基于机器学习的我国天然气进口量预测及其运输安全评价

Prediction of China's Natural Gas Import Volume and Transportation Safety Evaluation Based on Machine Learning

在线阅读下载PDF

导出

摘要天然气作为一种高效、低碳、清洁的化石能源,具有易存储、运输和储量丰富的特性,被视为其他化石能源的最佳替代品之一。在确保我国天然气供应的连续性及实现碳达峰与碳中和目标的过程中,准确预测我国天然气进口量并对其运输安全进行评价显得尤为关键。针对单一特征选择方法特征筛选结果不稳定的问题,本文提出了一种基于多属性决策的特征选择方法,并根据特征选择结果建立高斯过程回归模型对我国天然气进口量进行预测。同时,提出了一种结合熵权与t-SNE降维的K均值聚类模型,并将其应用于评估我国天然气进口运输安全。结果表明:本文所提出的基于多属性决策的特征选择方法所得到的特征重要程度排序更加稳定,对我国天然气进口量影响最大的3个指标分别是:能源消费总量中天然气占比、居民消费水平和国内天然气管道长度;高斯过程回归模型对我国天然气进口量预测的绝对百分比误差均在5%以下,预测精度高;与传统t-SNE降维相比,结合熵权的t-SNE降维方法具有更低的KL散度,其降维结果更贴近原始数据的分布;结合熵权与t-SNE降维的K均值聚类结果可信度更高,我国进口天然气运输安全状态越来越好,并且对其影响最大的3个指标分别是:自有船队运力供给、管道气年运输量和LNG年运输量。 Natural gas,as an efficient,low-carbon,and clean fossil energy source,is recognized as one of the best alternatives to other fossil fuels due to its characteristics of being easy to store,transport,and abundant in reserves.In the process of ensuring the continuity of China's natural gas supply and achieving the goals of peaking carbon emissions and carbon neutrality,accurately predicting China's natural gas imports and assessing the safety of its transport are particularly crucial.This paper addresses the instability of feature selection results from single feature selection methods by proposing a multi-attribute decision-making-based feature selection method,and based on the results of feature selection,a Gaussian Process Regression model is established to predict China's natural gas import volume.Additionally,a K-means clustering model combining entropy weighting and t-SNE dimension reduction is proposed and applied to assess the safety of China's natural gas import transportation.The results show that the feature selection method based on multi-attribute decision-making provides a more stable ranking of feature importance.The three most influential indicators on China's natural gas import volume are the proportion of natural gas in total energy consumption,residential consumption level,and domestic natural gas pipeline length.The Gaussian Process Regression model predicts China's natural gas import volume with an absolute percentage error below 5%,indicating high prediction accuracy.Compared to traditional t-SNE dimension reduction,the entropy-weighted t-SNE method achieves lower KL divergence,with its dimension reduction results being closer to the distribution of the original data.The K-means clustering results combining entropy weighting and tSNE dimension reduction are more credible.The safety status of China's imported natural gas transportation is improving,with the three most significant indicators being the supply capacity of owned fleet,annual pipeline gas transportation volume,and annual LNG transportation volume.

作者邹黎敏唐永欣 Zou Limin;Tang Yongxin(School of Mathematics and Statistics,Chongqing Technology and Business University,Chongqing 400067,China)

机构地区重庆工商大学数学与统计学院

出处《工业技术经济》北大核心 2025年第2期108-118,共11页 Journal of Industrial Technology and Economy

基金国家自然科学基金“矩阵酉不变范数不等式的理论及应用研究”(项目编号:12261030) 重庆市自然科学基金面上项目“基于机器学习的电能质量综合评价模型研究”(项目编号:CSTB2023NSCQ-MSX0223) 重庆工商大学研究生创新型科研项目“面板数据变系数固定效应模型的经验似然检验”(项目编号:yjscxx2024-284-247)。

关键词机器学习天然气进口量运输安全熵权特征选择高斯过程回归 t-SNE降维 K均值聚类 machine learning natural gas imports transportation safety entropy weight feature selection gaussian process regression t-SNE dimensionality reduction K-means clustering