Optimization of IBBCEAS Spectral Retrieval Range Based on Machine Learning and Genetic Algorithm
LING Liu-yi1,3*, HUANG You-rui1,2*, WANG Chen-jun1, HU Ren-zhi3, LI Ang3, XIE Pin-hua3
1. School of Artificial Intelligence, Anhui University of Science and Technology, Huainan 232001, China
2. Anhui Science and Technology University, Fengyang 233100, China
3. Anhui Institute of Optics and Fine Mechanics, Key Laboratory of Environmental Optics & Technology, Chinese Academy of Sciences, Hefei 230031, China
Abstract：Incoherent broadband cavity-enhanced absorption spectroscopy (IBBCEAS) can highly sensitively detect trace gases by using an optical resonator to enhance the absorption path. At present, IBBCEAS mainly uses light-emitting diode (LED) as its incoherent light source. The fitting result of the measured gas concentration with improper spectral retrieval range may have a large deviation if the reflectivity curve of the resonator’s mirror does not match well with the LED radiation spectrum with limited bandwidth. Taking the case of quantitative detection for atmospheric NO2, the influence of retrieval range on NO2 fitting results is analyzed. It is found that the relative fitting error will increase rapidly when the retrieval range is narrowed to a certain extent. In this paper, a method for optimizing retrieval range based on machine learning using RBF neural network and genetic algorithm is proposed in order to minimize the error. 435 sample data are obtained by retrieving NO2 concentrations with various spectral subranges, which are members of 430~480 nm and have different widths and center wavelengths. 80% of the sample data are used to train the RBF neural network, and the rest for the testing network. The nonlinear mapping relationship between input parameters, starting and ending wavelengths of retrieval range, and output parameter, relative fitting error, is obtained by the trained network. The optimal retrieval range is searched using a genetic algorithm, in which starting and ending wavelength of the retrieval range are encoded into an individual, and a population is generated with a number of random individuals. After the evolution of multi-generation populations, the optimal retrieval range is obtained by the genetic algorithm, which uses the output of the RBF neural network, i.e. relative fitting error, as individual fitness. Every population has 100 individuals, and the maximum evolution generation is set to 100. When the populations evolve in the 61st generation, the optimal individual, corresponding to 445.78~479.44 nm of the optimal retrieval range, appears, and the optimal fitness is 3.584%. The NO2 fitting results with the other four typical and the optimal retrieval ranges with the same width are compared. The results show that fitting error, relative fitting error and standard deviation of fitting residual with the optimal retrieval range are lower than those with the other four retrieval ranges. The results demonstrate the feasibility of using machine learning to determine the optimal retrieval range of an IBBCEAS system.
凌六一，黄友锐，王成军，胡仁志，李 昂，谢品华. 机器学习的IBBCEAS光谱反演波段优化[J]. 光谱学与光谱分析, 2021, 41(06): 1869-1873.
LING Liu-yi, HUANG You-rui, WANG Chen-jun, HU Ren-zhi, LI Ang, XIE Pin-hua. Optimization of IBBCEAS Spectral Retrieval Range Based on Machine Learning and Genetic Algorithm. SPECTROSCOPY AND SPECTRAL ANALYSIS, 2021, 41(06): 1869-1873.
[1] Fang B, Zhao W X, Xu X Z, et al. Optics Express, 2017, 25(12): 26910.
[2] Min K E, Washenfelder R A, Dube W P, et al. Atmos. Meas. Tech., 2016, 9(2): 423.
[3] Liang S X, Qin M, Xie P H, et al. Atmos. Meas. Tech., 2019, 12(4): 2499.
[4] DUAN Jun, QIN Min, FANG Wu, et al(段 俊, 秦 敏, 方 武, 等). Acta Physica Sinica(物理学报), 2015, 64(18): 180701.
[5] LIANG Shuai-xi, QIN Min, DUAN Jun, et al(梁帅西, 秦 敏, 段 俊, 等). Acta Physica Sinica(物理学报), 2017, 66(9): 090704.
[6] Yi H M, Wu T, Wang G S, et al. Optics Express, 2016, 24(10): A781.
[7] Duan J, Qin M, Ouyang B, et al. Atmos. Meas. Tech., 2018, 11(7): 4531.
[8] Jordan N, Ye C Z, Ghosh S, et al. Atmos. Meas. Tech., 2019, 12(2): 1277.
[9] Liu J W, Li X, Yang Y M, et al. Analytical Chemistry, 2020, 92(3): 2697.
[10] Washenfelder R A, Attwooda A R, Flores J M, et al. Atmos. Meas. Tech., 2016, 9(1): 41.
[11] Bahrini C, Gregoire A C, Obada D, et al. Optics Laser Technol., 2018, 108(12): 466.
[12] Meng L S, Wang G X, Augustin P, et al. Optics Lett., 2020, 45(7): 1611.
[13] Lamgridge J M, Ball S M, Shillings A J L, et al. Rev. Sci. Instrum., 2008, 79(12): 123110.
[14] LING Liu-yi, WEI Ying, HUANG You-rui, et al(凌六一, 韦 颖, 黄友锐, 等). Spectroscopy and Spectral Analysis(光谱学与光谱分析), 2018, 38(3): 670.
