基于数据挖掘的天然气市场收益率预测

以2000—2018年美国天然气价格为研究对象,基于动态时间规整算法（DTW）、模拟退火算法（SA）、支持向量机模型（SVM）构建DTW|SVM|SA天然气价格预测组合模型,并在不同预测步长下将其与对照模型的预测结果进行对比,分别从预测精度和预测误差两方面对模型的预测性能进行评估。结果表明：利用模拟退火算法可以优化SVM模型的自由参数和混合模型的权重参数;DTW-SVM-SA组合预测模型在天然气价格收益率预测方面表现出良好的泛化能力,对比其他模型,其在不同步长上的预测精度均有显著提升,预测误差均有降低,是一种有效的天然气价格预测模型。DTW-SVM-SA组合预测模型不仅能够为政府进行宏观调控提供参考,而且可以帮助企业尤其是能源相关企业更好地预测和管理价格变动的风险。

FORECAST OF NATURAL GAS MARKET YIELD BASED ON DATA ANALYSIS

This paper, based on Americas natural gas price from 2000 to 2018, establishes DTW-SVM-SA(dynamic time warping-support vector machine-simulated annealing) combined model to forecast the natural gas price. It also compares the DTM-SVM-SA modeling results with comparison modeling results at different intervals, and evaluates its forecasting accuracy and error. Simulated annealing algorithm can optimize the free parameters of SVM and weight parameters of the combined model. DTM-SVM-SA model shows a good generalized capacity in forecasting natural gas yield, with outstandingly rising accuracy at difference intervals and decreasing error, which is an effective gas price forecasting model. DTW-SVM-SA model can not only provide references for governmental macroscopic controls, but also help enterprises better forecast and manage risks in price changes, especially for energy enterprises.