Development of the method of forecasting the atmospheric air pollution parameters based on error correction by neural-like structures of the model of successive geometric transformations

Oleksandra Mishchuk

doi:10.15587/2312-8372.2019.188743

Abstract

The article describes the importance of improving existing and exploring new algorithms for predicting environmental parameters to improve the quality of environmental monitoring. Because the organization and management of production require the development of new approaches to the problem of control and management of industrial sources of harmful substances based on new information technologies. One of the most problematic places in industrial air quality control and management systems is the development of advanced prospective air pollution forecasting algorithms. These algorithms must take into account t situational changes in data distribution and do not require retraining of atmospheric air pollution parameters. With the advent of neural-like structures, there is a need for their study, including the task of predicting the parameters of air pollution. The object of research is the neural-like structures of the Model of Successive Geometric Transformations. A method for predicting the parameters of atmospheric air pollution based on error correction with the help of a committee of different types of neural-like structures is proposed. In the course of the study, three methods for predicting the parameters of atmospheric air pollution were analyzed: a Generalized Regression Neural Network, a Radial Basis Function, and a neural-like structure of Sequential Geometric Transformations Model. A combination of these methods was performed and the results of the three methods were compared. It is experimentally determined that the prediction of atmospheric air pollution parameters based on the error correction using the committee of neural-like structures of the Sequential Geometric Transformations Model provides a prediction error reduction by 7 % of the General Regression Neural Network and by 2.6 % of the Radial Basis Function with extended General Regression Network. The obtained results increase the reliability and speed of forecasting of atmospheric air parameters to improve the quality of monitoring of emissions of harmful impurities in production and to make environmental management decisions.

Highlights

IntroductionEnvironmental monitoring is an intelligent system with a wide variety of modules that provides the collection and processing of information obtained in the selected space-time field, further interpretation of the material, modeling, forecasting and management decisions [1]
Environmental monitoring is an intelligent system with a wide variety of modules that provides the collection and processing of information obtained in the selected space-time field, further interpretation of the material, modeling, forecasting and management decisions [1].The purposes of environmental monitoring are identification of potential hazards, development of measures to protect and prevention of the occurrence of critical situations, harmful or dangerous to human health and the existence of living organisms
Analyzing and evaluating the state of the air environment is important for choosing optimal management decisions, but they are based on the use of information that reflects present and past states. This is usually not enough to formulate a strategy, so trends in atmospheric air pollution must be taken into account to identify issues that may be encountered in the future

Summary

Introduction

Environmental monitoring is an intelligent system with a wide variety of modules that provides the collection and processing of information obtained in the selected space-time field, further interpretation of the material, modeling, forecasting and management decisions [1]. Analyzing and evaluating the state of the air environment is important for choosing optimal management decisions, but they are based on the use of information that reflects present and past states This is usually not enough to formulate a strategy, so trends in atmospheric air pollution must be taken into account to identify issues that may be encountered in the future. There is a need to create more accurate neural network prediction algorithms which will take into account a large amount of environmental monitoring data and will require less time in application mode for use on mobile devices and controllers. The aim of research is development of a high-speed method for predicting the parameters of atmospheric air pollution based on neural-like structures of the model of sequential geometric transformations and to compare the proposed method with the statistical methods (based on General regression neural network)

Methods of research

Research results and discussion

Findings

Conclusions