Multi-mode Integrated Prediction Of Pollutants Based On Improved Genetic Algorithm

At present,China's atmospheric pollutant concentration forecasting technology has entered a period of rapid and steady development.With the continuous improvement of numerical forecasting technology,the level of atmospheric pollutant concentration has been significantly improved,as numerical forecasting matures,it currently depends only on numerical models.It has not been able to greatly increase the forecast of atmospheric pollutant concentration.In response to the above issues,this paper does the following:Firstly,The prediction ability of numerical forecast products of Tianjin Meteorological Observatory is different in different environments.This paper proposes a screening discriminant analysis method based on the analysis of numerical forecast products:First,I use the eight numerical models in 2015 for data analysis and compare the eight numerical models in different.The seasonal forecasting ability was verified to verify the effectiveness of the integration;then,eight kinds of numerical models were tested for significance in different seasons,and a screening method for correlation-associated bias was introduced as a classifier to perform screening and classification in the four seasons.Finally,the selected numerical model is subjected to principal component analysis,and the analyzed principal components are used as the input of the integrated improvement algorithm.Secondly,aiming at the problem that the concentration accuracy of atmospheric pollutants is not high in the numerical model,an effective multi-model pollutant integrated forecasting method based on the extreme learning machine(ELM)improved genetic algorithm with genetic operators is proposed.The improved algorithm firstly uses the excellent nonlinear mapping capability of extreme learning machine to quantitatively simulate the operation of genetic algorithm population evolution to construct the ELM evolution mechanism.Then the ELM evolution mechanism is combined with the genetic algorithm to establish a multi-model pollutant integrated prediction model.The algorithm uses a 30-day rolling forecast.This article compares experiments and analyzes the accuracy of different integrated algorithms and single-mode tests.The experimental results show that the improved genetic algorithm model of extreme learning machine(ELM)with genetic operators has a more reliable forecasting ability.The prediction error is obviously less than any numerical model and other integrated algorithm models,and it is closer to the real data.At the same time,the establishment of atmospheric pollutant concentration forecasting system will provide forecasters with accurate and reliable forecasting reference in a short period of time,which has a good application prospect.