Research On Carbon Price Prediction By Deep Learning Based On Multi-modal Features Integrating Multi-source Dat

In the context of the increasingly severe global climate change problem,the ultimate goal of the global energy transition,which is mainly characterized by clean and low-carbon energy,is to gradually achieve carbon dioxide emission reduction,and finally achieve net zero emissions or zero emissions.The effective analysis and prediction of carbon price not only contributes to the reasonable allocation of carbon resources,but also promotes the development of carbon trading market and the transformation of low-carbon economy.However,carbon market is a complex market,and many factors affect the fluctuation of carbon price.Especially with the rapid development of the Internet,a large amount of unstructured information related to carbon price has emerged.Carbon price prediction based only on traditional carbon market time series data is increasingly showing data limitations.In addition,existing carbon price prediction studies also ignore the importance of feature selection,carbon price and its exogenous variables preprocessing,intelligent optimization and deep learning models in improving the effectiveness of prediction,which may lead to poor prediction performance.Therefore,starting from the intersection of computer and finance,this study constructs a multimodal feature deep learning forecasting technology framework integrating multi-source data information,and applies this framework to the specific carbon market price prediction problem,so as to solve the shortcomings of existing studies.Firstly,this research empirically investigates the correlation and regional differences between various driving factors and regional carbon prices in China from the perspectives of linear and nonlinear correlations.The impact of energy,finance,environment and international carbon market on the price of China’s carbon market is analyzed respectively.Pearson correlation coefficient and nonlinear Granger causality test are used to empirically analyze the correlation between driving factors and the price of carbon market,and preliminary screening of input variables is achieved.Secondly,to overcome the performance limitations of existing carbon price prediction models,a novel hybrid carbon price prediction model that integrates multi-source data is proposed.Based on the modeling idea of "divide and rule",this model designs a novel hybrid carbon price prediction framework that combines advanced data preprocessing technology,effective feature selection methods,intelligent optimization algorithms and excellent deep learning models.In addition,from the perspective of behavioral finance research,this research constructs a news sentiment index based on the financial field sentiment dictionary and an attention index based on network search behavior data to address irrational investment behaviors such as the "herd effect" in the financial market.Empirical analysis is conducted based on the proposed mixed prediction model.The results indicate that the multimodal feature deep learning carbon price prediction framework proposed in this paper,which integrates multi-source data,can effectively improve the shortcomings of existing carbon price prediction research.Finally,a visualization system of carbon price prediction for Chinese regional carbon market is realized by using the data mining method and the constructed model.