基于多Agent传动关系的股市趋势预测

Prediction of Stock Market Trend Based on Multi-Agent Transmission Relationship

股市趋势预测是机器学习领域中一个具有挑战性的任务。由于一些因素对于股市的影响是动态且不确定的,导致股市趋势难以预测。针对已有方法在股市预测时存在的灵敏性差、适应力弱等问题,从快变量和慢变量的传动关系出发,利用Agent技术对股市中的快周期和慢周期进行联合建模,提出一种多Agent传动影响图(MATID)股市趋势预测方法。给出股市中快周期和慢周期的划分标准,并引入周期能量的概念;通过对相关趋势指标的特征融合,给出周期能量的量化计算方法;通过分析快周期和慢周期的动态作用过程,给出传动因子的表示方法;将快周期和慢周期分别对应成不同的Agent,利用多Agent影响图模型建模快周期和慢周期的传动过程;利用股市振子模型表示快Agent和慢Agent之间的传动效用,利用联合树的自动推理技术对股市趋势进行预测。在不同样本数量和不同股市趋势下进行实验,结果表明,与门控循环单元、S-LSTM和Hybrid-RNN预测方法相比,MATID方法预测精确率提升1.5%~7.0%,召回率提升5.4%~6.7%,F1值提升3.7%~6.2%,具有良好的灵敏性和适应力。

Stock market trend prediction is a challenging problem in the field of machine learning.Owing to the dynamic and uncertain nature of some factors affecting the stock market,changes in stock market trends are difficult to predict.In response to the issues of poor sensitivity and weak adaptability in stock market prediction,starting from the transmission relationship between fast and slow variables,a Multi-Agent Transmission Influence Diagram(MATID)stock market trend prediction method is proposed using Agent technology to jointly model the fast and slow cycles in the stock market.First,the classification criteria of fast and slow periods in the stock market are provided,and the concept of period energy is introduced.Subsequently,a quantitative calculation method for periodic energy through feature fusion of relevant trend indicators is provided;the transmission factor representation method is provided by analyzing the dynamic action process of fast and slow periods.Thereafter,fast and slow periods are corresponded to different Agent,and a multi-Agent influence diagram model is used to model the transmission process of the fast and slow cycles.Furthermore,the stock market oscillator model is used to represent the transmission utility between fast and slow Agent.Finally,the automatic inference technique of a joint tree is used to predict the stock market trend.Experimental results under different sample sizes and different stock market trends demonstrate that compared with Gated Recurrent Unit(GRU),Stacked Long Short-Term Memory(S-LSTM),and Hybrid Recurrent Neural Network(Hybrid-RNN)prediction methods,the MATID method has good sensitivity and adaptability with approximately 1.5%‒7.0%,5.4%‒6.7%,and 3.7%‒6.2%improvements in prediction,recall,and F1 value,respectively.