Research And Application Of Indoor PM2.5 Prediction Based On Neural Networks

In recent years,with the acceleration of industrialization and urbanization in China,the gradual improvement of the state's attention to the environment and people's good expectations for a livable living environment,air quality has increasingly become the focus of attention of every citizen.PM2.5 is one of the most important pollutants in China's ambient air due to its small particle size,strong activity and harmful substances.Existing air pollution research focuses on outdoor air.However,people spend more than 80% of their time indoors every day.Older,younger and chronically ill people spend more time indoors.Indoor air pollutants are not only affected by outdoor air pollutants,but also related to the indoor environment itself.Indoor air pollution often causes more lasting harm to human body than outdoor air pollution,so it is more meaningful to study indoor air quality for health to some extent.But the particularity of indoor environment also significantly increases the difficulty of air quality prediction.In order to solve problems of inaccurate prediction of indoor PM2.5 and insufficient research on relevant characteristics,this paper explores the hidden non-linear relationship between indoor PM2.5 and other air pollutants.Using multiinstance technology to solve the problem of balance between sampling interval and prediction time and constructing indoor PM2.5 prediction model based on deep neural network.The main work of this paper is as follows.(1)According to the regulations of air pollutant collection methods,monitoring sensors are made indoors and outdoors to obtain air quality parameters such as indoor and outdoor PM2.5,and then the collected data are analyzed and preprocessed.In the construction of the neural network model,the sample data are divided into training set and test set according to the appropriate proportion.(2)Analyzing the basic principle of neural network and processes of forward transmission and backward feedback.The network structure of the model is determined according to the actual situation of indoor PM2.5.The relative error of the prediction result is 14.77% when the test set is input into the model.It is found that the prediction of indoor PM2.5 concentration using BP neural network is feasible,but the prediction results still need to be improved due to the simplicity of the model.(3)The genetic algorithm is used to optimize the BP neural network to avoid the network falling into local minimum.Compared with BP neural network,the average relative error of the final prediction result of genetic neural network is 13.15%,which is lower than that of BP neural network.This shows that genetic algorithm has played an optimization role in the neural network model.(4)Analyzing the existing problems of BP neural network and genetic neural network,considering the complexity of indoor PM2.5 pollution formation mechanism,shallow learning method is difficult to mine its deep time series characteristics,build a deep neural network model and use multi-example technology to solve the inconsistency between sample size and predicted size.The results show that the relative error of the prediction results of the depth neural network model is 5.60%,which is better than the learning methods such as BP neural network,genetic neural network and support vector machine.(5)Develop a set of APP and its background program for monitoring and forecasting indoor PM2.5,which is helpful to grasp the trend of dynamic change of air quality.At present,the system has been put into use in a kindergarten in Kunshan City.The monitoring system has been installed in seven classrooms.Up to now,the system has collected more than 290,000 data and successfully warned many times,which meets the growing health needs of schools and parents for children's growing environment.In conclusion,the indoor PM2.5 prediction models based on BP neural network,genetic neural network and deep neural network are constructed respectively with actual collected data.The best prediction results are obtained by deep neural network.Finally,a set of indoor PM2.5 monitoring and prediction system is designed,which has been put into practical use.