A Study Of Air Quality Prediction In Lu’an City Base On Combination Of The Back Propagation Artificial Neural Networks And Genetic Algorithm

The current air quality environment has a lot of problems in our country, including PM2.5 and PM10 atmospheric pollutants. Air pollution not only affects the physical and mental health of local people, but also seriously breaks the ecological environment. It has great influence on the development of urban economy and the investment environment. To carry out air quality forecast, we can monitor the air pollution situation in time, according to the changes of the future meteorological conditions, forecast the air quality, and consciously to reduce or reduce the pollutant emissions, so as to achieve the goal from passive defense to active prevention. The State Council issued the " air pollution prevention action plan ", In the plan clearly put forward all over the country need to " establish a monitoring and early warning system, improve the accuracy of monitoring and early warning, timely release monitoring and warning information."The return of the classical statistical model and the numerical prediction model is the most cities at home and abroad taken for air quality forecast method, but it can’t satisfy the huge amounts of data management and utilization of low accuracy, not form a good decision support functions. Back Propagation(BP) artificial neural networks is mainly used for function approximation, with the non-linear processing power and capacity of noise. But in practice there are many flaws in BP, Network training may fall into local minima, learning algorithm convergence is slow. Therefore, this paper is told that the genetic algorithm to optimize the initial weight threshold of BP neural network model in order to further improve air quality forecasting capacity. This paper studies the implementation in Delphi environment based on genetic algorithm and BP neural network combination of air quality forecast model.The main work of this paper is listed as follows:(1) According to the collected data of air pollutant concentration in 5 years (2011-2015), From January 2015 to February the main pollutants were PM10 and PM2.5, and the concentration of pollutants increased. In 2011-2014 year, the average annual SO2 concentration in the first two years, the average annual increase in the two years; NO2 annual average concentration in the first 2 years of increase,13 young,14 years and a substantial increase; The annual concentration of PM10 in 11-14 years, has been growing. Regardless of the air quality standards, only think in terms of the concentration of the pollutants, so we put the particulate matter of course become the main pollutants of Lu’an city. Analysis of the climate characteristics of Lu’an, the air quality of how to divide the level, for the back based on the time series of the network model to establish the input factor.(2) Research on the Back Propagation(BP) neural network of the structure and training methods, using Dephi compiled the BP network training interface. Summarizes the classical statistical models, forecast the SO2 pollution density in January 2011 to December. With the 5-5-1 BP network structure model, in October to November the value of the training data to predict, and compared with classical statistical models to predict the value experiment comparison in November, the result shows that the BP neural network model has better performance.(3)Analyzing the defects of BP neural network model, researching on the genetic algorithm and how to choose the algorithm implementation, crossover and mutation. The genetic algorithm how to solve the coding and the issue of fitness function. In this paper, it shows that how to use genetic algorithm to optimize the BP neural network initial weight threshold. The optimized BP model compared with before optimization model, the experimental results show that the optimized network error in further reduced.(4) The air quality forecast model of BP-GA is developed by Delphi. With the 5-5-1 BP network structure model, BP-GA model and statistical model to predict the pollutant density of SO2, the experimental results of three kinds of model are analyzed; In the 5-5-1 network structure of BP model and BP-GA model forecasting PM10, PM2.5, the value of NO2, two models for comparison; Using the 10-5-1 network structure of BP-GA model forecasting PM10 and the 5-5-1 network structure of BP-GA model forecasting PM10, two kinds of network structure are compared.