【作者】 姚凯源； 陆荣秀； 杨辉； 张昊正； 徐伟进；

【Author】 KaiYuan Yao;Rongxiu Lu;Hui Yang;Haozheng Zhang;Weijin Xu;School of Electrical and Automation Engineering, East China Jiaotong University;Key Laboratory of Advanced Control and Optimization of Jiangxi Province;

【机构】 华东交通大学电气与自动化工程学院； 江西省先进控制与优化重点实验室(华东交通大学)；

【摘要】 鉴于稀土生产属于复杂长周期流程工业,市场响应及生产调度周期长,准确预测稀土产品价格的长期变化趋势更具实际价值.为此,提出了一种基于随机森林改进的长时间序列预测模型SRF.该模型通过重构决策树节点划分机制实现改进:首先,在分支节点划分过程中,采用LCSS算法计算时间序列间的形态相似性,替代传统平方误差准则,构建基于时序相似度的信息增益评价函数;其次,以中心序列作为叶子节点输出表征,增强模型对价格波动趋势的解析能力.通过四个具有代表性的稀土产品价格数据进行实验对比分析,结果表明本文提出的SRF模型具有更高的预测精度,为处理长时间序列预测任务提供了一种新的有效途径.更多还原

【Abstract】 Considering that rare earth production belongs to complex long-cycle process industry, and the market response and production scheduling cycle are long, it is more practical to accurately predict the long-term trend of rare earth product prices. To this end, we proposes a Long Sequence Time-Series Forecasting model SRF based on improved random forest. The model is improved by reconstructing the decision tree node partitioning mechanism: Firstly, in the process of branching node partitioning, the LCSS algorithm is used to calculate the morphological similarity between time series, replacing the traditional squared error criterion, and constructing an information gain evaluation function based on time series similarity; Secondly, the central sequence is used as the output representation of leaf nodes to enhance the model’s ability to analyze price fluctuation trends. Experimental comparisons and analyses using four representative rare earth product price datasets demonstrate that the proposed SRF model exhibits higher prediction accuracy, providing a new and effective approach for handling long time series prediction tasks.更多还原