Study On Artificial Neural Network For Forecasting Airborne Particulate Concentration

In recent years,the issue of atmospheric pollution has attracted widespread public attention.It has become a research hotspot in academia that the airborne particle concentration prediction model can provide valuable reference data for air pollution control work.Based on the basic principles and algorithms of artificial neural network and deep learning,this paper establishes a double-layer BP neural network model based on restricted Boltzmann machine and applies it to the prediction research of PM_2.5.5 concentration.The main work of this article includes the following three aspects:1)A restricted Boltzmann machine with a double-layer BP neural network is combined to establish a double-layer BP neural network model based on a constrained Boltzmann machine.In this model,the restricted Boltzmann machine can obtain the characteristic information of the input parameters during the training process and use this information as the initial value of the double-layer BP neural network,which makes up for the double-layer BP neural network easily fall into the local optimum.This article uses the monitoring data of Shanghai,Hangzhou and Chengdu from December 2013 to April 2017 to study,including air quality index?AQI?,PM_2.5,PM₁₀,SO₂,CO,NO₂,temperature,dew point,and air pressure.After correlation analysis and normalization of the data,all data were grouped according to the ratio of 80%,10%,and 10%for training,verification,and testing of the model,respectively.2)The numerical results show that compared with the traditional BP neural network model?BPNN?,the prediction results of the double-layer BP neural network model based on constrained Boltzmann machine?RBM-DL-BPNN?are closer to the real monitoring value and the error is smaller.This article uses the performance indicators RMSE,MAE,and MAPE to objectively evaluate the network model.Taking Hangzhou as an example,the calculated values of RMSE,MAE,and MAPE for the RBM-DL-BPNN model are 13.947?g/m3,5.945?g/m3,and 12.637%,respectively,and for the BPNN model are 17.653?g/m3,13.150?g/m3,and27.655%.In order to deeply analyze the prediction results,we also conducted absolute error rate analysis and prediction accuracy analysis.Taking Hangzhou as an example again,the accuracy of the prediction results of the RBM-DL-BPNN model in the 0-20%,20%-50%,and>50%are 107,12,5,respectively,and the BPNN model are 64,44,16.3)We developed a PM2.5 concentration prediction software in MATLAB environment.Users can enter the relevant historical data in the software to easily obtain the predicted concentration value of PM_2.5,so that appropriate preventive measures can be made in advance.