| For a long time,the increasing human demand for fossil energy has led to the excessive emission of greenhouse gases,mainly carbon dioxide.While accelerating the abnormal changes of the global climate and the process of global warming,it has also made the mitigation of greenhouse gas emissions to solve the deteriorating climate and environmental problems has become a serious challenge facing countries around the world.With the promulgation of the United Nations Framework Convention on Climate Change,the Kyoto Protocol,carbon trading has gradually become one of the effective means and mechanism for reducing greenhouse gas emissions.In the carbon trading market,as the changing trend of carbon trading price plays a fundamental role in the decision-making of relevant market participants,the scientific forecasting of carbon trading price becomes a hot issue concerned by market participants and scholars.At the same time,due to the complex influencing factors of carbon trading price and its inherent non-linear and non-stationary characteristics,accurate carbon trading price forecasting has become a difficult problem in the field of statistical prediction.In recent years,domestic and foreign scholars have conducted a lot of research on carbon trading price forecasting models,which can be roughly divided into two types:forecasting models based on the historical value of carbon trading price,and carbon trading price forecasting models based on multi-factors.Although the existing carbon trading price forecasting models can obtain better forecasting results,there are still some problems:Firstly,the great potential of machine learning model to improve the forecasting performance of carbon trading price is not fully understood.Secondly,most of the researches only use the traditional single-obj ective optimization algorithm,but the meta-heuristic optimization algorithm is easy to fall into the local optimal solution,and most of the previous carbon price forecasting models focused on the use of the single-objective optimization algorithm to improve the forecasting accuracy of the model,ignoring the importance of the forecasting stability,leading to few studies involving multi-objective optimization algorithms.Thirdly,without considering the importance of feature selection in the forecasting model,the accuracy and efficiency of forecasting are reduced to some extent.In addition,ignoring the importance of analyzing the characteristics of the carbon trading price to improve the forecasting performance.Besides,most studies mostly focus on point forecasting and one-step forecasting,ignoring interval forecasting and multi-step forecasting,while interval forecasting and multi-step forecasting can provide more valuable information for the carbon trading market.Finally,most carbon trading price forecasting models based on multiple factors only consider preprocessing for carbon trading price data,but ignore the importance of data preprocessing of other external influence variables.Aiming at the problems existing in the existing research,this paper introduces two machine learning models that solve the defects of traditional artificial intelligence models as the basic forecasting model,including the machine learning prediction model extended to the extreme learning machine theory and the machine learning prediction model extended to the fuzzy inference system theory,and fusion of data preprocessing algorithms,artificial intelligence optimization algorithms,feature selection algorithms,etc.,to establish and apply carbon transaction price prediction models from different perspectives,and to make up for the shortcomings of existing research.Specifically,the research content of this paper is mainly divided into seven parts:The first chapter introduces the background and research significance of the topic,the research at home and abroad,the main research content and the main innovations and shortcomings.The second chapter discusses the basic theory of the carbon trading price and the basic carbon trading price forecasting model introduced in this paper.The third chapter introduces the evaluation system of forecasting model constructed in this paper.Chapter four,chapter five and chapter six propose the carbon trading price forecasting model based on two-stage feature selection and multi-objective optimization,the carbon trading price forecasting model based on chaos theory and hybrid optimization algorithm,and the carbon trading price forecasting model based on multiple factors,separately,and apply them to actual carbon trading markets.Experiments show that the carbon trading price forecasting models proposed in this paper can obtain better prediction accuracy.Chapter seven is the summary of the full text and prospects for future research directions.The work of this paper mainly focuses on the construction and application of carbon trading price forecasting models:(1)In view of most previous carbon trading price forecasting researches that did not consider feature selection,only focused on the accuracy of the forecasting model,and rarely considered the stability of the forecasting results,this paper proposes a carbon trading price forecasting model based on two-stage feature selection and multi-objective optimization algorithm.The model reflects the effective combination of the advantages of algorithms such as machine learning models,data preprocessing algorithms,feature selection,and multi-objective optimization algorithms.Specifically,the model firstly preprocesses the original carbon trading price data using the decomposition and reconstruction strategy to effectively reduce the impact of noise on the prediction performance.Then,the proposed two-stage feature selection algorithm is used to determine the optimal input variables for the prediction model.In addition,in order to obtain the prediction results with better forecasting accuracy and forecasting stability,the multi-objective grasshopper optimization algorithm is used to optimize the weighted regularized extreme learning machine and predict the decomposed-reconstructed sequence.Finally,the final prediction results are obtained by summing up the results of each subsequence.(2)In view of the existing carbon trading price forecasting only considers the point forecasting research,the lack of analysis on the characteristics of carbon trading price and the research on the interval forecasting model,a carbon trading price forecasting model based on chaos theory and hybrid optimization algorithm is proposed,and the model includes analysis module and forecasting module.In the analysis module,the chaotic analysis and the determination of the optimal distribution function of the original carbon trading price can effectively analyze the characteristics of the carbon trading price and lay a foundation for the accurate establishment of the forecasting model.The forecasting module includes point forecasting and interval forecasting.In point forecasting,the model uses the phase space reconstruction method to determine the optimal input-output variables of the forecasting model,and uses the hybrid butterfly optimization algorithm-sine cosine algorithm proposed in this paper to optimize the adaptive neuro-fuzzy inference system.For interval forecasting,the interval forecasting results can be obtained by the point forecasting results and optimal distribution function of the carbon trading price sequence determined by the analysis module.(3)In view of the most of the existing researches do not consider the importance of the factors such as feature selection,data preprocessing for external impact variables,improved multi-objective optimization algorithm and multi-step forecasting in the carbon trading price forecasting,a carbon trading price forecasting model based on multiple factors is proposed.Specifically,advanced data preprocessing algorithm is firstly used for preprocessing of carbon trading price series and its exogenous impact variables,which can effectively solve the negative impact of noise.Then,effective feature selection algorithm is used to determine the optimal input features of the forecasting model,which can improve the forecasting performance of the model and providing more valuable information for carbon market participants.In addition,in order to overcome the limitations of the extreme learning machine and the artificial neural network models,based on the multi-objective chaotic sine cosine algorithm proposed in this paper,optimal kernel based extreme learning machine model with good generalization ability and stability is established.Finally,the performance of single-step forecasting and multi-step forecasting is verified based on the established forecasting model.In general,the main innovations of this paper are:(1)Based on advanced machine learning method,artificial intelligence optimization algorithm,feature selection algorithm,advanced data preprocessing algorithm et al.,this paper fully integrates the advantages of each algorithm to build a theoretical framework of carbon trading price forecasting methods.(2)Different from the existing researches that only build one carbon trading price forecasting model,this paper builds multiple carbon trading price forecasting models from multiple perspectives,which has stronger application value.In addition,in order to verify the superiority of the proposed models,this paper constructs a comprehensive evaluation system of carbon trading price forecasting model,which includes evaluation of model forecasting accuracy,evaluation of model forecasting effectiveness,and evaluation of model forecasting significance.(3)A carbon trading price forecasting model based on two-stage feature selection and multi-objective optimization is proposed.This model proposes a two-stage feature selection algorithm to provide a choice for determining the optimal input feature.In addition,the model adopts multi-objective optimization algorithm to optimize the weighted regularized extreme learning machine,which can improve the forecasting accuracy and the stability of the forecasting results at the same time.(4)A carbon trading price forecasting model based on chaos theory and hybrid optimization algorithm is proposed.First,A novel hybrid optimization algorithm is proposed in this model,which provides a new feasible option for solving the optimization problem.Secondly,an improved adaptive neuro-fuzzy inference system is proposed and applied to carbon trading price forecasting for the first time.Thirdly,the characteristics of carbon trading price are analyzed before modeling,which lays a foundation for accurate modeling.Finally,the forecasting module constructed in this study considers both point forecasting and interval forecasting,which can quantify certainty and uncertainty at the same time.(5)A carbon trading price forecasting model based on multiple factor is proposed.Firstly,this model uses advanced data preprocessing technology to preprocesses the carbon trading price data and its external impact variables,which can effectively reduce the impact of noise on the forecasting performance.Secondly,an effective feature selection algorithm is used to determine the optimal input feature of the forecasting model.Thirdly,this model successfully proposed a new multi-objective optimization algorithm:multi-objective chaotic sine cosine algorithm,which provides a new choice for processing multi-objective optimization problems.Moreover,this study uses an optimal kernel-based extreme learning machine model with good generalization ability for forecasting,which can effectively improve the forecasting ability.F:inally,this proposed forecasting model can improve the performance of single-step and multi-step forecasting at the same time,which can provide sufficient information for the carbon trading market.The research of this paper has important theoretical and practical significance:From a theoretical point of view,this paper proposes several improved carbon trading price forecasting models based on machine learning,data preprocessing algorithm,feature selection,artificial intelligence optimization algorithm et al.,which can theoretically make up for the deficiency of the existing carbon trading price forecasting research.From a practical point of view,accurate forecasting of carbon trading price is helpful to understand the characteristics of carbon trading price,to grasp the fluctuation law of carbon trading price,and to lay a foundation for establishing a stable price pricing mechanism.In addition,accurate carbon trading price forecasting has important guiding role for relevant participants in the carbon trading market.Moreover,accurate forecasting of carbon trading price can also help to grasp the dynamics of carbon trading market and provide theoretical reference for changing the upper limit of carbon emissions.The main shortcomings of this paper are as below:(1)There are still some parameters in the proposed carbon trading price forecasting model that need to be predefined manually.In the future research,the problem of parameter setting in the forecasting model needs to be further improved.(2)The calculation time of the model of the proposed forecasting model may increase due to the fusion of multiple algorithms.However,with the development of science and technology,it is believed that this problem can be solved to some extent.(3)The prediction model established in this paper did not consider the application of deep learning in carbon trading price prediction.In future research,the application of deep learning in carbon trading price forecasting will be further explored. |
