统计与数据科学学院知识库

理想状态下的数据一般具有线性、平稳性及复杂性低等特点，数据的准确预测，不仅可以为投资者提供决策支持，也可以让政府制定相关政策有参考依据。然而，现实中的数据由于外部各种因素的影响而呈现出复杂性高的特性。此外，再加上突发事件的影响，使数据的预测变得越来越困难，因此寻找一种可靠且有效的方法来预测数据至关重要。

当前金融数据预测的方法主要有传统的计量经济方法、人工智能方法和分解集成方法。传统的计量经济方法在非线性、非平稳性及一些复杂数据的处理中无能为力。人工智能方法通过让机器模拟人类智力来解决一些比较复杂的任务，从而获得更准确的预测，但人工智能方法存在参数敏感、容易陷入局部极小值、过度拟合等问题。分解集成方法是当前研究的主流方法，它将复杂的数据分解为易于描述的简单分量，从而降低建模的难度，提高模型的预测性能，达到“化繁为简，各个击破”的目的。但是，现有的分解集成方法，当分量过多对各个分量分别进行预测时，会增加计算成本，最终在结果集成时可能会出现误差累加的问题。

针对以上问题，本文基于“先分解后集成”的思想，从数据的分解、分量的重构和预测方法的优化三个方面出发，构建基于“分解-重构-集成”范式的预测方法，在此基础上开展相关实证研究：基于集合经验模态分解、重构和粒子群优化最小二乘支持向量机的美元兑人民币汇率预测；基于二次分解（CEEMDAN- CEEMDAN）、重构、混沌麻雀搜索算法优化KELM的比特币价格预测；基于二次分解（ICEEMDAN-EMD）、二次重构及混沌麻雀搜索算法优化KELM的原油期货价格预测。利用误差评价指标和DM检验，与其他多种模型的预测结果对比，实证结果表明，本文构建的模型预测结果更好，精度更高。

The data in the ideal state generally has the characteristics of linearity, stability and low complexity. The accurate prediction of the data can not only provide decision support for investors, but also provide a reference basis for the government to formulate relevant policies. However, the real data show the characteristics of high complexity due to the influence of various external factors In addition, coupled with the impact of emergencies, the prediction of data becomes more and more difficult. Therefore, it is very important to find a reliable and effective method to predict data.

At present, the methods of financial data prediction mainly include traditional econometric methods, artificial intelligence methods and decomposition integration methods Traditional econometric forecasting methods are powerless in the processing of nonlinearity, nonstationarity and some complex data Artificial intelligence methods solve some complex tasks by making machines simulate human intelligence, so as to obtain more accurate prediction. However, artificial intelligence methods are sensitive to parameters, easy to fall into local minimum, over fitting and so on Decomposition integration method is the mainstream method in current research. It decomposes complex data into simple components that are easy to describe, so as to reduce the difficulty of modeling, improve the prediction performance of the model, and achieve the purpose of "turning complexity into simplicity and breaking each one" However, in the existing decomposition integration methods, when there are too many components to predict each component separately, it will increase the calculation cost, and finally there may be the problem of error accumulation in the result integration.

Aiming at the above problems, based on the idea of "decomposition first and then integration", this paper constructs a prediction method based on the "decomposition-reconstruction-integration" paradigm from three aspects: data decomposition, component reconstruction and prediction method optimization. On this basis, relevant empirical research is carried out: Based on set empirical mode decomposition Reconstruction and particle swarm optimization of least squares support vector machine to predict the exchange rate of US dollar against RMB; Based on quadratic decomposition (CEEMDAN-CEEMDAN), reconstruction and chaotic sparrow search algorithm, kelm's bitcoin price prediction is optimized; KELM's crude oil futures price forecast is optimized based on ICEEMDAN-EMD, quadratic reconstruction and chaotic sparrow search algorithm. Using the error evaluation index and DM test, compared with the prediction results of other models, the empirical results show that the prediction results of the model constructed in this paper are better and more accurat.