Method, system and medium for extracting time series distinguishing patterns based on parallel computing

Abstract

The present invention discloses a method, system, and medium for extracting distinguishable patterns from time series based on parallel computing. The method comprises determining a data set D and a set of candidate sequences S; performing a DTW distance calculation on a GPU to obtain a distance calculation result F and storing it in global memory; and if the number N of time series in data set D is greater than a set value, then the CPU is selected to calculate the optimal segmentation point based on the distance calculation result F in global memory and perform candidate sequence screening; otherwise, the GPU is selected to calculate the optimal segmentation point based on the distance calculation result F in global memory and perform candidate sequence screening. To address the time-consuming Shapelet discovery process, the present invention proposes Shapelet GPU parallel optimization based on the dynamic time warping distance (DTW) metric, reducing the time required for Shapelet parallel discovery and improving the efficiency of Shapelet parallel discovery.

Description

Technical Field

[0001] The present invention relates to the field of rail transit data mining, and in particular to a method, system and medium for extracting time series distinction patterns based on parallel computing. Background Art

[0002] The rail transit sector is flooded with a vast amount of data, most of which exists in the form of time series. More and more people are studying time series mining in order to obtain the intrinsic value of data. Time series classification is one of the important branches of time series mining. Currently, similarity / distance calculation still occupies a dominant position in the field of time series classification. The Shapelet classification algorithm is one of the more important time series classification algorithms based on similarity / distance metrics. The Shapelet algorithm also has its drawbacks. The training process is time-consuming and resource-intensive. Shapelets is essentially a local pattern that can be used to classify and cluster subsequences with distinguishing capabilities. It has two advantages: (1) local patterns; (2) strong interpretability. Therefore, Shapelets can not only classify time series, but also output patterns with distinguishing capabilities.

[0003] In order to solve the problem that Shapelet classification has high accuracy but long training time, many people have proposed many methods such as pruning and dimensionality reduction, but time consumption is still a key issue of Shapelet algorithm. In practical application, Shapelet algorithm requires both offline training module and online prediction module. Therefore, the overall solution of Shapelet algorithm consists of two parts: offline training part and online prediction part. Figure 1 As shown in the figure. The offline training part of the Shapelet algorithm is a process of training a model, which outputs a subsequence with the highest classification ability; the online prediction part of the Shapelet algorithm is mainly used to classify unknown time series and evaluate the accuracy of the prediction model. Among them, the offline training part of the Shapelet algorithm has the following functions: input the time series training set, generate a candidate sequence set based on the training set, and then execute the Shapelet parallel discovery process on the candidate sequence set, and output the candidate sequence with the highest classification ability and its corresponding optimal segmentation point as the prediction model. The process of predictive model training, that is, the Shapelet discovery process, is very time-consuming. Predictive model training is the focus of the overall parallel solution. How to accelerate the offline training part of the Shapelet algorithm has become a key technical problem that needs to be solved urgently. Summary of the Invention

[0004] The technical problem to be solved by the present invention is as follows: In response to the above-mentioned problems in the prior art, a method, system and medium for extracting time series distinguishing patterns based on parallel computing are provided. In response to the time-consuming problem of the Shapelet discovery process, the present invention proposes Shapelet GPU parallel optimization based on dynamic time warping (DTW) distance metric, which reduces the time required for Shapelet parallel discovery and improves the efficiency of Shapelet parallel discovery.

[0005] In order to solve the above technical problems, the technical solution adopted by the present invention is:

[0006] The time series distinguishing pattern extraction method based on parallel computing includes:

[0007] S101, determining a data set D and a candidate sequence set S, where the data set D consists of N time series and their corresponding categories, and the candidate sequence set S is obtained by enumerating sub-time series of the data set D that meet a specified length;

[0008] S102, performing DTW distance calculation on the GPU to obtain a distance calculation result F and storing it in the global memory;

[0009] S103: If the number N of time series in the data set D is greater than the set value, the CPU is selected to calculate the best segmentation point based on the distance calculation result F in the global memory and perform candidate sequence screening; otherwise, the GPU is selected to calculate the best segmentation point based on the distance calculation result F in the global memory and perform candidate sequence screening.

[0010] Optionally, the function expression of the data set D in step S101 is:

[0011] D={(T1,y1),(T2,y2),…,(T N ,y N )},

[0012] In the above formula, T1~T N Represent the 1st to Nth time series, y1~y N They represent the categories corresponding to the 1st to Nth time series respectively, and N is the number of time series in the dataset D.

[0013] Optionally, the function expression of the candidate sequence set S in step S101 is:

[0014]

[0015] In the above formula, T j s->s+len Represents the jth time series T in the dataset D jThe corresponding sub-time series contains the sth to s+lenth time series, where N is the number of time series in the dataset D, s represents the starting position of the sub-time series, len is the length of the sub-time series, and L is the length of a single time series in the dataset D.

[0016] Optionally, step S102 includes the step of performing DTW distance calculation with parameter w=0 by GPU:

[0017] S201, calculate any sub-time series based on the following formula using GPU The Euclidean distance between:

[0018] S202, using L GPU threads to perform DTW distance calculation with parameter w=0 according to the following formula:

[0019]

[0020] In the above formula, subDist(S,T j ) represents the candidate sequence set S and the time series T j The distance between Indicates that p starts from the value 1 and ends at L-|S|+1. The minimum value of Represents calculation Euclidean distance Euclid; Represents a sub-time series, where the subscript j represents the serial number of the corresponding time series in the data set D, the subscript p represents the starting position of the sub-time series, and the superscript || represents the length of the candidate sequence set S.

[0021] Optionally, in step S201, any sub-time series is calculated by GPU The function expression of the Euclidean distance Euclid between is:

[0022]

[0023] In the above formula, and They represent sub-time series respectively, where i and j in the subscripts represent the serial numbers of the corresponding time series in the dataset D, s, p, s+1 and p+1 in the subscripts represent the starting position of the sub-time series, and len represents the length of the sub-time series; T i,s 、T j,p 、T i,s+len and T j,p+lenThey represent a value in the time series respectively, where the i and j in the subscripts represent the serial numbers of the corresponding time series in the dataset D, the s, p, s+len and p+len in the subscripts represent the positions of the time series, Euclid represents the calculation of the Euclidean distance between two time series and sub-time series, and d represents the calculation of the Euclidean distance between two values.

[0024] Optionally, step S102 includes the step of performing DTW distance calculation with parameter w>0 by GPU:

[0025] S301, calculating the DTW distance D between any two time series a and b using the GPU;

[0026] S302, using L GPU threads to perform DTW distance calculation with parameter w>0 according to the following formula:

[0027]

[0028] In the above formula, subDist(S, T j ) represents the candidate sequence set S and the time series T j The distance between Indicates that p starts from the value 1 and ends at L-|S|+1. The minimum value of Indicates the value calculated in step S301 The Euclidean distance D of represents a sub-time series, where the subscript j represents the sequence number of the corresponding time series in the dataset D, the subscript p represents the starting position of the sub-time series, the superscript |S| represents the length of the candidate sequence set S, and w is the parameter for DTW distance calculation.

[0029] Optionally, in step S301, the function expression for calculating the DTW distance D between any two time series a and b by the GPU is:

[0030]

[0031] In the above formula, D(a 1：x , b 1：y , w) represents a 1：x and b 1：y The DTW distance D between even t Indicates the tth element on the line x+y=M when x+y is an even number, envn t+1 When x+y is an even number, the t+1th element on the x+y=M line, d(a x , b y ) means to calculate two elements (a x , b y ), min(event , odd t , odd t+1 ) means in even t , odd t , odd t+1 The minimum value of the three, if x+y is even means the value of x+y is an even number, a x 、b y are the subsequences of x and y elements before two time series a and b, respectively. x and y are the sequence numbers of the two time series. odd t Indicates the tth element on the line x+y=M when x+y is an odd number, odd t+1 If x+y is odd, it indicates the t+1th element on the line x+y=M. If x+y is odd, it indicates that the value of x+y is an odd number. w is the parameter for DTW distance calculation.

[0032] Optionally, the set value in step S103 is 2000.

[0033] In addition, the present invention also provides a time series distinguishing pattern extraction system based on parallel computing, including a microprocessor and a memory connected to each other, and the microprocessor is programmed or configured to execute the steps of the time series distinguishing pattern extraction method based on parallel computing.

[0034] In addition, the present invention also provides a computer-readable storage medium, in which a computer program is stored. The computer program is used to be programmed or configured by a microprocessor to execute the steps of the time series distinguishing pattern extraction method based on parallel computing.

[0035] Compared with the prior art, the present invention has the following main advantages: the method of the present invention includes determining a data set D and a candidate sequence set S; performing DTW distance calculation on a GPU to obtain a distance calculation result F and storing it in a global memory; if the number N of time series in the data set D is greater than a set value, the CPU is selected to calculate the best segmentation point based on the distance calculation result F in the global memory and perform candidate sequence screening, otherwise the GPU is selected to calculate the best segmentation point based on the distance calculation result F in the global memory and perform candidate sequence screening. In response to the time-consuming problem of the Shapelet discovery process, the present invention proposes a Shapelet GPU parallel optimization based on the dynamic time warping distance DTW metric, and designs a Shapelet discovery parallel implementation method jointly executed by the CPU and GPU, in which the CPU is only responsible for scheduling (responsible for a small amount of data calculation), thereby reducing the time required for Shapelet parallel discovery and improving the efficiency of Shapelet parallel discovery. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] Figure 1This is a flowchart of the overall scheme of the Shapelet algorithm in the prior art.

[0037] Figure 2 Schematic diagram of the basic process of the method of the embodiment of the present invention.

[0038] Figure 3 Schematic diagram of the basic principle of the method of the embodiment of the present invention.

[0039] Figure 4 This is a dependency diagram during DTW distance calculation with parameter w=0 in an embodiment of the present invention.

[0040] Figure 5 This is a dependency diagram for DTW distance calculation when parameter w>0 in an embodiment of the present invention. DETAILED DESCRIPTION

[0041] The following text uses data collection from rail vehicle inverters in rail transit as an example to further explain the present invention's method, system, and medium for extracting time series patterns based on parallel computing. Domain experts label the input current data from the vehicle inverters (marking whether it's faulty or not). This labeled data is used to create training data, ensuring a balanced mix of positive and negative samples and a fixed sample length. This method, system, and medium for extracting time series patterns based on parallel computing can then be used to perform parallel Shapelet discovery.

[0042] like Figure 2 and Figure 3 As shown, the time series distinguishing pattern extraction method based on parallel computing in this embodiment includes:

[0043] S101, determining a data set D and a candidate sequence set S, where the data set D consists of N time series and their corresponding categories, and the candidate sequence set S is obtained by enumerating sub-time series of the data set D that meet a specified length;

[0044] S102, performing DTW distance calculation on the GPU to obtain a distance calculation result F and storing it in the global memory;

[0045] S103: If the number N of time series in the data set D is greater than the set value, the CPU is selected to calculate the best segmentation point based on the distance calculation result F in the global memory and perform candidate sequence screening; otherwise, the GPU is selected to calculate the best segmentation point based on the distance calculation result F in the global memory and perform candidate sequence screening.

[0046] In this embodiment, the function expression of the data set D in step S101 is:

[0047] D={(T1,y1),(T2,y2),…,(T N ,yN )},

[0048] In the above formula, T1~T N Represent the 1st to Nth time series, y1~y N They represent the categories corresponding to the 1st to Nth time series respectively, and N is the number of time series in the data set D. In this embodiment, the time series is a sequence composed of the current data of the railway vehicle converter. Each value represents the current data collected at a moment, and the category is used to represent the fault category of the current data. Assuming that any time series T j The corresponding category is y j , then we have the time series T j =t1,…,t L is an ordered set of L real numbers, y j ∈{A, B}, in this embodiment, A / B represent normal and abnormal respectively.

[0049] In this embodiment, the function expression of the candidate sequence set S in step S101 is:

[0050]

[0051] In the above formula, Represents the jth time series T in the dataset D j The corresponding sub-time series contains time series from s to s+len, where N is the number of time series in dataset D, s represents the starting position of the sub-time series, len is the length of the sub-time series, and L is the length of a single time series in dataset D. The candidate sequence set S is obtained by enumerating sub-sequences in dataset D that meet a certain length. The length of the sub-time series, len, also varies within a certain range.

[0052] In this embodiment, step S102 includes the step of performing DTW distance calculation with parameter w=0 by the GPU:

[0053] S201, calculate any sub-time series based on the following formula using GPU The Euclidean distance between:

[0054] S202, using L GPU threads to perform DTW distance calculation with parameter w=0 according to the following formula:

[0055]

[0056] In the above formula, subDist(S, T j ) represents the candidate sequence set S and the time series T j The distance between Indicates that p starts from the value 1 and ends at L-|S|+1. The minimum value of Represents calculation Euclidean distance Euclid; Represents a sub-time series, where the subscript j represents the serial number of the corresponding time series in the data set D, the subscript p represents the starting position of the sub-time series, and the superscript |S| represents the length of the candidate sequence set S.

[0057] In this embodiment, in step S201, the GPU is used to calculate any sub-time series. The function expression of the Euclidean distance Euclid between is:

[0058]

[0059] In the above formula, and They represent sub-time series respectively, where i and j in the subscripts represent the serial numbers of the corresponding time series in the dataset D, s, p, s+1 and p+1 in the subscripts represent the starting position of the sub-time series, and len represents the length of the sub-time series; T i，s 、T j，p 、T i，s+len and T j，p+len Respectively represent a value in the time series, where the subscripts i and j represent the serial numbers of the corresponding time series in the dataset D, the subscripts s, p, s+len, and p+len represent the position of the time series, len represents the length of the sub-time series, Euclid represents the calculation of the Euclidean distance between two time series and sub-time series, and d represents the calculation of the Euclidean distance between two values. For w = 0, the dependence of the distance calculation is as follows Figure 4 As shown, see Figure 4 It can be seen that in this embodiment, the GPU is used to calculate any sub-time series based on the following formula: When the Euclidean distance between the two is Euclid, it is first calculated separately by multi-threading. and Then all You can Completed in constant time through cooperation between threads.

[0060] In this embodiment, step S102 includes the step of performing DTW distance calculation with parameter w>0 by the GPU:

[0061] S301, calculate the DTW distance D between any two time series a and b through GPU;

[0062] S302, using L GPU threads to perform DTW distance calculation with parameter w>0 according to the following formula:

[0063]

[0064] In the above formula, subDist(S, T j ) represents the candidate sequence set S and the time series T j The distance between Indicates that p starts from the value 1 and ends at L-|S|+1. The minimum value of Indicates the value calculated in step S301 The Euclidean distance D of represents a sub-time series, where the subscript j represents the sequence number of the corresponding time series in the dataset D, the subscript p represents the starting position of the sub-time series, the superscript |S| represents the length of the candidate sequence set S, and w is the parameter for DTW distance calculation.

[0065] In this embodiment, the function expression for calculating the DTW distance D between any two time series a and b by the GPU in step S301 is:

[0066]

[0067] In the above formula, D(a 1：x , b 1：y , w) represents a 1：x and b 1：y The DTW distance D between even t Indicates the tth element on the line x+y=M when x+y is an even number, envn t+1 When x+y is an even number, the t+1th element on the x+y=M line, d(a x , b y ) means to calculate two elements (a x , b y ), min(even t , odd t , odd t+1 ) means in even t , odd t , odd t+1 The minimum value of the three, if x+y is even means the value of x+y is an even number, a x 、b y are the subsequences of x and y elements before two time series a and b, respectively. x and y are the sequence numbers of the two time series. odd t Indicates the tth element on the line x+y=M when x+y is an odd number, odd t+1If x+y is odd, it means the t+1th element on the line x+y=M. If x+y is odd, it means the value of x+y is odd. w is the parameter for DTW distance calculation. For w>0, the distance calculation dependency is as follows: Figure 5 As shown, DTW(A 1：m+1 , B 1：m+1 ,w) represents D(A 1：m+1 , B 1：m+1 ,w), see Figure 5 It can be seen that DTW(A 1：m+1 , B 1：m+1 ,w) calculation depends on DTW(A 1：m , B 1：m ,w),DTW(A 1：m , B 1：m+1 ,w),DTW(A 1：m+1 , B 1：m , w) is calculated to avoid DTW(A 1：m+1 , B 1：m+1 , w) starts from 0. Figure 5 The parallel dependency relationships marked with symbols in the lower left corner of the figure are shown in the "Parallel Dependency Computation" box on the right. "even" represents the tth element on the line x+y=M when x+y is even, and "odd" represents the tth element on the line x+y=M when x+y is odd. The five example threads, threads 1 through 5, perform parallel computations in an alternating pattern of odd and even. For example, the figure shows a parallel round of "odd" followed by a parallel round of "even," and so on. In actual computations, matrix values ​​are not updated in the x and y directions. Instead, elements with the same x+y values ​​are updated in batches, iterating in the direction of increasing x+y values.

[0068] like Figure 2 As shown, the set value in step S103 of this embodiment is 2000. If the number N of time series in the data set D is greater than 2000, the CPU is selected to calculate the best segmentation point based on the distance calculation result F in the global memory and perform candidate sequence screening. Otherwise, the GPU is selected to calculate the best segmentation point based on the distance calculation result F in the global memory and perform candidate sequence screening. It should be noted that the calculation of the best segmentation point based on the distance calculation result F in the global memory and the screening of candidate sequences include: first, calculating the information gain according to the following formula:

[0069]

[0070] In the above formula, g(D,(S,d th )) represents information gain, D is the data set, S is the candidate sequence set, d th is the distance threshold, (S, d th) as a feature to classify the data set into two categories, D1 and D2; H(D|(S,d th )) is the value in (S, d th ) under the empirical conditional entropy, H(D) is the information entropy of data set D, H(D1) is the information entropy of data set D1, H(D2) is the information entropy of data set D2, |D| is the size of data set D, |D1| is the size of data set D1, |D2| is the size of data set D2; then the maximum information gain corresponds to the distance threshold d th As the corresponding optimal segmentation point, the sequence with the largest information gain can be selected.

[0071] In summary, the time series distinguishing pattern extraction method based on parallel computing in this embodiment includes determining a data set D and a candidate sequence set S; performing DTW distance calculation on a GPU to obtain a distance calculation result F and storing it in a global memory; if the number N of time series in the data set D is greater than a set value, then the CPU is selected to calculate the best segmentation point based on the distance calculation result F in the global memory and perform candidate sequence screening, otherwise the GPU is selected to calculate the best segmentation point based on the distance calculation result F in the global memory and perform candidate sequence screening. In response to the time-consuming problem of the Shapelet discovery process, the present invention proposes a Shapelet GPU parallel optimization based on the dynamic time warping distance DTW metric, and designs a Shapelet discovery parallel implementation method jointly executed by the CPU and GPU, wherein the CPU is only responsible for scheduling (responsible for a small amount of data calculation), thereby reducing the time required for Shapelet parallel discovery and improving the efficiency of Shapelet parallel discovery.

[0072] In addition, this embodiment also provides a time series distinguishing pattern extraction system based on parallel computing, including a microprocessor and a memory connected to each other, and the microprocessor is programmed or configured to execute the steps of the aforementioned time series distinguishing pattern extraction method based on parallel computing.

[0073] In addition, this embodiment also provides a computer-readable storage medium, in which a computer program is stored. The computer program is used to be programmed or configured by a microprocessor to execute the steps of the aforementioned time series pattern extraction method based on parallel computing.

[0074] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present application may take the form of a computer program product implemented on one or more computer-readable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code. The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, may be implemented by computer program instructions. These computer program instructions may be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the steps in the process. Figure 1 a process or multiple processes and / or boxes Figure 1 These computer program instructions can also be stored in a computer-readable memory that can guide a computer or other programmable data processing device to work in a specific way, so that the instructions stored in the computer-readable memory produce a product including the instruction device, which implements the function specified in the process. Figure 1 a process or multiple processes and / or boxes Figure 1 These computer program instructions can also be loaded onto a computer or other programmable data processing device, so that a series of operation steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process in the process. Figure 1 a process or multiple processes and / or boxes Figure 1 The steps for the function specified in one or more boxes.

[0075] The above description is merely a preferred embodiment of the present invention. The scope of protection of the present invention is not limited to the above embodiment. All technical solutions based on the concept of the present invention are within the scope of protection of the present invention. It should be noted that for those skilled in the art, various improvements and modifications that do not depart from the principles of the present invention should also be considered within the scope of protection of the present invention.

Claims

1. A method for extracting time series distinguishing patterns based on parallel computing, characterized in that: include: S101, determine the data set D and candidate sequence set S, where the dataset D Depend on N time series and their corresponding categories, the candidate sequence set S is the dataset D The sub-time series that meet the specified length are enumerated; S102, performing DTW distance calculation on the GPU to obtain a distance calculation result F and store it in global memory; S103, if the data set D The number of time series in N If the value is greater than the set value, the CPU will calculate the distance based on the global memory. F Calculate the best segmentation point and perform candidate sequence screening, otherwise the GPU calculates the result based on the distance in global memory F Calculate the optimal split point and screen candidate sequences; Step S102 includes the step of performing DTW distance calculation with parameter w=0 by the GPU: S201, calculate any sub-time series based on the following formula using GPU Euclidean distance between : S202, passed L Each GPU thread performs the DTW distance calculation with parameter w=0 according to the following formula: ， In the above formula, Represents the candidate sequence set S With time series T j The distance between express p Starting from value 1 to End The minimum value of Represents calculation Euclidean distance ; Represents a sub-time series, where the subscript j Indicates in the dataset D The corresponding time series number in the subscript p Indicates the starting position of the sub-time series. Represents the candidate sequence set S length.

2. The method for extracting time series distinguishing patterns based on parallel computing according to claim 1, characterized in that: Data set in step S101 D The function expression is: ， In the above formula, T 1~ T N Respectively represent the 1st to N A time series, y 1~ y N Respectively represent the 1st to N The categories corresponding to the time series, N For the dataset D The number of time series in .

3. The method for extracting time series distinguishing patterns based on parallel computing according to claim 1, characterized in that: The candidate sequence set in step S101 S The function expression is: ， In the above formula, T j s->s+len Representation dataset D Middle j time series T j The corresponding sub-time series contains s ~ s+len A time series, N For the dataset D The number of time series in , s Indicates the starting position of the sub-time series, len is the length of the sub-time series, L For the dataset D The length of a single time series in .

4. The method for extracting time series distinguishing patterns based on parallel computing according to claim 3, characterized in that: In step S201, the GPU is used to calculate any sub-time series Euclidean distance between The function expression is: In the above formula, 、 、 and Represent sub-time series, where i 、 j Indicates in the dataset D The corresponding time series number in the subscript s 、 p 、 s +1 and p +1 indicates the starting position of the sub-time series, len Indicates the length of the sub-time series; 、 、 and Respectively represent a value in the time series, where the subscript i 、 j Indicates in the dataset D The corresponding time series number in the subscript s 、 p 、 s + len as well as p + len represents the position of the time series, Indicates calculating the Euclidean distance between two time series and sub-time series. Calculates the Euclidean distance between two values.

5. The method for extracting time series distinguishing patterns based on parallel computing according to claim 3 is characterized in that: Step S102 includes the step of performing DTW distance calculation with parameter w>0 by GPU: S301, calculate the DTW distance between any two time series a and b through GPU ; S302, passed L Each GPU thread performs the DTW distance calculation with parameter w>0 according to the following formula: ， In the above formula, Represents the candidate sequence set S With time series T j The distance between express p Starting from value 1 to End The minimum value of Indicates the value calculated in step S301 Euclidean distance ; Represents a sub-time series, where the subscript j Indicates in the dataset D The corresponding time series number in the subscript p Indicates the starting position of the sub-time series. Represents the candidate sequence set S length, Parameters for DTW distance calculation.

6. The method for extracting time series distinguishing patterns based on parallel computing according to claim 5, characterized in that: In step S301, the DTW distance between any two time series a and b is calculated by GPU. The function expression is: ， In the above formula, Indicates a 1:x and b 1:y The DTW distance between , It means the tth element on the line x+y=M when x+y is an even number. It means the t+1th element on the line x+y=M when x+y is an even number. Indicates calculation of two elements The distance between Indicates The minimum of the three, express The value of is an even number, are the subsequences of the first x and y elements of the two time series a and b respectively. are the serial numbers of the two time series, It means the tth element on the line x+y=M when x+y is an odd number. It means the t+1th element on the line x+y=M when x+y is an odd number. express The value of is an odd number, Parameters for DTW distance calculation.

7. The method for extracting time series distinguishing patterns based on parallel computing according to claim 1, characterized in that: The set value in step S103 is 2000.

8. A time series distinguishing pattern extraction system based on parallel computing, comprising a microprocessor and a memory connected to each other, characterized in that: The microprocessor is programmed or configured to execute the steps of the time series distinguishing pattern extraction method based on parallel computing according to any one of claims 1 to 7.

9. A computer-readable storage medium storing a computer program, wherein: The computer program is used to be programmed or configured by a microprocessor to execute the steps of the time series distinguishing pattern extraction method based on parallel computing according to any one of claims 1 to 7.

Images