Public Foundational Courses Department, Nanjing Vocational University of Industry Technology, Nanjing, 210023, China.
Organization Department, Nanjing Vocational University of Industry Technology, Nanjing, 210023, China.
Sci Rep. 2021 Oct 27;11(1):21173. doi: 10.1038/s41598-021-00804-7.
Studies have shown that there is a certain correlation between air pollution and various human diseases, especially lung diseases, so it is very meaningful to monitor the concentration of pollutants in the air. Compared with the national air quality monitoring station (national control point), the micro air quality detector has the advantage that it can monitor the concentration of pollutants in real time and grid, but its measurement accuracy needs to be improved. This paper proposes a model combining the least absolute selection and shrinkage operator (LASSO) regression and nonlinear autoregressive models with exogenous inputs (NARX) to calibrate the data measured by the micro air quality detector. Before establishing the LASSO-NARX model, correlation analysis is used to test whether the correlation between the concentration of air pollutants and its influencing factors is significant, and to find out the main factors that affect the concentration of pollutants. Due to the multicollinearity between various influencing factors, LASSO regression is used to further screen the influencing factors and give the quantitative relationship between the pollutant concentration and various influencing factors. In order to improve the prediction accuracy of pollutant concentration, the predicted value of each pollutant concentration in the LASSO regression model and the measurement data of the micro air quality detector are used as input variables, and the LASSO-NARX model is constructed using the NARX neural network. Several indicators such as goodness of fit, root mean square error, mean absolute error and relative mean absolute percent error are used to compare various air quality models. The results show that the prediction results of the LASSO-NARX model are not only better than the LASSO model alone and the NARX model alone, but also better than the commonly used multilayer perceptron and radial basis function neural network. Using this model to calibrate the measurement data of the micro air quality detector can increase the accuracy by 61.3-91.7%.
研究表明，空气污染与各种人类疾病之间存在一定的相关性，尤其是肺部疾病，因此监测空气中污染物的浓度具有重要意义。与国家空气质量监测站（国控点）相比，微型空气质量监测仪的优势在于可以实时、网格化监测污染物浓度，但需要提高其测量精度。本文提出了一种将最小绝对收缩和选择算子（LASSO）回归与非线性自回归模型与外生输入（NARX）相结合的模型，用于校准微型空气质量监测仪测量的数据。在建立 LASSO-NARX 模型之前，使用相关性分析来测试空气污染物浓度与其影响因素之间的相关性是否显著，并找出影响污染物浓度的主要因素。由于各影响因素之间存在多重共线性，因此使用 LASSO 回归来进一步筛选影响因素，并给出污染物浓度与各影响因素之间的定量关系。为了提高污染物浓度的预测精度，将 LASSO 回归模型中各污染物浓度的预测值和微型空气质量监测仪的测量数据作为输入变量，使用 NARX 神经网络构建 LASSO-NARX 模型。使用拟合优度、均方根误差、平均绝对误差和相对平均绝对百分比误差等几个指标来比较各种空气质量模型。结果表明，LASSO-NARX 模型的预测结果不仅优于单独的 LASSO 模型和 NARX 模型，而且优于常用的多层感知机和径向基函数神经网络。使用该模型校准微型空气质量监测仪的测量数据可以将精度提高 61.3%-91.7%。
