基于正余弦算法和误差修正组合方法的汇率预测研究

Research of Exchange Rate Prediction Based on Combinatorial Approach of Sine Cosine Algorithm and Error Correction

针对汇率数据具有高波动性、非线性和高噪声等特点,提出了一种基于分解算法和误差修正的汇率时间序列预测方法(VMD-SCA-ELM-EC)。该方法首先利用变分模态分解(VMD)算法对原始数据进行分解,然后采用正余弦算法优化的极限学习机神经网络(SCA-ELM)对各个分量进行预测,并采用误差修正法(EC)拟合误差因素,最后将各项预测结果进行线性集成。选择英镑兑美元每日汇率序列进行分析预测,并将提出的模型与多种常见预测模型进行对比,实证分析得出,该方法在一步和多步提前预测中均取得更优的预测精度和方向准确率。

A time series prediction method for exchange rate data(VMD-SCA-ELM-EC)based on decomposition algorithm and error correction is proposed to address the problems of high complexity,nonlinearity and high noise of exchange rate data.The method first decomposes the raw data by using the variational mode decomposition(VMD)algorithm,then predicts each component by the extreme learning machine neural network optimized by the sine cosine algorithm(SCA-ELM),and fits the error factors with the error correction method(EC),and finally linearly integrates the results of each prediction.The daily exchange rate series of GBP/USD is selected for analysis and prediction,and the proposed model is compared with a variety of some common prediction models,and empirical analysis shows that this method achieves better prediction accuracy and directional accuracy in both one-step-ahead and multi-step-ahead.