PDF下载

[1]温泉,王宁,魏学华.基于MF-DFA与BorutaShap的天然气需求预测模型[J].油气储运,2024,43(10):1-15.
　WEN Quan,WANG Ning,WEI Xuehua.Natural gas demand forecast model based on MF-DFA and BorutaShap[J].Oil & Gas Storage and Transportation,2024,43(10):1-15.

点击复制

基于MF-DFA与BorutaShap的天然气需求预测模型

《油气储运》[ISSN:1000-8241/CN:13-1093/TE] 卷: 43 期数: 2024年10期页码: 1-15 栏目: 出版日期: 2024-10-25

Title:: Natural gas demand forecast model based on MF-DFA and BorutaShap

作者:: 温泉; 王宁; 魏学华

Author(s):: WEN Quan; WANG Ning; WEI Xuehua

关键词:: 天然气需求; 多重分形消除趋势波动分析; 随机森林; Sobol; BorutaShap; 蜜獾优化算法; 莱维飞行; XGBoost

Keywords:: Natural gas demand; MF-DFA; ; RF; ; Sobol; ; BorutaShap; ; HBA; ; Levy; ; XGBoost

文献标志码:: A

摘要:: 【目的】天然气需求受诸多因素影响，为了有效获取天然气月度需求时序数据的局部特征信息，提升天然气需求预测模型非线性拟合能力与预测精度展开研究。【方法】首先，引入多重分形消除趋势波动分析（Multi-Fractal Detrended Fluctuation Analysis, MF-DFA）对天然气月度需求时序数据进行分形研究。其次，采用二次插值法与随机森林（Random Forest, RF）插值法处理影响因素特征序列数据中时间粒度不一致与缺失情况。再次，选择极限梯度提升（eXtreme Gradient Boosting, XGBoost）模型，分别对插值前后原特征序列及经Boruta、SHAP、BorutaShap筛选后的新特征序列进行计算误差比对，以确定最佳特征序列筛选降维方式，进一步降低模型输入数据的维度与规模。最后，引入Sobol低差异序列、改进密度因子及莱维飞行策略，以提升蜜獾优化算法（Honey Badger Algorithm, HBA）种群初始化覆盖范围的均匀分布度、扩大迭代搜索范围及跳出局部最优的能力，从而增强改进HBA算法对XGBoost模型中决策树数量、决策树深度、学习速率等决定模型拟合能力的参数寻优效果。【结果】采用BorutaShap算法进行特征序列筛选降维最佳，新提出的预测模型预测精度优于对比模型，其MAPE、MAE、RMSE及R²分别为：2.87%、9.350 9、11.335 3及0.890 9。【结论】该方法适用于多种影响因素条件下的天然气需求预测工作，可为天然气行业发展规划决策提供参考依据。

Abstract:: [Objective] Natural gas demand is influenced by numerous factors. This study aims to effectively capture the local characteristics of monthly natural gas demand time series data, thereby enhancing the nonlinear fitting capability and prediction accuracy of natural gas demand forecasting models. [Methods] First, the Multi-Fractal Detrended Fluctuation Analysis (MF-DFA) was introduced to perform a fractal analysis of the monthly natural gas demand time series data. Next, quadratic interpolation and Random Forest (RF) interpolation methods were employed to address issues of inconsistent time granularity and missing data in the feature sequence of influencing factors. Then, the eXtreme Gradient Boosting (XGBoost) model was applied to compare the prediction errors of the original feature sequences before and after interpolation and the newly selected feature sequences screened by Boruta, SHAP, and BorutaShap algorithms. This comparison aimed to determine the optimal feature selection and dimensionality reduction method, thereby further reducing the dimensions and scale of the model's input data. Finally, the Sobol low-discrepancy sequence, an improved density factor, and the Lévy flight strategy were introduced to enhance the Honey Badger Algorithm (HBA) by improving the uniformity of the population initialization coverage, expanding the search range, and increasing the algorithm's ability to escape local optima. These enhancements were intended to improve the optimization of key parameters in the XGBoost model, such as the number of decision trees, tree depth, and learning rate, which are crucial for the model's fitting ability. [Results] The BorutaShap algorithm proved to be the best for feature selection and dimensionality reduction. The newly proposed prediction model demonstrated superior accuracy compared to the benchmark models, with MAPE, MAE, RMSE, and R²values of 2.87%, 9.3509, 11.3353, and 0.8909, respectively. [Conclusion] This method is suitable for predicting natural gas demand under various influencing factors and can provide valuable references for planning and decision-making in the natural gas industry.

相似文献/References:

[1]殷建成,袁宗明,岑康.天然气需求自适应优化组合预测模型的改进[J].油气储运,2005,24(10):17.[doi:10.6047/j.issn.1000-8241.2005.10.005]
　YIN Jiancheng,YUAN Zongming.Improving for the Model of Demand Selfadaption Optimization[J].Oil & Gas Storage and Transportation,2005,24(10):17.[doi:10.6047/j.issn.1000-8241.2005.10.005]

更新日期/Last Update: 2024-08-15