| Air pollution analysis is a hot topic in current urban modernization and atmospheric environment research.How to make good use of the historical observed pollution data and extract critical information to predict the trend of pollution change is of great significance for pollution prevention,pollution warning and pollution control.The numerical prediction method that can fully reflect the influence of meteorological and emission sources on other physical and chemical processes is an important method to carry out air pollution prediction.However,the current bottlenecks in air pollution analysis include:(1)bottlenecks in the rapid and efficient processing of big atmospheric environmental data;(2)bottlenecks in simulating the dynamic process of pollution change and calculating the interaction between various meteorological factors.The traditional theory-based air quality prediction model can simulate the air pollution process.But due to the large uncertainty of data dimensions,initial conditions and physical and chemical parameters,these traditional models have limited ability to fit air pollution..In recent years,artificial intelligence algorithms have succeeded in image and natural language processing,which provides new thinking and methods for air pollution analysis and prediction.Deep learning is one of the most popular intelligent algorithms currently.Deep learning has a large number of neurons and powerful data fitting capabilities.Deep learning algorithms can extract features and rules from rich and massive air pollution data,and deeply reveal the relationship between pollutants and meteorological factors,and achieve accurate prediction results.Deep learning technology brings the following opportunities for air pollution analysis:(1)deep feature analysis based on neural network;(2)time series correlation analysis based on multidimensional data;Based on the above considerations,this paper introduces deep learning technology,fully exploits the advantages of deep learning in big data processing,and studies the crossapplication of intelligent algorithms in the field of environmental science.Firstly,this paper proposes a pollution prediction model based on deep learning,and then improves the model from different aspects in order to achieve better prediction results.The main contents of this article are as follows:(1)Research on time series pollution prediction method.In this paper,an AutoEncoderbased Pollutant Prediction(RASP)model based on Auto-encoder neural network is proposed.The model consists of two parts: Encoder and Decoder.Encoder is used to extract the distribution characteristics of the previous time series pollution data;Decoder uses the extracted features to predict the pollution concentration in the next unknown time.Encoder and Decoder in the model use multi-layer recurrent neural network(RNN)to achieve long-term dependent prediction targets.At the same time,this paper attempts to modify classic RNNs,and proposes a new type of recurrent neural network based on Long Short-Term Memory(LSTM)to improve the performance of the RASP model.(2)Research on time series prediction methods based on Attention mechanism.This paper believes that air pollution observations at different times not only have impact,but also have different influences.Therefore,predicting air pollution concentrations requires a focus on air observations at one or more of the previous moments.Based on the RASP model,this paper introduces attention mechanism to calculate the influences at different moments.Attention mechanism make RASP model pay more attention on those moments that have great influence on the subsequent moments,and integrate these concerns into the prediction.The simulation results show that compared with the traditional air prediction model,the proposed RASP model can improve the accuracy of pollution prediction.At the same time,although the complexity has increased,the addition of Attention Mechanism in RASP model can further improve the accuracy of pollution prediction. |
PDF Full Text Request