基于邻域互信息的组合预测最优子集选择算法
Novel optimal sub-models selection algorithm for combination forecasting based on neighborhood mutual information
吕兴 1李倩 1张大斌 2曾莉玲 1凌立文3
作者信息
- 1. 华南农业大学数学与信息学院,广东广州 510642
- 2. 华南农业大学数学与信息学院,广东广州 510642;华南农业大学 乡村振兴研究院,广东 广州 510642;广东白云学院 大数据与计算机学院,广东广州 510450
- 3. 华南农业大学数学与信息学院,广东广州 510642;华南农业大学 乡村振兴研究院,广东 广州 510642
- 折叠
摘要
为在候选模型集中高效地选择时间序列组合预测的最优子集,提出一种CSPSO-NMI-MRMR最优子集选择算法.利用邻域互信息(neighborhood mutual information,NMI)度量相关性和冗余度,避免数值型数据的离散化,按最大相关最小冗余原则(minimal redundancy and maximal relevance,MRMR)筛选最优子集;邻域互信息中的邻域参数与子集选择效果密切相关,采用CSPSO算法寻找最优邻域参数,充分利用布谷鸟算法(cuckoo search,CS)和粒子群优化算法(par-ticle swarm optimization,PSO)的优势,兼顾搜索效率和全局搜索能力;在寻参过程中设计一种淘汰策略,优化邻域参数的寻优区间并淘汰部分单模型,减少计算量.仿真结果表明,所提方法在预测精度、运行时间和稳健性上效果更优.
Abstract
To select the optimal sub-models from the candidate models efficiently,an algorithm named CSPSO-NMI-MRMR was proposed in combination forecasting of time series.The neighborhood mutual information(NMI)was used to measure the relevance and redundancy,avoiding the discretization of numerical data.The optimal subsets were selected according to the principle of minimal redundancy and maximal relevance(MRMR).The result of subset selection was closely related to the neigh-borhood parameter in NMI.The CSPSO algorithm was used to search for optimal neighborhood parameter.The advantages of cuckoo search(CS)algorithm and particle swarm optimization(PSO)algorithm were utilized,giving consideration to global search ability and high search efficiency.To reduce the computational cost,an elimination strategy was designed to optimize the optimization interval of neighborhood parameters and eliminate some single models.Simulation results show that the algorithm is more effective in prediction accuracy,running time and robustness compared to other methods.
关键词
时间序列/组合预测/子模型选择/邻域互信息/参数优化/启发式算法/布谷鸟算法Key words
time series/combination forecasting/sub-model selection/neighborhood mutual information/parameter optimiza-tion/heuristic algorithm/cuckoo search algorithm引用本文复制引用
基金项目
国家自然科学基金面上项目(71971089)
国家自然科学基金青年基金(72001083)
广东省自然科学基金面上基金(2022A1515011612)
广东省普通高等学校重点领域专项(2020ZDZX3009)
出版年
2024
NETL
NSTL
维普
万方数据