基于机器学习方法的我国天然气进口预测

Natural Gas Import Forecast of China Based on Machine Learning Method

在“双碳”政策引导下,大力使用低碳、清洁的天然气资源是我国能源转型的重要举措,我国天然气供应主要依赖进口,准确预测天然气进口趋势对实现“双碳”目标具有重大意义。本文采用曲线投影寻踪动态聚类模型从经济发展、人口、天然气行业和能源消费4个方面确定天然气进口影响因素评价指标体系,分别构建多变量灰色GM(1,N)模型、支持向量机回归(SVR)、卷积神经网络(CNN)3种机器学习模型并对2006~2020年我国天然气进口数据进行拟合,对比拟合精度选取最优模型进行预测,以降低预测风险。研究结果表明:(1)曲线投影寻踪动态聚类模型不受指标数量和样本容量限制,按照影响因素的重要程度进行排序构建了天然气进口预测评价指标体系,剔除人为任意性因素的影响,实现预测指标体系的客观性;(2)CNN模型对天然气进口量的时间序列预测具有较高的预测精度;(3)2021~2026年我国天然气进口量呈上升趋势,增速下降,2026年天然气进口量突破2000亿立方米,预测结果与当前国内外能源政策导向吻合,最后针对预测结果提出相关建议。研究结论科学、可靠,可为我国天然气进口及风险管理提供一定的科学依据。

Under the guidance of the“double carbon”policy,vigorously using low-carbon and clean natural gas resources is an important measure of China's energy transformation.China's natural gas supply is mainly dependent on imports.Accurate prediction of natural gas import trend is of great significance to achieve the“double carbon”goal.The evaluation index system of influencing factors of natural gas import is determined from four aspects of economic development,population,natural gas industry and energy consumption by using curve projection pursuit dynamic clustering model.Three machine learning models,including multi-variable grey GM(1,N)model,support vector machine regression(SVR)and convolutional neural network(CNN),are constructed respectively to fit the data of China's natural gas import from 2006 to 2020,and select the optimal model to predict the accuracy of comparison,so as to reduce the prediction risk.The results show that:(1)the curve projection pursuit dynamic clustering model is not limited by the number of indicators and sample size,and the prediction and evaluation index system of natural gas import is constructed according to the importance degree of influencing factors,and the influence of artificial arbitrary factors is eliminated to achieve the objectivity of the prediction index system.(2)CNN model has high prediction accuracy for the time series of natural gas imports.(3)China's natural gas imports will increase from 2021 to 2026,and the growth rate will decrease.In 2026,the natural gas imports will exceed 200 billion cubic meters.The forecast results are consistent with the current domestic and foreign energy policy guidance.Finally,some suggestions are put forward for the prediction results.The conclusions are scientific and reliable,which can provide a scientific basis for natural gas import and risk management in China.