| 摘要: | 在多元時間序列上探索事件的趨勢，挖掘趨勢之間的相連關係，對業者來說可能可以發現新的異常發生時所應該關注的其他事件，進而加以調整機台，舉例來說，我們能夠在機台發生張力劇烈下降時，透過規則的分析得知這樣的情況常常跟隨著劇烈的某作用輪轉速上升或是呈現某情況的起伏。我們希望透過我們所提的方法，可以讓往後公司進行機台調整時，能多方考慮可能影響的因素，讓機台更穩健運作。
然而傳統上的關聯規則以處理類別資料為主，資料的前後順序是被忽視的。後來當研究上欲把關聯規則應用在工程數值資料時，常見的處理方式是利用Piecewise Aggregate Approximation (PAA)、Symbolic Aggregate Approximation (SAX)的方法，分別是利用PAA將資料進行降維，以減少關聯規則運算中所需的龐大作用時間，再利用SAX對資料進行符號化，以利關聯規則所需的交易集資料的製作。但是和傳統關聯規則中資料型態不同的是，在多元時間序列資料上，資料數值的前後發生順序往往代表一定的意義，而且前後筆資料可能是有連帶關係的，若降維可能會稀釋甚至忽略掉資料本身的特性。
而在解決這部分問題時，舉例來說，A事件是先發生a再發生b，B事件是先發生b再發生a，本研究希望能在時序資料中實現這個欲區分A = (a, b) 和B = (b , a)差異的概念，並且區分出a事件是來自於A，b事件是來自於B，也就是區分資料欄位之間的不同，本研究中透過不同的交易集資料建立手法，以利關聯規則在作分析時可以更進一步關注資料集上的欄位差異，進而整合，做資料欄位上趨勢的分析，讓管理者可以根據分析結果，進行適當的維修工作或其他決策。
;Exploring the trend of events on a multivariate time series and digging up the connection between the trends, it is possible for the engineer to find other events that could be concerned when a new anomaly occurs, and then adjust machine, for example, we can know that when the tension of the machine drops sharply, it often follows a sharp increase in the speed of wheel. We hope that there are more adjustment method can be used through the idea we proposed.
However, the traditional association rules mainly deal with category data, such as transaction data of stores, and the order of the data is ignored. Recently, it often to find the Piecewise Aggregate Approximation (PAA) and Symbolic Aggregate Approximation (SAX) is applied to engineering numerical data, which uses PAA to reduce the dimensionality of the data to reduce the huge process time required in the calculation of association rules, and then uses SAX to symbolize the data to present the transaction set data required by the association rules. Unlike the data types in traditional association rules, the order of occurrence of the data values often represents a certain meaning on multivariate time series data, and it may be related between previous one and the next one. If the dimension is reduced, it may be diluted or even ignored the characteristics of the data itself.
We hope to achieve the idea to distinguish which the event of A = (a, b) and B = (b, a) are different, and also distinguish the difference between data fields in our research. Different transaction set data was established to facilitate the association rules about the field differences on the data set in our research. The analysis based on our method allows managers to not only find out the field trends that affect each other, but also use the method of our research to make the engineer know more about the connection between the machine columns and adjust the machine for proper maintenance work or other decisions. |
