基于注意力机制的CNN-LSTM模型的碳市场价格预测  

Carbon market price prediction by CNN-LSTM model based on attention mechanism

作　　者：简明宇 任苏灵 王静 JIAN Mingyu;REN Suling;WANG Jing(School of Statistics and Data Science,Lanzhou University of Finance and Economics,Lanzhou 730020,China;School of Science,Lanzhou Jiaotong University,Lanzhou 730070,China)

机构地区：[1]兰州财经大学统计与数据科学学院,兰州730020 [2]兰州交通大学数理学院,兰州730070

出　　处：《哈尔滨商业大学学报(自然科学版)》2025年第2期226-232,共7页Journal of Harbin University of Commerce:Natural Sciences Edition

基　　金：国家自然科学基金青年基金项目(No.12101279);甘肃省自然科学基金(25JRRA978);甘肃省教育厅青年博士支持项目(2025QB-057)。

摘　　要：在以往研究的基础上,针对碳价格数据具有金融时间序列的不确定性及高噪声的特点,利用卷积神经网络以及LSTM模型对时间序列进行特征分量提取和序列预测,提出一种基于注意力机制的CNN-LSTM模型进行碳市场价格的预测.实验选取来自广州碳排放权交易中心(CEEX)公示的2014年5月4日~2024年4月30日的历史碳价交易数据对LSTM、LSTM-attention、CNN-LSTM以及CNN-LSTM-attention模型进行多种模型和多种特征输入变量的精度对比和图像分析.本文所提多变量CNN-LSTM-attention模型的总体拟合预测效果最好,其预测精度高于单变量预测模型以及基础对比模型,且模型具有较好的泛化性能.Based on previous studies,considering the uncertainty and high noise of financial time series of carbon price data,this paper utilized convolutional neural network and LSTM model to extract feature components and predict series of time series.Proposed a CNN-LSTM model based on attention mechanism for predicting carbon market price.In this experiment,historical carbon price trading data published by Guangzhou Carbon Emission Trading Center(CEEX)from May 4,2014 to April 30,2024 were selected for LSTM,LSTM-ATTENTION,CNN-LSTM,and CNN-LSTM-ATTENTION models.A precision comparison and image analysis were conducted among multiple feature input variables.It was found that the multi-variable CNN-LSTM-attention model proposed in this paper had the best overall fitting prediction effect,and its prediction accuracy was higher than that of the single variable prediction model and the basic comparison model,indicating better generalization performance of the model.

关 键 词：注意力机制 卷积神经网络 LSTM CNN-LSTM 碳市场价格 价格预测 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程]