Prediction Of China's Daily Carbon Emission Based On EMD-GA-LSTM

With the development of science and technology,people's living standards are improving day by day,and environmental problems are becoming more and more prominent.Over the past 170 years of industrialization,industrial products have greatly boosted productivity and humans have dependent on them more and more,but industrial emissions of greenhouse gases have increased and global temperatures have also climbed.Turning our attention to 2019,it holds the title of the highest global average temperature of the past decade,and the second-highest temperature of any year on record.The emission of greenhouse gases has caused great damage to the environment and affected the survival and development of human beings seriously.Therefore,from the perspective of environmental protection and sustainable development,the study of carbon emissions is conducive to the country's control of carbon emissions effectively,reducing environmental pollution,and reducing greenhouse gas emissions while maintaining economic development.The main work of this paper is to use deep learning and other methods to study China's daily carbon emissions data from 2019 to 2020.Taking 75% of the original data as the training set and 25% of the original data as the test set,three deep learning-based daily carbon emission prediction models were established and compared.The first model is the LSTM neural network.This model has a good fitting for the data's trend,but there is a large deviation in some extreme points.The second model is the GA-LSTM model,this model is improved from LSTM model,using genetic algorithm to search the optimal parameters of LSTM network,using test set data for predicting,compared with the LSTM model,the mean square error fell by 5.6%.The advantage of this model can save some manual steps.The last model is the EMD-GA-LSTM model,which optimises the structure of LSTM network with the help of empirical mode decomposition and genetic algorithm,getting a EMD-GA-LSTM model for everyday carbon predicting from LSTM.Compared with the previous two models,the prediction error of this model decreases by more than30%.In addition,genetic algorithm is used to optimize the structural parameters of the parameters,eliminating most of the parameter adjustment work.Finally,compared with the traditional time series model,the prediction error of this model is reduced by 31.2%compared with the time series model.The improvement of prediction effect of this model is mainly attributed to the innovative combination of two methods,namely empirical mode decomposition and genetic algorithm.Genetic algorithm can get the better LSTM network parameters.Empirical mode decomposition can remove some noise.By comparing the prediction errors of each model,it is concluded that the prediction effect of EMD-GA-LSTM model is better,which proves that the combination of EMDGA-LSTM model is effective.