人工智能对金融时间序列模式的识别——股价模式的DTW研究

IDENTIFICATION OF FINANCIAL TIME SERIESPATTERN BY AI: DTW RESEARCH ON STOCKPRICE PATTERN

陈瀚谋

11649033

硕士

金融

向巨

2018-05-31

2018-7-11

哈尔滨工业大学

深圳

模式识别是人工智能的重要分支，其广泛应用于语音识别中。孤立词汇的识别是语音识别的重要组成部分，其方法是将语音信号转化为时间序列数值，再运用动态时间规整 (Dynamic Time Warping, DTW)法识别长度不同的时间序列的相似性，从而判断语音词汇的相似性。股票价格数据也是时间序列数据，语音识别的技术可以运用于股票价格序列的识别。由于股价序列的特定模式类似于语音中的孤立词汇，因此本文选择了DTW法来识别股票价格序列。本文选取上证指数历史数据与4个模板序列作为查询序列，运用快速动态时间规整（FDTW）法挖掘与查询序列挖掘相似度较高的子序列，在挖掘到足够数量的相似子序列的前提下，识别的成功率在65%想，且运用经过识别的子序列预测股价的成功率也在55%以上，超出预期。

Pattern recognition is an important branch of AI, and it is widely used in speech recognition. The recognition of isolated words is an important part of speech recognition. The method is to transform the speech signal into the time sequence value, and then use the Dynamic Time Warping (DTW) to identify the similarity of the time series with different length, so as to judge the similarity of the speech words.Stock price data are also time series data, and speech recognition technology can be applied to stock price sequence recognition. Because the specific pattern of the stock price sequence is similar to the isolated vocabulary in the voice, this paper chooses the DTW method to identify the stock price sequence.In this paper, the history data of Shanghai stock index and 4 template sequences are selected as query sequences, and fast dynamic time warping (FDTW) mining and query sequences are used to mine the subsequences with higher similarity. The success rate of recognition is more than 65% on the premise of a sufficient number of similar subsequences, and the results of the test are ideal, and The success rate of using the identified sub sequence to predict stock price is over 55%, exceeding expectations.

人工智能 模式识别 DTW算法

AI pattern recognition DTW algorithm

Chinese

联合培养

南方科技大学

陈瀚谋. 人工智能对金融时间序列模式的识别——股价模式的DTW研究[D]. 深圳. 哈尔滨工业大学,2018.