Prediction Of Air Quality Based On BP Neural Network North Qinling Mountains Central

Compared to the cities, the region of north central of Qinling which islocated in the south of Xi’an has higher ambient air capacity, and it is of greatsignificance for us to study its air quality status and forecast its air quality change.As the green barrier for the cities, the air quality status and its change of theregion of north central of Qinling play an essential role in promoting theeconomic development of the surrounding cities. Further, they offer a warningfunction on the deterioration of urban environment air quality, and theestablishment a series of policies (such as urban planning, pollutants regulation,etc.) will be influenced directly. Zhouzhi County which is located in the centralof the region is still predominantly agricultural, and its environment air quality isless affected by external factors. In our study, Zhouzhi County was chosen as thetypical research subject in the region of north central of Qinling, its air qualityforecasts was studied via the help of the rather strong non-linear processingability of BP nepal network. The results showed that, there was a goodagreement between the forecast results and the actual experimental results,78.83%of the actual experimental data (≈283d) matched well with the forecastdata. It was considered that our research results could give expression to thebasic situation of the air quality in the region of north central of Qinling. Besides,the results also proved that, it is feasible to forecast the air quality in the regionof north central of Qinling via BP nepal network.The main research contents and results are shown as follows:(1)14factors which were significantly related to the index of AQI werescreened out from18primary predictors by using a linear regression method, andthen the input nodes of Air Quality Index AQI forecast model of Zhouzhi Countywere obtained.(2) In our study, seven kinds of algorithms were compared, includingmomentum BP algorithm, variable learning rate BP algorithm, resilient BPalgorithm, BFGS quasi-Newton algorithm, LM algorithm, SCG algorithm and Bayesian normalization method. And the results showed that, Bayesiannormalization method was suit to be the training algorithm for the air qualityforecasting model of the north central of Qinling.(3) The network training was carried out by Bayesian normalization method,the effects of network training and forecast under different conditions werecompared and analyzed via the change of hidden nodes and training times. Theresults showed that, the best hidden nodes of the air quality forecasting model ofthe north central of Qinling was4.(4) However, there was a certain limitation for BP nepal network as it wasused for the forecast of extreme pollution. Accordingly, in our study, a revisionmethod was suggested tentatively in order to correct the forecast results when BPnepal network was used for the forecast of extreme pollution.