Research And Application Of A Kind Of Hybrid Forecasting Model

PM2.5 is one of the most serious pollutants in the air.The factors affecting PM2.5content are complex and varied,and there is not only a linear relationship but also a nonlinear relationship.How to establish a model that combines the linear and nonlinear characteristics skillfully and effectively improves the prediction accuracy is the core problem of the research.First consider to extract the linear feature in the PM2.5 data,time series model of ARIMA model can be very good to extract the linear feature,the experiment proved that multivariate ARIMAX model prediction effect is better than univariate ARIMA model of prediction effect,suggests that changes in concentrations of PM2.5 is not only affected by their own change,but also related to many complicated factors.The extraction of nonlinear features involves the gated neural network GRU model.The activation function inside the neural unit can arbitrarily approximate the nonlinear characteristics of the output data.The model residuals output by ARIMAX model were input into GRU model(GRU1)to obtain the predicted value of the nonlinear part,and through the comparison of several combinations,the final combination of linear predicted value and nonlinear predicted value in a ratio of 3:1 was better,and the model prediction after combination was better than that of each single model.This way,although the two parts are taken into account,but still emphasizes the linear part of the role,and supervision over all data into question after establishing models GRU(GRU2)is a nonlinear model,and by adopting the idea of model integration will be two separate models together in a certain way to get the final forecast.The former actually combines two parts of the data,while the latter combines two models of the complete data.The prediction experiment of PM2.5 content was conducted with the meteorological data of Beijing from 2014 to 2019,including minimum temperature,maximum temperature,average temperature,humidity,wind speed,wind level,air pressure,visibility,precipitation,and average cloud cover,etc.,indicating that the prediction model of(ARIMAX + GRU1)-GRU2 can better predict the trend of daily PM2.5 content change in Beijing.