Air quality plays a vital role in people’s health, and air quality forecasting can assist in decision making for government planning and sustainable development. In contrast, it is challenging to multi-step forecast accurately due to its complex and nonlinear caused by both temporal and spatial dimensions. Deep models, with their ability to model strong nonlinearities, have become the primary methods for air quality forecasting. However, because of the lack of mechanism-based analysis, uninterpretability forecasting makes decisions risky, especially when the government makes decisions. This paper proposes an interpretable variational Bayesian deep learning model with information self-screening for PM2.5 forecasting. Firstly, based on factors related to PM2.5 concentration, e.g., temperature, humidity, wind speed, spatial distribution, etc., an interpretable multivariate data screening structure for PM2.5 forecasting was established to catch as much helpful information as possible. Secondly, the self-screening layer was implanted in the deep learning network to optimize the selection of input variables. Further, following implantation of the screening layer, a variational Bayesian gated recurrent unit (GRU) network was constructed to overcome the complex distribution of PM2.5 and achieve accurate multi-step forecasting. The high accuracy of the proposed method is verified by PM2.5 data in Beijing, China, which provides an effective way, with multiple factors for PM2.5 forecasting determined using deep learning technology.

/pdf/variational-bayesian-network-with-information-4jzhs71k.pdf

Variational Bayesian Network with Information Interpretability Filtering for Air Quality Forecasting

Influence and prediction of PM2.5 through multiple environmental variables in China.

Vibration analysis process based on spectrogram using gradient class activation map with selection process of CNN model and feature layer

Fine particulate matter (PM2.5) is a major pollutant in Guanzhong Urban Agglomeration (GUA) during the winter, and GUA is one of China’s regions with the highest concentrations of PM2.5. Daily surface PM2.5 maps with a spatial resolution of 1 km × 1 km can aid in the control of PM2.5 pollution. Thus, the Random Forest and eXtreme Gradient Boosting (RF-XGBoost) model was proposed to fill the missing aerosol optical depth (AOD) at the station scale before accurately estimating ground-level PM2.5 using the recently released MODIS AOD product derived from Multi-Angle Implementation of Atmospheric Correction (MAIAC), high density meteorological and topographic conditions, land-use, population density, and air pollutions. The RF-XGBoost model was evaluated using an out-of-sample test, revealing excellent performance with a coefficient of determination (R2) of 0.93, root-mean-square error (RMSE) of 12.49 μg/m3, and mean absolution error (MAE) of 8.42 μg/m3. The result derived from the RF-XGBoost model indicates that the GUA had the most severe pollution in the winter of 2018 and 2019, owing to the burning of coal for heating and unfavorable meteorological circumstances. Over 90% of the GUA had an annual average PM2.5 concentrations decrease of 3 to 7 μg/m3 in 2019 compared to the previous year. Nevertheless, the air pollution situation remained grim in the winter of 2019, with more than 65% of the study area meeting the mean PM2.5 values higher than 35 μg/m3 and the maximum reaching 95.57 μg/m3. This research would be valuable for policymakers, environmentalists, and epidemiologists, especially in urban areas.

/pdf/estimating-pm2-5-concentrations-using-the-machine-learning-1x4t3dhh.pdf

Estimating PM2.5 Concentrations Using the Machine Learning RF-XGBoost Model in Guanzhong Urban Agglomeration, China

Multi-modal image feature fusion-based PM2.5 concentration estimation

A deep learning-based PM2.5 concentration estimator

The invention provides a polar-harmonic sine and cosine transformation algorithm. A new radial kernel is constructed by utilizing a sine-cosine function, such that the stability is realized; both the zero number and distribution of the radial kernel are superior to the traditional polar-harmonic transformation algorithm based on two-dimensional transformation; compared with the traditional method, in the aspect of graphical representation, the algorithm is more suitable for a high-order moment condition; and simultaneously, the algorithm has rotational invariance and excellent graphical representation performance. The polar-harmonic sine and cosine transformation algorithm disclosed by the invention has the advantages of the traditional polar-harmonic transformation algorithm based on two-dimensional transformation and overcomes the disadvantage that the numerical value of the traditional polar-harmonic transformation algorithm based on two-dimensional transformation is non-steady simultaneously; and the polar-harmonic sine and cosine transformation algorithm disclosed by the invention has wide application prospect and can be widely applied in the fields of image retrieval, face recognition and the like.

Cairong Lou

Papers

A deep learning-based PM2.5 concentration estimator

Polar-harmonic sine and cosine transformation algorithm