Function 2) and hyperbolic cosine transformation (Objective Function 3) was compared. The results showed that the inversion performance of PSO was related to the number of particles, but the overall performance implied that it was not suitable for the inversion optimization under the current conditions. NM and DA had better inversion accuracy (MARE<30%), but the inversion computation efficiency of NM was about 11-20 times higher than that of DA, and NM had the best inversion performance. All the three optimization objective functions were suitable for inversion optimization under current conditions (MARE<30%). Objective Function 1 and Function 3 performed better when the random error intensity was small and the number of failure sites was small, while Objective Function 2 performed better when the random error intensity was relatively large and the number of failure sites was relatively large.

     

/

返回文章
返回