PM2.5 (particulate matter 2.5) concentration value prediction method based on memory neuron network

Abstract

The invention relates to a PM2.5 (particulate matter 2.5) concentration value prediction method based on a memory neuron network. The method comprises the following steps of: step 1, acquiring raw data, step 2, preprocessing various raw data by an unsupervised training method to generate an initial optimization parameter, and step 3, forecasting a PM2.5 concentration value by using the memory neuron network, wherein the raw data comprises PM2.5 concentration value history data, pollutant index (such as AQI (air quality index), PM10 (particulate matter 10), NO2, CO, SO2 and O3) history data andmeteorological history data. The PM2.5 concentration value prediction method based on the memory neuron network accurately describes a long-term change rule of the PM2.5 concentration value and improves the precision of prediction.

Description

technical field [0001] The invention relates to the technical field of prediction of air particulate matter PM2.5 concentration value, in particular to a method for predicting PM2.5 concentration value based on memory neural network. Background technique [0002] PM2.5 refers to particulate matter with a diameter less than or equal to 2.5 microns in the atmosphere. It is rich in a large amount of toxic and harmful substances and has a long residence time in the atmosphere and a long transportation distance. Therefore, it has a greater impact on human health and the quality of the atmospheric environment. Excessive PM2.5 also brought another impact - haze weather. Air pollution has become the focus of people's attention, and among the air pollution indicators, PM2.5 concentration has become a symbolic detection index to measure air quality. Nowadays, the prediction of PM2.5 concentration value in the future time period based on historical data has become a research problem w...

AI Technical Summary

Problems solved by technology

[0005] In order to overcome the inadequacy of existing PM2.5 concentration value prediction methods that cannot remember and screen the dependency relationship in the PM2.5 concentration value prediction training, the present invention is based on historical data on PM2.5 concentration values, historical data on PM2.5 concentration value-related indicators, and In addition to the nonlinear correlation analysis of historical meteorological data, an unsupervised initial optimization parameter generation method is also introduced to preprocess the original data, combined with a memory neural network that can memorize and screen training dependencies, and connect each storage unit with an input gate , an output gate is associated with an internal state that is fed into itself without interference across time steps, providing a PM2.5 concentration value prediction based on a memory neural network that accurately describes the long-term change law of PM2.5 concentration values ​​and improves prediction accuracy method

Images