Air pollutant concentration prediction method based on graph attention mechanism

Abstract

The invention discloses an air pollutant concentration prediction method based on a graph attention mechanism. The method comprises steps of constructing a spatial pollutant concentration prediction model based on a graph attention mechanism; and taking the meteorological data, the air monitoring data and the environmental factor data as model input data, constructing a graph adjacency matrix through a graph attention mechanism, extracting graph information characteristics by utilizing a graph convolutional neural network layer and a multi-layer perceptron network layer, and outputting a predicted air pollutant concentration value. According to the method, the air pollutant concentration prediction is more accurate, and the process is more efficient.

Description

technical field [0001] The invention belongs to the technical fields of graph convolutional neural network technology and air quality monitoring, and relates to a technology for predicting the concentration of air pollutants at prediction points, in particular to a method for predicting the concentration of air pollutants based on a graph attention mechanism. Background technique [0002] Air quality has always been an important component in the study of changes in environmental pollution. Changes in air quality are determined by the concentration of air pollutants. Studying the concentration of air pollutants can better grasp the changes in air quality. Most of the prediction of air pollutant concentration is to collect data of various related influencing factors, and carry out the correlation analysis of pollutants. The collected data of influencing factors are used as independent variables, and the data of air pollutant concentration are used as dependent variables for co...

AI Technical Summary

Problems solved by technology

[0006] Existing prediction algorithms, prediction algorithms based on machine learning mainly include multiple linear regression (Multiple Linear Regression, MLR), support vector machine (Support Vector Machine, SVM), random forest (RandomForest, RF) method, among them, multiple linear regression algorithm The calculation is relatively simple, fast, and the results are easy to understand, but it requires high data quality and poor fitting; the support vector machine algorithm is robust and can reduce the probability of overfitting, but it is difficult to train large-scale data ; The random forest algorithm has strong anti-overfitting ability, stable algorithm, and strong data adaptability, but it is sensitive to noisy data, and the calculation cost is high and time-consuming

The prediction algorithms based on deep learning mainly include BP neural network and recurrent neural network algorithm (Recurrent Neural Network, RNN). Among them, BP neural network has strong fault tolerance, nonlinear mapping and self-learning ability, but the amount of data is large and the algorithm convergence speed is slow. , Strong sample dependence; the cyclic neural network algorithm is easy to solve time series problems, but the processing effect on non-time series data is not good

Graph Attention Network Algorithm (GAT, Graph Attention Network) adds an attention mechanism to give importance to the edges between nodes and help the model learn structural information, but the relative disadvantage is that the training method is not very good, the efficiency is not high enough, and the effect is poor

Images