Research On Carbon Trading Price Prediction Method Based On Deep Learning

Carbon emission trading is an important tool that uses the market economy mechanism to promote global greenhouse gas emission reduction.Carbon trading price refers to the market price of carbon emission trading and is the core indicator of the carbon emission trading market.Carbon trading price prediction can provide scientific guidance and evaluation for the operation of the carbon emission trading market and the goal of carbon emission reduction,as well as provide theoretical basis and empirical support for the formulation of government emission reduction policies and the development plan of the energy industry.The current work on predicting the price of carbon trading primarily uses machine learning and deep learning techniques,such as multi-layer perceptron,long short-term memory networks,etc.,while to some extent accounting for influencing factors like energy prices,environmental changes,and macroeconomics,which improves the accuracy of prediction to a certain extent.However,the differences in the above-mentioned influencing factors between the source domain and the target domain are not fully considered in the carbon trading price prediction research of cross-regional cross-market migration,which affects the effective migration of features.Therefore,there are two difficulties with carbon trading price prediction in cross-regional and crossmarket migration:(1)How to conduct an effective correlation analysis on the multi-source external factors that affect carbon trading prices,and screen out the indicators that play a leading role in carbon trading price fluctuations;(2)How to effectively extract features from multi-source data that affect carbon trading prices,and solve the problem of domain shift between different carbon trading markets,so as to realize the effective migration of features.In response to the above problem,this paper first proposes an influence factor analysis algorithm based on VMD-DMCA,which combines variational mode decomposition(VMD)and detrended moving average cross-correlation analysis(DMCA)to remove noise interference in the sequence while preserving the main frequency components and dynamic variation characteristics of the sequence,reducing the error and instability in the DMCA calculation process,and thus achieving a more effective and robust evaluation of the true correlation between nonlinear non-stationary sequences.Through this algorithm,this paper screens the indicators that affect the fluctuation of carbon trading price,and selects the most significant indicators for subsequent prediction work.Secondly,in order to effectively extract and transfer the multi-source data features that affect the carbon trading price,this paper designs and implements a carbon trading price prediction model based on domain adaptation network.This model adopts a source domain data filtering algorithm based on time series similarity,which obtains source domain data with low distribution difference,and avoids the impact of large distribution difference between source domain data and target domain data on the adaptive performance of domain adaptation network;meanwhile,this model also adopts a feature extraction module based on attention mechanism,which uses a trend attention module based on derivative dynamic time warping to filter and assign weights to the key information of the features learned by the network,highlighting more important contextual information,and achieving effective feature extraction of multi-source data that affect carbon trading price.This model can learn the deep invariant latent features between source carbon trading market and target carbon trading market,thus realizing effective feature transfer.Finally,this paper establishes a cross-market carbon trading price multi-source dataset,and designs a series of comparative experiments to verify the effectiveness of the constructed network model and each module in reducing carbon trading price prediction error.