摘要
通过探索生鲜价格波动研究,构建了基于STL-XGBoost-KDE区间预测模型。对于单STL模型难以分离灵活的趋势和季节性数据的问题,引入Loess方法进行处理再分解,将分解得到的分量与原始价格数据作为输入,选择灵活性和可扩展性的XGBoost模型进行训练,得到分量并相加得到预测值,对预测误差使用拟合精度高的高斯核密度估计(KDE)来估计其概率分布函数。对于一定置信水平下,最后计算价格预测区间。经过比较评价指标,构建的STL-XGBoost-KDE区间预测模型比其他单模型和组合模型在点预测和区间预测的精准度上都有明显的提高。
Through the exploration of fresh produce price fluctuations,an interval prediction model based on the STL-XGBoost-KDE combination was constructed.To address the issue of the difficulty faced by the single STL model in flexibly separating trend and seasonal data,the Loess method was introduced for further decomposition.The decomposed components,along with the original price data,were served as inputs,and the flexible and extendable XGboost model was selected for training.The components were summed to gain the predicted values.To estimate the the probability distribution function of the prediction error,the Gaussian kernel density estimation(KDE)were used.Finally,at a given level of confidence,the price prediction interval were calculated.After comparing the evaluation indicators,the STL-XGBoost-KDE interval prediction model constructed in this article has significantly improved the accuracy of point prediction and interval prediction compared to other single and combination models.
作者
陈兴琪
詹棠森
CHEN Xingqi;ZHAN Tangsen(School of Information Engineering,Jingdezhen Ceramic University,333403,Jingdezhen,Jiangxi,PRC)
出处
《江西科学》
2025年第1期126-131,共6页
Jiangxi Science
基金
国家自然科学基金项目(71763013)
江西省教育厅学位办项目(JXYJG-2023-160)。
关键词
时间序列分解
区间预测
高斯核密度估计
累积分布函数
time Series decomposition
interval prediction
Gaussian Kernel Density Dstimation
cumulative distribution function
