首页|基于LLM的金融市场波动率高频数据异常检测方法

基于LLM的金融市场波动率高频数据异常检测方法

扫码查看
原文链接
  • 国家科技期刊平台
  • NETL
  • NSTL
  • 万方数据
金融市场高频数据包括时间序列数据和其他宏观经济指标,通常具有高维特征.其处理需要更复杂的算法,易产生较高的模型过拟合风险.基于此,提出基于局部线性映射(Local Linear Mapping,LLM)的金融市场波动率高频数据异常检测方法,对各个高频数据目标的日平均序列数据进行标准化处理,在数据筛选时,使用标准化处理设定相关阈值,将不同维度的数据转化为相同的尺度,并利用连通图算法,将具有边连接的金融市场波动率高频数据划分至一个群组内,计算待检测高频数据阈值,采用局部线性映射,完成金融市场波动率高频数据异常检测.实验结果表明:所提方法在TPR为0.98时,ROC曲线稳定运行,贡献因子为1.287,重构误差为1.6％,能够以最快速度使训练集异常检测的损失值达到稳定.
Method of Detection of High-frequency Data Anomaly in Financial Market Volatility Based on LLM
High-frequency data in financial markets usually has high-dimensional characteristics, including time-series data and other macroeconomic indicators. The processing of high-dimensional data requires more sophisticated algorithms, leading to increased computational complexity and the risk of model over-fitting. In view of this, the anomaly detection method of financial market volatility high-frequency data based on LLM is proposed. Specifically, the method sets to standardization of the daily average sequence of each high-frequency data. While screening data, the method uses standardized processing to set the relevant threshold, transforms the data of different dimensions into the same scale, and applies the connected algorithm to put edge-connected financial market volatility high-frequency data into a same group. When calculating high-frequency data threshold to be detected, the method uses local linear mapping so as to bring the financial market volatility high-frequency data anomaly detection to an end. The experimental results show that when the proposed method is 0.98, the ROC curve displays a stable run and, with the contribution factor being 1.287, and the reconstruction error being 1.6％, the loss value of anomaly detection in the training set reaches a steady state at the fastest speed.

local linear mappingfinancial marketsvolatilityhigh-frequency dataanomaly detection

何远景、李光龙

展开 >

安徽工业经济职业技术学院财经学院,合肥 230000

安徽大学经济学院,合肥 230000

局部线性映射 金融市场 波动率 高频数据 异常检测

2021年安徽省省级质量工程项目教育部职业教育提质培优行动计划(2020-2023)安徽省职业与成人教育学会2022年度教育教学研究规划课题2020年院级质量工程项目

2021jpk028皖教秘高[2021]35号Azj20220912020yxxkc01

2024

常熟理工学院学报
常熟理工学院

常熟理工学院学报

影响因子：0.206
ISSN：1008-2794
年,卷(期)：2024.38(2)
  • 15