A Study On Appllcatlon Of Neural Network Based On Genetic Optlmization And Bayeslan Regularlzation In Air Quality Predlction

Since the1760s, the human society has gone through industrial revolutions one after another for three times, which brought us the rapid development of economic and social, and also a serious environmental problem at the same time. With the continuous improvement of the level of industrialization of the world, the rapid expansion of the urban population and the daily growth in the number of motor vehicles have led to the air quality (such as sulfur dioxide (SO2), nitrogen oxides (NOx), carbon monoxide (CO), respirable particulate matter (PM10, and PM2,s) etc. as major pollutants) from bad to worse. Air pollution problems has become increasingly prominent, and aroused more and more general concern of the society.Environmental Protection Monitoring Station of some city in Shandong Province was found. Since then, a large number of air-quality monitoring data was collected. Especially after the rapid development in recent years, the number of monitoring sites in the city area increasing, and the quality of the automatic monitoring of environmental parameters has been greatly improved, which provide an important source of data for analysis and forecast of ambient air quality of the entire city. With these large quantities of high-quality monitoring data, a problem based on how best to improve ambient air-quality forecast accuracy is an important issue that must be taken into account and be studied in the current period of time. As not reflecting the inpact of meteorological events, the traditional prediction methods have been unable to meet the people’s growing urgent need. The artificial intelligence network has a strong nonlinear mapping ability and generalization characteristics, and it can take full advantages of characteristics such as the diverse environmental meteorological factors. So it has begun to be used in the research and application of the ambient air quality forecasts.In this paper, we design and implement an urban air quality prediction model, which is based on artificial neural network theory. The model is used to predict ambient air quality of the city in Shandong province, with day concentration of SO2 NO2and PM10, and the result is converted to the form of air pollution index according to the ambient air quality standards [1]. The model uses a hybrid learning algorithm, LM (Levenberg-Marquardt) algorithm to train the neural network based on the introduction of genetic algorithms (Genetic Algorithm, GA) to optimize the initial weights of the neural network, to a certain extent avoiding the neural network is premature convergence to local minimum value, the introduction of a Bayesian regularization algorithm (Bayesian Regularization Algorithm) to adjust the neural network training objective function, directly improving the generalization ability of the neural network. The concentration of pollutants in the most recent period of the year and the most recent time of the same period in the previous year are selected as training sample data, and the neural network input and output data is used as normalization and anti-normalization form before and after the training and forecast, etc. Finally, the structural parameters of the network are determined through repeated and a large number of trials, and the experimental results are compared with the experimental results of the reference model based on the validation set. The results show that, the prediction model designed in this paper has a better predictive effect, and is better able to meet the requirements of the ambient air quality forecast for the city.