基于BiLSTM-AM-ResNet组合模型的山西焦煤价格预测  

Forecast of Shanxi coking coal price based on BiLSTM-AM-ResNet combined model

作　　者：樊园杰[1] 睢祎平 张磊[3] FAN Yuanjie;SUI Yiping;ZHANG Lei(School of Business,Shanxi Datong University,Datong,Shanxi 037009,China;College of Resources and Environmental Engineering,Guizhou University,Guiyang,Guizhou 550025,China;School of Coal Engineering,Shanxi Datong University,Datong,Shanxi 037003,China)

机构地区：[1]山西大同大学商学院,山西省大同市037009 [2]贵州大学资源与环境工程学院,贵州省贵阳市550025 [3]山西大同大学煤炭工程学院,山西省大同市037003

出　　处：《中国煤炭》2025年第3期42-51,共10页China Coal

基　　金：2023年山西省高等学校科技创新项目资助(2023L255);山西省大同大学2022年度云冈学专项资助(2022YGZX026)。

摘　　要：煤炭作为我国重要的基础能源,其价格的波动会直接影响国民经济发展与能源市场稳定,因此对煤炭价格进行预测具有重要意义。针对我国煤炭价格受政策与供求关系影响大、多呈现非线性的变化趋势,且目前存在的煤价预测方法存在滞后性大等问题,以山西焦煤价格为研究对象,分析影响煤炭价格的多种因素,并利用先进的人工智能机器学习算法来解决煤价预测问题。综合双向长短期记忆网络、注意力机制和残差神经网络的优势,构建双向长短期残差神经网络(BiLSTM-AM-ResNet)进行山西焦煤价格预测实验。采集2012-2023年的山西焦煤价格周度数据作为实验数据,对其进行空缺值处理和归一化处理,绘制相关系数热图并确定模型输入特征类型,进而简化模型并提高预测准确率与预测速度。通过模型预测实验得出,经BiLSTM-AM-ResNet模型预测的山西焦煤价格与实际煤价的发展趋势有着较高的线性拟合性,且预测结果与真实煤价在数值上非常接近,预测准确率达到了95.08%。Coal is an important basic energy in China,the fluctuation of coal price directly affects the stability of energy market and the development of national economy.Therefore,it is of great significance to predict coal price.Aiming at the problems of non-linearity change trend of domestic coal price due to the great influence by policies and supply-demand relationship,and the significant lag of current coal price forecast methods,taking Shanxi coking coal price as the research object,various factors affecting coal price were analyzed,and advanced artificial intelligence machine learning algorithm was used to solve the problem of coal price forecast.By combining the advantages of bidirectional long-term and short-term memory network(BiLSTM),attention mechanism(AM)and residual neural network(ResNet),a bidirectional long-term and short-term memory residual neural network(BiLSTM-AM-ResNet)was established for Shanxi coking coal price forecast experiment.The weekly price data of Shanxi coking coal from 2012 to 2023 were collected as experimental data,and the vacancy value processing and normalization processing of the data were carried out,the correlation coefficient heat map was drawn to determine the input feature type of the model to simplify the model and improve the forecast accuracy and speed.Through the model forecast experiment,it was concluded that the Shanxi coking coal price forecasted by the BiLSTM-AM-ResNet model had a high linear fitting with the development trend of the actual coal price,and the forecast result was very close to the real coal price in value,and the accuracy rate reached 95.08%.

关 键 词：焦煤价格预测 长短期记忆网络 注意力机制 残差神经网络 相关性分析 

分 类 号：TD-9[矿业工程]