Search

PM 2.5 is the main component of most haze, and the presence of high concentrations of PM 2.5 in the air for an extended time can cause serious effects on human health, so there is an urgent need for research work related to PM 2.5. Traditional PM 2.5 monitoring uses ground-based monitoring stations with low spatial resolution. Other studies have retrieved the Moderate Resolution Imaging Spectroradiometer aerosol optical depth product by the dark-target algorithm. However, the estimated PM 2.5 concentration on the ground will produce missing values, which will lead to the reduction of spatial and temporal resolution. Based on this, this study proposes a machine learning algorithm to estimate PM 2.5 by using the fifth generation reanalysis (ERA5) data set published by the European Center for Medium-range Weather Forecasts (ECMWF). In this study, two different methods of back propagation neural network (BPNN) and random forest (RF) were used to develop the models. Firstly, the meteorological parameters (precipitable water vapor, water vapor pressure and relative humidity, etc.) and pollution parameters (O 3 , CO, NO 2 , SO 2 , PM 10 , and PM 2.5) were used to establish PM 2.5 model in 2021. The results showed that the R2 and RMSE for BPNN and RF were 0.94/0.96 and 10.37/8.77 µg/m3, respectively. Then, due to the lack and the low spatial resolution of the pollution parameters, using only the ERA5 meteorological data with the high spatiotemporal resolution to develop the PM 2.5 model in winter, the R2 of the RF model (0.93) was 0.05 higher and the RMSE (12.50 µg/m3) was 4.19 µg/m3 lower than that of the BPNN model, which indicates that it is feasible to develop the PM 2.5 model using only meteorological parameters. Finally, using the RF model of the second stage and ERA5 meteorological data with a spatial resolution of 0.05° (obtained by cubic spline interpolation) to generate the hourly PM 2.5 map of Beijing and compare it with China High Air Pollutants dataset, the R2 and RMSE of Beijing were 0.78 µg/m3 and 14.85 µg/m3, respectively. On this basis, it is found that the areas with high PM 2.5 concentration are close to the areas with serious pollution in Hebei Province by analyzing the PM 2.5 map of Beijing, and area transport and human activities are important sources of air pollution. [ABSTRACT FROM AUTHOR]

• A novel nonlinear feature selective networks is designed. • Proposed method is applied for continuous values regression. • An optimizer joins back propagation and re-weighted is designed. The curse of dimensionality is a long-standing intractable issue in many machine learning and computer vision tasks. Feature selection, a data preprocessing technique, aims to select most discriminative feature subsets for improving performance of downstream machine learning tasks. However, most existing feature selection methods concentrate upon learning a linear relationship between data points and their labels, which causes that they are incapable of handling nonlinear complex data in real-world applications. In this paper, we first propose a novel end-to-end Nonlinear Feature Selective Networks (NFSN) that be able to select discriminative feature subsets while preserving their nonlinear structure by embedding a ℓ 2 , p -norm regularized hidden layer into designed continuous values regression networks. In addition, we propose an efficient optimization algorithm that joins back propagation algorithm and re-weighted optimization strategy to acquire derivative of all weights accurately. Experimental results on the nonlinear analog pulse signal and real-world datasets demonstrate the superiority of proposed method compared to some related methods on feature selection. Our source code available on: https://github.com/StevenWangNPU/NFSN. [ABSTRACT FROM AUTHOR]

Large-scale cluster extended reach wells (ERWs) are usually drilled from one platform to target distant multiple reservoirs in offshore oil & gas development. Due to the complicated downhole environments and complex geological conditions in various azimuths, the accurate prediction of the rate of penetration (ROP) in offshore large-scale cluster ERWs drilling is a challenging task. In this work, a ROP prediction method for offshore large-scale cluster ERWs drilling based on machine learning and big-data techniques is presented. 27 input parameters affecting ROP are collected from block N of China Bohai Oilfield, with a total of 9,193,408 data points. The internal relation between various parameters and ROP at the data level is revealed, which verifies the feasibility of the collected data set training. Based on various machine learning algorithms, 12 kinds of ROP prediction models based on support vector regression (SVR), random forest regression (RFR), back propagation neural network (BPNN), and recurrent neural network (RNN) are established respectively. Meanwhile, the prediction accuracy of the ROP prediction models is evaluated based on six performance indicators. It shows that the prediction results of the RNN model with LSTM as neuron structure are the best (all of the 6 indicators are optimal) among the 12 kinds of ROP prediction models, and the mean absolute error is only 6.12 m/h. The RNN model with LSTM as neuron structure is the best machine learning model in the prediction of ROP in offshore large-scale cluster ERWs drilling. It provides a practical method for enhancing cleaner development of offshore oil & gas. • A ROP prediction method for offshore large-scale cluster ERWs drilling is presented. • 12 kinds of ROP prediction models are established based on machine learning & big-data. • The prediction accuracy of the 12 kinds of ROP prediction models is evaluated. [ABSTRACT FROM AUTHOR]