PM2.5 concentration prediction method based on stack self-encoding and support vector regression

Abstract

The invention discloses a PM2.5 concentration prediction method which combines stack-type self-coding and support vector regression. Steps are as follows: selecting influential factors with strong correlation as independent variables of the PM2.5 concentration prediction model, continuously collecting the selected independent variables and corresponding PM2.5 concentration data for N hours, and constructing a training set; a deep learning network model of stack self-encoder with K hidden layers being constructed; the training set being input into the model for training and the weights of eachvariable being obtained; the SVR model being trained by using the eigenvalues of the learning network model as input vectors, and the output of the SVR model being obtained; the test set being obtained, the test sample being predicted by the predictive model, and the precision of the predictive result being calculated. The invention has the advantages of SAE extracting data set features and SVR excellent prediction ability, and the prediction performance and precision are better than widely used ANN algorithm and SVR algorithm.

Description

technical field [0001] The invention belongs to the field of air quality prediction, and in particular relates to a PM2.5 concentration prediction method. Background technique [0002] In recent years, with the serious decline in air quality around the world, in order to better warn the occurrence of haze weather, some scholars and business organizations at home and abroad have also begun to pay attention to the effective prediction of fine particles in the air. [0003] At present, the commonly used air quality forecasting techniques are mainly divided into two categories: statistical forecasting and numerical forecasting. Among them, the statistical forecast is generally based on the correlation model between historical data and meteorological conditions, and then uses the meteorological conditions in the future and the correlation model established to predict the future air quality. This method has low requirements for input data, but the prediction results are generally...

AI Technical Summary

Problems solved by technology

[0006] In order to solve the technical problems raised by the above-mentioned background technology, the present invention aims to provide a PM2.5 concentration prediction method that combines stacked self-encoding and support vector regression, and overcomes the non-convex optimization problem in the traditional ANN model that cannot go beyond one or two hidden layers , so that it is impossible to learn through deep neural networks to obtain higher accuracy problems

Images