KR20190024328A - Device and method for calculating correct alignment path of time series data using improved dynamic time warping - Google Patents

Abstract

The present invention relates to a device for calculating a correct response path of time series data using improved dynamic time warping (DTW) and a method thereof. A response path calculation device comprises: a feature extraction unit which performs discrete wavelet transform on each of a first time series signal and a second time series signal, extracts a first scale portion and a first detail portion from the first time series signal, extracts a second scale portion and a second detail portion from the second time series signal, divides the first detail portion based on an intersection of the first time series signal and the first scale portion, and divides the second detail portion based on an intersection of the second time series signal and the second scale portion; and a response path calculation unit which calculates a first response path by applying DTW to each of the divided first detail portions and the divided second detail portions and calculates a response path between the first time series signal and the second time series signal by using the first response path.

Description

BACKGROUND OF THE INVENTION 1. Field of the Invention [0001] The present invention relates to an apparatus and a method for accurately calculating a corresponding path of time series data using an improved DTW,

The present invention relates to a corresponding path calculation apparatus for time series data, and more particularly, to a corresponding path calculation apparatus and method for calculating a correct corresponding path of time series data such as a semiconductor process signal using an improved DTW (Dynamic Time Warping) algorithm .

The DTW (Dynamic Time Warping) algorithm is a widely used algorithm for determining the similarity between two dynamic patterns (signals) having different lengths varying with time or speed. The DTW algorithm is used in the field of pattern recognition in the fields of data mining, gesture recognition, robots, speech recognition, and the like, but has the following problems.

First, when a corresponding path is calculated using the DTW algorithm, a singularity may occur. The singular point means a phenomenon in which several corresponding points included in one pattern are concentrated at one corresponding point included in another corresponding pattern (see FIG. 3).

Second, there is a problem that it is impossible to know whether the corresponding path between the two dynamic patterns calculated using the DTW algorithm is correct. For example, when the corresponding path is determined by the point where the rising point of one dynamic pattern corresponds to the falling point of another dynamic pattern in the two dynamic patterns, the DTW algorithm determines the corresponding path only by the relationship of only two points, Can be generated.

The present invention provides a method for matching two time series data by reflecting the shape of time series data.

United States Patent No. 8,880,352 (published April 4, 2014) United States Patent Publication No. 2015-0095390 (published on October 29, 2012)

The technical problem to be solved by the present invention is to perform an intuitive matching between time series data using an improved DTW in order to solve a problem that may occur when time series data is matched in a DTW.

The corresponding path calculation apparatus according to an embodiment of the present invention discrete wavelet transforms each of the first time series signal and the second time series signal to generate a first scale portion and a first detail portion from the first time series signal Extracts a second scale portion and a second detail portion from the second time series signal and divides the first detail portion based on an intersection of the first time series signal and the first scale portion, A characteristic extracting unit for dividing the second detail portion based on the intersection of the time series signal and the second scale portion, and a dynamic time warping (DTW) method for each of the divided first detail portion and the divided second detail portion, 1 corresponding path and calculates a corresponding path between the first time series signal and the second time series signal using the first corresponding path, It includes the mountain part.

The corresponding path calculation apparatus according to another embodiment of the present invention performs discrete wavelet transform on each of the first time series signal and the second time series signal to extract a first scale portion from the first time series signal, 2 scale section, dividing the first time series signal by an intersection of the first time series signal and the first scale section, and dividing the first time series signal by the intersection of the second time series signal and the second scale section, A first corresponding path is calculated by applying a DTW to each of the characteristic extracting unit and the divided first time series signal and the divided second time series signal to divide the time series signal into a first corresponding path, And a corresponding path calculator for calculating a corresponding path between the first time series signal and the second time series signal.

The corresponding path calculation method according to an embodiment of the present invention is performed in the corresponding path calculation apparatus, and comprises: (a) performing discrete wavelet transform on each of the first time series signal and the second time series signal, And extracting a second scale portion and a second detail portion from the second time series signal, and (b) extracting a first scale portion and a second scale portion from the second time- Extracting at least one second intersection point that is an intersection of the second time series signal and the second scale portion, (c) extracting the first detail portion from the at least one first intersection point, And dividing the second detail portion with respect to the at least one second intersection point; (d) adding a DTW to each of the divided first detail portions and the divided second detail portions, And calculating a corresponding path between the first time series signal and the second time series signal using the first corresponding path.

Also, a corresponding path calculation method according to another embodiment of the present invention is performed in the corresponding path calculation apparatus, and comprises: (a) performing discrete wavelet transform on each of the first time series signal and the second time series signal, (B) extracting at least one first intersection point that is an intersection of the first time-series signal and the first scale portion, extracting a second scale portion from the second time- Extracting at least one second intersection point that is an intersection of the first time series signal and the second time series signal, (c) dividing the first time series signal with reference to the at least one first intersection point, (D) dividing the second time series signal with respect to the second time series signal by dividing the first time series signal into a first time series signal and a second time series signal, And step (e), which comprises a step of calculating a corresponding path between the first wave form signal and the second time-series signal by using the first path corresponds.

According to the apparatus and method for calculating a corresponding path according to the embodiment of the present invention, it is possible to reduce a singular point generated when a corresponding path of a time series signal such as a process signal is calculated.

Further, there is an effect that it is possible to provide a better signal matching method by calculating a corresponding path with a reduced singularity.

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS In order to more fully understand the drawings recited in the detailed description of the present invention, a detailed description of each drawing is provided.
1 is a functional block diagram of a corresponding path calculation apparatus according to an embodiment of the present invention.
2 shows an example of a signal generated in a semiconductor process.
FIG. 3 is a view for explaining a singular point occurring in applying the DTW algorithm. FIG.
4 is a diagram for explaining a specific operation of the characteristic extracting unit and the corresponding path calculating unit shown in FIG.
5 is a flowchart illustrating a corresponding path calculation method performed by the corresponding path calculation apparatus shown in FIG.

It is to be understood that the specific structural or functional description of embodiments of the present invention disclosed herein is for illustrative purposes only and is not intended to limit the scope of the inventive concept But may be embodied in many different forms and is not limited to the embodiments set forth herein.

The embodiments according to the concept of the present invention can make various changes and can take various forms, so that the embodiments are illustrated in the drawings and described in detail herein. It should be understood, however, that it is not intended to limit the embodiments according to the concepts of the present invention to the particular forms disclosed, but includes all modifications, equivalents, or alternatives falling within the spirit and scope of the invention.

The terms first, second, etc. may be used to describe various elements, but the elements should not be limited by the terms. The terms may be named for the purpose of distinguishing one element from another, for example, without departing from the scope of the right according to the concept of the present invention, the first element may be referred to as a second element, The component may also be referred to as a first component.

It is to be understood that when an element is referred to as being "connected" or "connected" to another element, it may be directly connected or connected to the other element, . On the other hand, when an element is referred to as being "directly connected" or "directly connected" to another element, it should be understood that there are no other elements in between. Other expressions that describe the relationship between components, such as "between" and "between" or "neighboring to" and "directly adjacent to" should be interpreted as well.

The terminology used herein is for the purpose of describing particular embodiments only and is not intended to be limiting of the invention. The singular expressions include plural expressions unless the context clearly dictates otherwise. In this specification, the terms "comprises" or "having" and the like are used to specify that there are features, numbers, steps, operations, elements, parts or combinations thereof described herein, But do not preclude the presence or addition of one or more other features, integers, steps, operations, components, parts, or combinations thereof.

Unless defined otherwise, all terms used herein, including technical or scientific terms, have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs. Terms such as those defined in commonly used dictionaries are to be interpreted as having a meaning consistent with the meaning of the context in the relevant art and, unless explicitly defined herein, are to be interpreted as ideal or overly formal Do not.

Hereinafter, embodiments of the present invention will be described in detail with reference to the drawings attached hereto.

The present invention relates to algorithm improvement of DTW (Dynamic Time Warping) used to measure distance between time series data. In order to improve the erroneous correspondence path that occurs when DTW is applied to non-stationary time-series data whose mean and variance are changed, a scale portion of the time-series data and a detail portion of the time-series data using MODWT (Maximal Overlap Discrete Wavelet Transform) (detail) portion is extracted. Then, the intersection is extracted by comparing the time series data and the scale portion, and the sub-step of the detail portion is extracted based on the extracted intersection point. It is possible to calculate the corresponding path by applying the DTW to the divided detail portion and express the calculated corresponding path in the time series data to calculate the correct corresponding path between the time series data rather than the DTW.

FIG. 1 is a functional block diagram of a corresponding path calculating apparatus according to an embodiment of the present invention. FIG. 2 shows an example of a signal generated in a semiconductor process, FIG. 3 shows a singular point generated in applying the DTW algorithm Fig.

1 to 3, the corresponding path calculation apparatus 10 includes a feature extraction unit 300 and a corresponding path calculation unit 500. According to an embodiment, the corresponding path calculation device 10 may further include at least one of the signal receiving unit 100 and the DB (database) 700.

The corresponding path calculation device 10 can calculate a corresponding path of two signals (time series data or time series signals) that are received or stored in advance. One of the two signals may refer to a reference signal, and the other may refer to a process signal to be compared.

A signal used in the present invention may be, for example, a semiconductor process signal as a pattern moving according to time or speed, a time series signal or time series data. Referring to FIG. 2 showing an example of a signal generated in a semiconductor process, the semiconductor process signal has a moving average and a non-uniform dispersion, and it is known that the length of the measurement portion differs for each signal, have. For this reason, the use of indiscreet DTW algorithm can cause singularity. The singularity means that non-intuitive matching takes place as shown in FIG. 3 when matching time series data to other time series data. Such a singularity may cause the performance of the algorithm to deteriorate at the signal matching of the process.

The signal receiving unit 100 can receive at least two time series signals (time series data) for calculating a corresponding path through a predetermined input device or a wired / wireless network. Each of the at least two time series signals may refer to a reference signal and / or a comparison target signal.

In addition, the signal receiving unit 100 may store each of the received signals in the DB 700. According to an embodiment, the at least two time series signals may be stored in the DB 700 in advance. In this case, the property extraction unit 300 can perform a signal conversion operation on a signal stored in the DB 700. [

The feature extraction unit 300 may extract the feature of each of the two signals. The property extraction unit 300 can extract the characteristics of each of the input signals using a predetermined algorithm, for example, a DWT (Discrete Wavelet Transform) algorithm, in particular, a MAXWT (Maximal Overlap Discrete Wavelet Transform) algorithm. That is, the feature extraction unit 300 may perform discrete wavelet transform on each of the input signals.

Specifically, the feature extractor 300 performs wavelet transformation of the transformation target signal (time series data) using MODWT, thereby extracting a scale portion (or scale coefficients) and a detail portion (or a detail coefficient (Detailed coefficients). In addition, the feature extraction unit 300 extracts an intersection through comparison of the signal to be converted and a scale portion, and extracts a sub-step of the detail portion based on the extracted intersection. The specific operation of the feature extraction unit 300 will be described later with reference to FIG.

The corresponding path calculation unit 500 can calculate the corresponding path of two time series data. At this time, the corresponding path calculation unit 500 can calculate a corresponding path between the corresponding detail parts by applying DTW (Dynamic Time Warping) to the divided detail parts of each of the two time series data. Also, the corresponding path calculation unit 500 can express the calculated corresponding path in the time series data, and calculate the correct corresponding path between the time series data. The specific operation of the corresponding path calculation unit 500 will also be described later with reference to FIG.

The DTW algorithm refers to a pattern matching algorithm for two vector sequences of different lengths using a problem solving algorithm called dynamic programming.

Let the two dynamic pattern inputted _{Q (Q = q 1, q} 2, ..., q k, ..., q m) with _{C (C = c 1, c} 2, ..., c k, ..., c n) . The value (W _k ) of the corresponding path for obtaining the corresponding relation (d) of the k-th Q and C between them can be defined as shown in Equation (1).

The path value DTW (Q, C) that minimizes the corresponding path is expressed by Equation (2).

The corresponding path (warping path, r) is calculated as shown in Equation (3).

Each of the plurality of signals is stored in the DB 700. The plurality of signals may include a reference signal. In addition, data generated in the corresponding path calculation process may be stored in the DB 700. [

Each of the configurations of the corresponding path calculation apparatus 10 shown in FIG. 1 indicates that it is functionally and logically separable, and it means that each configuration is divided into separate physical devices or written in separate codes Or may be easily inferred by an average expert in the field of the present invention.

Also, in this specification, "part" may mean a functional and structural combination of hardware for carrying out the technical idea of the present invention and software for driving the hardware. For example, the above-mentioned "part" may mean a logical unit of a predetermined code and a hardware resource for executing the predetermined code, and it does not necessarily mean a physically connected code or a kind of hardware .

4 is a diagram for explaining a specific operation of the characteristic extracting unit and the corresponding path calculating unit shown in FIG.

1 to 4, the feature extraction unit 300 extracts two pieces of time series data (refer to FIG. 4A) stored in the DB 700 or received by the signal reception unit 100 using the MODWT The discrete wavelet transform can be performed (see Fig. 4 (b)). That is, the feature extraction unit 300 extracts scale factors and detail coefficients from each of the two time series data, and extracts a scale portion (which may mean a set of scale coefficients) and a detail portion Which may mean a set of detail coefficients).

In addition, the feature extraction unit 300 may extract the intersection by comparing the time series data with the corresponding scale part (scale factors) (see FIG. 4C). More specifically, when the feature extraction unit 300 arranges the time series data and the scale coefficients on the same axis (for example, time axis), it can extract points where the time series data and the scale coefficients cross each other at an intersection.

In addition, the feature extraction unit 300 may divide the detail parts (detail coefficients) of the time series data into a plurality of sections based on the extracted intersection points (see Fig. 4 (d)). According to the embodiment, the feature extraction unit 300 may divide the time series data into a plurality of sections based on the extracted intersection points.

The corresponding path calculation unit 500 can calculate the corresponding path of the two time series data by applying the DTW to the divided detail parts of each of the two time series data (see e in FIG. 4). In other words, the corresponding path calculation unit 500 calculates DTW (DTW) corresponding to each of the divided detail parts of the time series data of the two time series data and the divided detail parts of the other time series data of the corresponding two time series data The corresponding path can be calculated. However, if the feature extraction unit 300 divides the time series data itself instead of dividing the detail portion, the corresponding path calculation unit 500 can calculate the corresponding path by applying DTW to the divided time series data.

In addition, the corresponding path calculation unit 500 can calculate a correct corresponding path between the two time series data by applying (or expressing) the calculated corresponding path to the two time series data (see f in FIG. 4). Comparing the corresponding path between the finally computed time series (refer to FIG. 4F) with the corresponding path of FIG. 3, it can be seen that the singularity has drastically decreased.

5 is a flowchart illustrating a corresponding path calculation method performed by the corresponding path calculation apparatus shown in FIG. In the description of the corresponding path calculation method, a detailed description of the contents overlapping with those described above will be omitted.

Referring to FIGS. 1 to 5, the signal receiving unit 100 of the corresponding path calculation apparatus 10 can receive at least two signals to be analyzed (S100). According to an embodiment, the analysis target signal may be stored in advance in the DB 700 of the corresponding path calculation device 10. [

The property extraction unit 300 of the corresponding path calculation device 10 can extract the scale portion and the detail portion by discrete wavelet transform of the analysis object signal (S200). In addition, the feature extraction unit 300 may extract an intersection between the time-series data as the analysis target signal and the extracted scale part (S300), and divide the extracted detail part based on the extracted intersection point into a plurality of parts ( S400). The feature extraction unit 300 may divide the time series data as the analysis target signal in place of the detail portion.

The corresponding path calculation unit 500 of the corresponding path calculation apparatus 10 can calculate the corresponding path by applying the DTW to the divided detail part or the divided time series data (S500).

Lastly, the corresponding path calculation unit 500 can calculate the corresponding path between the time series data by applying the calculated corresponding path to the time series data (S600).

While the present invention has been particularly shown and described with reference to exemplary embodiments thereof, it is evident that many alternatives, modifications and variations will be apparent to those skilled in the art. Accordingly, the true scope of the present invention should be determined by the technical idea of the appended claims.

10: Corresponding path calculation device
100: Signal receiver
300: Feature extraction unit
500: corresponding path calculating unit
700: DB

Claims (9)

Extracting a first scale portion and a first detail portion from the first time series signal by performing a discrete wavelet transform on each of the first time series signal and the second time series signal, And dividing the first detail portion with reference to an intersection point of the first time series signal and the first scale portion and dividing the first detail portion with respect to an intersection of the second time series signal and the second scale portion A feature extraction unit for dividing the second detail portion; And
Calculating a first corresponding path by applying DTW (Dynamic Time Warping) to each of the divided first detail parts and the divided second detail parts, and using the first corresponding path to calculate the first time series signal and the second time series And a corresponding path calculation unit for calculating a corresponding path between the signals.
The first time series signal and the second time series signal are subjected to discrete wavelet transform to extract a first scale portion from the first time series signal and a second scale portion from the second time series signal, A characteristic extractor for dividing the first time series signal with reference to an intersection of the first scale portion and dividing the second time series signal based on an intersection of the second time series signal and the second scale portion; And
A first corresponding path is calculated by applying DTW to each divided first time series signal and a divided second time series part, and a corresponding path between the first time series signal and the second time series signal is calculated using the first corresponding path, And a corresponding path calculation unit for calculating a corresponding path.
3. The method according to claim 1 or 2,
Wherein the corresponding path calculation apparatus further comprises a signal reception section for receiving the first time series signal and the second time series signal,
Corresponding path calculation device.
3. The method according to claim 1 or 2,
The feature extraction unit may perform a discrete wavelet transformation using a MODWT (Maximal Overlap Discrete Wavelet Transform)
Corresponding path calculation device.
A corresponding path calculation method performed in a corresponding path calculation apparatus,
(a) discrete wavelet transforming each of the first time series signal and the second time series signal to extract a first scale portion and a first detail portion from the first time series signal, and extracting a second scale portion and a second scale portion from the second time series signal, 2 extracting a detail portion;
(b) extracting at least one first intersection point that is an intersection of the first time series signal and the first scale portion, and extracting at least one second intersection point that is an intersection of the second time series signal and the second scale portion step;
(c) dividing the first detail portion based on the at least one first intersection point and dividing the second detail portion based on the at least one second intersection point;
(d) applying a DTW to each of the divided first detail portions and the divided second detail portions to calculate a first corresponding path; And
(e) calculating a corresponding path between the first time series signal and the second time series signal using the first corresponding path.
A corresponding path calculation method performed in a corresponding path calculation apparatus,
(a) discrete wavelet-transforming each of the first time series signal and the second time series signal to extract a first scale portion from the first time series signal and extracting a second scale portion from the second time series signal;
(b) extracting at least one first intersection point that is an intersection of the first time series signal and the first scale portion, and extracting at least one second intersection point that is an intersection of the second time series signal and the second scale portion step;
(c) dividing the first time series signal with reference to the at least one first intersection point and dividing the second time series signal with reference to the at least one second intersection point;
(d) calculating a first corresponding path by applying DTW to each of the divided first time series signals and the divided second time series signals; And
(e) calculating a corresponding path between the first time series signal and the second time series signal using the first corresponding path.
The method according to claim 5 or 6,
Wherein the first time series signal and the second time series signal are semiconductor process signals having measured values according to time or speed,
Corresponding path calculation method.
The method according to claim 5 or 6,
In the step (a), the discrete wavelet transform is performed using the MODWT,
Corresponding path calculation method.
The method according to claim 5 or 6,
Wherein the corresponding path calculation method further comprises receiving the first time series signal and the second time series signal before the step (a)
Corresponding path calculation method.

Images