基于SSA-LSSVM-KDE 的农产品价格区间预测
Interval Prediction of Agricultural Price Based on SSA-LSSVM-KDE
张学竞 1胡焕玲 1凌立文 1张大斌1
作者信息
- 1. 华南农业大学数学与信息学院,广州 510642
- 折叠
摘要
为了更好地把控农产品价格波动的波动范围,提高预测精度,文章基于分解集成思想,提出一种基于奇异谱分析(SSA)、最小二乘支持向量机(LSSVM)与核密度估计(KDE)的区间预测组合模型,简称SSA-LSSVM-KDE模型.首先针对SSA方法中窗口维度难以确定的问题,引入Cao方法优化SSA最小嵌入的窗口维度,通过奇异值分解重构出多条分量;其次,选择学习能力强的LSSVM,将各分量作为LSSVM的输入,得到组合预测输出;最后利用B-样条基的最小二乘交叉验证法(B-spline-LSCV)优化KDE模型,估计组合预测输出的不同区间误差概率分布函数,得到给定置信水平下的最终预测区间.为了验证提出模型的有效性,对小麦现货价和玉米现货价进行区间预测,与四个单模型、三个组合模型和四个分布函数进行多种预测性能评价指标的对比,结果显示提出的模型在点预测和区间预测的精度都得到了明显的提高.
Abstract
In order to better control the fluctuation range of agricultural price fluc-tuations and improve the prediction accuracy,this paper proposes a combined interval prediction model based on singular spectrum analysis(SSA),least squares support vector machine(LSSVM)and kernel density estimation(KDE)based on the idea of decomposition and integration,referred to as SSA-LSSVM-KDE model.Firstly,to solve the problem that the window dimension is difficult to be determined in SSA method,Cao method is introduced to optimise the window dimension of SSA min-imum embedding,and the multiple components are reconstructed by singular value decomposition;Secondly,LSSVM with strong learning ability is selected,and the components are used as inputs to the LSSVM to obtain the combined prediction outputs;and finally,KDE model is optimised by using least squares cross validation based on B-spline bases(B-spline-LSCV)optimised KDE model to estimate the er-ror probability distribution function for different intervals of the combined prediction output,and obtain the final prediction intervals at a given confidence level.In order to verify the effectiveness of the proposed model,interval forecast of wheat spot price and corn spot price are made,and compared with four single models,three combined models and four distribution functions with various forecasting performance evalua-tion criteria.The results show that the accuracy of the proposed model in both point and interval forecasts has been significantly improved.
关键词
农产品价格/奇异谱分析/最小二乘支持向量机/核密度估计/区间预测Key words
Agricultural product price/singular spectrum analysis/least squares support vector machine/kernel density estimation/interval prediction引用本文复制引用
基金项目
国家自然科学基金(71971089)
国家自然科学基金(72001083)
广东省自然科学基金(2022A1515011612)
广州市基础与应用基础研究项目(2023A04J0317)
出版年
2024
NETL
NSTL
万方数据