基于一维搜索的ICA自适应算法及其在股票分析中的应用

ICA Adaptive Algorithm Based on Line Search and its Application in the Stock Data Analysis

在固定步长的ICA极大似然估计自适应算法的基础上,通过一维搜索引入了步长修正方案,使新算法可在收敛速度和稳定状态时的失调误差这两个性能指标上达到最佳结合点,具有较好的时变系统跟踪能力。仿真结果证实了本文所提出的算法可以有效地提高ICA的自适应性,能够更准确地完成盲源分离。在此基础上将算法用在时变性很强的股票数据上,以验证该算法的有效性和可行性。

This paper introduces step-size modified strategy on the base of fixed step size algorithm of ICA by maximum likelihood estimation through line search. Then the new algorithm attains the best integrated point in the two performance indices between the convergence speed and the mismatch error at the steady state. And it has the preferable ability to track a changing mixing environment. The simulation results demonstrate that this algorithm improves the adaptability of ICA effectively and performs blind signal separation accurately. Then the new algorithm is used to analyze the time-variable stock data to validate effectiveness and feasibility of the proposed algorithm.