区间映射规则下的时间序列相似形态搜索算法——基于改进的正则化损失函数

Time Series Similar Morphology Search Algorithm under Interval Mapping Rules:Based on Improved Regularization Loss Function

时间序列数据是一种随机过程,历史的波动趋势在不同的时期看来往往似曾相似。本文使用用可解释性的符号来刻画时间序列变化形态,改进了基于符号聚合相似的搜索模型,在原始搜索模型中引入改进的参数优化准则HIC,并提供了将字符转义为数值的变换方法,用于度量两个形态间的相似程度。结果表明,改进的模型实现了字符、数值的相互转化,且满足距离下界原理;参数的优化准则稳健的提高了模型的搜索精,有效的降低了算法复杂度。

Time series data is a kind of stochastic process.The trend of historical volatility seems to be similar in different periods.In this paper,we use interpretive symbols to depict the time series variation,improve the similar search model based on symbolic aggregation,introduce the improved parameter optimization criterion HIC into the original search model,and provide the transformation method of translating characters into numerical values,to measure the similarity between the two forms.The results show that the improved model realizes the mutual transformation of characters and values and satisfies the lower bound principle of distance.The optimization criterion of parameters steadily improves the searching precision of the model and reduces the complexity of the algorithm effectively.