The Prediction Of Pollutants’ Emissions Modeling And Implementation Based On Improved Gray Neural Network

With the development of China, the emission of pollution become even moreserious, so the total pollutant control has also became the focus of the work ofGovernment. The main work of pollutant’s reduction is predicting the emission of thenext few years, building the scientific and rational pollutant emission reduction policies.Accordance with national requirements, the emission count of sulfur dioxide, nitrogenoxides, ammonia and chemical oxygen are needed. This research’s goal is to buildingthe model which can predict the pollutant’s emission according to the result ofanalyzing the factors such as industry distribution, population. This model wascombined with gray systems theory and neural-network theory to predict these fourpollutions’ emissions precisely.In this thesis, the GM(1,1), GM(0,4) and BP-neural net was used to build thecombination forecasting model according to the characteristics of the emission ofpollutants in Chengdu. This model analyzed the factors which affecting the emissions ofthese four pollutants and combined with the historical data of pollutant emission tocalculate the emission of these four pollutants in the next years. The data waspreprocessed before input into the model to prevent the error-data from disturbing. Andthrough the computation of GM(1,1), the result of preliminary forecast was got. Inorder to get the precisely result, the improved Euler’s formula was used to solve the graydevelopment parameter and grey action parameter. Secondly, The GM(0,4) modelchoses three factors from industry distribution, population as the main influencingfactors, and these main influencing factors was used to get the preliminary predictivevalue with the result of GM(1,1). At last, by correcting the residuals in BP neuralnetwork, the finally prediction data would be acquired.For testing the correctness of the model’s outcome, the Chengdu’s pollutantemissions data of2002to2008and the factors related were used as the input data of themodel,and the model’s outputs were compared with the other models. The simulationresults showed that, the average error of the improved combined forecasting modelsimulation of the emissions was less than4%, and the minimum error was0.10%, the error rate was far less than other single model’s. Finally, the prediction-system was builtby author in programming language, the system through simple operations which couldbe intuitive display forecast results, the results of classification, and provide decisionsupport for environmental staff reduction.