Research On PM_2.5 Prediction Based On Time And External Factors

In recent years,with the rapid development of China’s industrialization and urbanization,the smog problem has become increasingly serious.As one of the main indicators of air pollution,PM_2.5 seriously threatens human health.Long-term exposure to air polluted by high concentrations of PM_2.5 will not only increase the prevalence of lung cancer,acute respiratory diseases,and asthma,but also seriously affect human life.PM_2.5 will also have direct or indirect impacts on climate change and human living environment.Therefore,PM_2.5 concentration prediction is an indispensable task.Scientifically and accurately predicting the PM_2.5 concentration can guide the country’s air pollution prevention work,and the organization of social activities.The daily concentration changes of PM_2.5 is affected by many factors,and has the characteristics of non-linear and time-varying,which is difficult to accurately predict.In response to this problem,this paper proposes a PM_2.5 daily concentration prediction method based on external influences and time series factors,namely the BP-EEMD-LSTM combined prediction method.This method separates the main external factors and time factors of PM_2.5 daily concentration,and establishes a BP neural network preliminary prediction model based on the main external factors and a residual correction model based on Ensemble Empirical Model Decomposition（EEMD）-Long-Short Term Memory（LSTM）.First,this paper analyzes the characteristics of the data of external factors affecting the concentration of PM_2.5 and preprocesses the data.Correlation analysis is carried out on the main external factors affecting PM_2.5 concentration,and the main external factors affecting the daily concentration of PM_2.5 are screened out.Secondly,through the BP model,the main factors of external influence are used as the input,and the BP predicted value of the daily PM_2.5 concentration of the prediction group and the historical residual time series of the experimental group are obtained.Then,the Ensemble Empirical Model Decomposition（EEMD）is used to decompose the residual time series into a series of relatively stable component time series in different frequency bands to reduce the complexity of the data,and then construct different Long-Short Term Memory（LSTM）models according to the characteristics of each component time series.The residual value of PM_2.5 daily concentration in the prediction set is obtained by linear addition of the predicted values of each component;finally,the BP forecast value and the residual correction value are combined to form the final PM_2.5 daily concentration forecast result.This paper uses the data of PM_2.5 daily concentration and other related factors in Hangzhou from 2014 to 2019 to conduct simulation experiments.The results show that compared with other models,the root mean square error of the combined forecasting model is 2.74,and the average percentage error is 8.01.Correlation The coefficient is closer to 1,and the prediction accuracy is higher.