Multivariate time series processing method, device, equipment, medium, and program

Abstract

The present application discloses a method for processing a multivariate time series, for mining causal relationships between each variable in the multivariate time series, by obtaining a time series set, the time series set including a plurality of first time series segments, where each first time series segment has the same length and is a multivariate time series. For any one of the first time series segments, input the first time series segment into a graph neural network to predict a multivariate parameter value corresponding to a first time, where the first time is the next time adjacent to the latest time in the first time series segment. Based on the multivariate parameter values ​​corresponding to the plurality of first times and the multivariate series labels (true values) corresponding to the plurality of first times, an optimization function is determined, the optimization function including a loss function and a causal matrix of the graph neural network. A value of each parameter in the causal matrix is ​​determined with the goal of minimizing the optimization function, and the causal relationships between the multivariates are determined based on the value of each parameter.

Description

[Technical field]

[0001] [CROSS-REFERENCE TO RELATED APPLICATIONS] This application claims priority to a Chinese patent application bearing application number 202210529330.4, filed on May 16, 2022, and entitled "Method, Apparatus, Equipment and Medium for Processing Multivariate Time Series," the entire contents of which are incorporated herein by reference.

[0002] The present application relates to the field of computer technology, and in particular to a method, apparatus, device and medium for processing multivariate time series. [Background technology]

[0003] A time series is a sequence formed by arranging the values ​​of a certain statistical indicator in chronological order. Generally, time series can be divided into single-variable time series and multivariable time series. Here, a multivariable time series refers to a value in which multiple different statistical indicators exist at the same time. Research on multivariable time series mainly includes forecasting, anomaly detection, causal mining, etc., where causal mining refers to mining the influence relationship between multiple variables.

[0004] Currently, analysis of multivariate time series is primarily focused on the study of multiple variables with simple linear relationships and cannot effectively analyze multivariate time series with nonlinear dependencies. Summary of the Invention [Problem to be solved by the invention]

[0005] In view of this, the present application provides a multivariate time series processing method, apparatus, device, and medium for processing multivariate time series having nonlinear relationships. [Means for solving the problem]

[0006] In order to achieve the above objectives, the technical solution of this application is as follows:

[0007] A first aspect of the present application provides a method for processing multivariate time series, said method comprising the steps of: obtaining a time series set, the time series set including a plurality of first time series segments, each first time series segment in the plurality of first time series segments having the same length, the first time series segments being a multivariate time series; For any one first time series segment in the set of time series, input the first time series segment into a graph neural network to predict a multivariate parameter value corresponding to a first time, the first time being the next time adjacent to a latest time in the first time series segment; determining an optimization function including a loss function and a causal matrix of the graph neural network based on multivariate parameter values ​​corresponding to the plurality of first times and multivariate sequence labels corresponding to the plurality of first times; determining values ​​for each parameter in the causal matrix with the goal of minimizing the optimization function; and determining causal relationships among multiple variables in the multivariate time series based on values ​​of each parameter in the causal matrix.

[0008] A second aspect of the present application provides an apparatus for processing multivariate time series, said apparatus comprising: an acquisition unit for acquiring a time series set, the time series set including a plurality of first time series segments, each of the first time series segments in the plurality of first time series segments having the same length, the first time series segments being a multivariate time series; a prediction unit for inputting a first time series segment in the set of time series into a graph neural network and predicting a multivariate parameter value corresponding to a first time, the first time being a next time to a latest time in the first time series segment; a determination unit for determining an optimization function including a loss function and a causal matrix of the graph neural network based on multivariate parameter values ​​corresponding to the plurality of first time instants and multivariate sequence labels corresponding to the plurality of first time instants; The determination unit is further adapted to determine values ​​of each parameter in the causal matrix with the goal of minimizing the optimization function; The determining unit is further used for determining a causal relationship between multiple variables in the multivariate time series based on values ​​of each parameter in the causal matrix.

[0009] A third aspect of the present application provides an electronic device, the device comprising a processor and a memory, the memory adapted to store instructions or computer programs, and the processor adapted to execute the instructions or computer programs in the memory to cause the electronic device to perform a method according to the first aspect.

[0010] A fourth aspect of the present application provides a computer readable storage medium having instructions stored thereon that, when run on an apparatus, cause the apparatus to perform a method according to the first aspect.

[0011] A fifth aspect of the present application provides a computer program product, the computer program product comprising computer programs / instructions which, when executed by a processor, cause the execution of the method according to the first aspect. Effect of the Invention

[0012] As can be seen thereby, the embodiments of the present application have the following beneficial effects:

[0013] In the present application, in order to mine causal relationships between each variable in a multivariate time series, a time series set is obtained, and the time series set includes a plurality of first time series segments, where each first time series segment has the same length and is a multivariate time series. For any one of the first time series segments in the time series set, the first time series segment is input into a graph neural network to predict a multivariate parameter value corresponding to a first time. Here, the first time is the next time adjacent to the latest time in the first time series segment. That is, the parameter value of each variable in the first time series segment is used to predict the parameter value of each variable at the next time. An optimization function is determined based on the multivariate parameter values ​​corresponding to the plurality of first times and the multivariate series labels (true values) corresponding to the plurality of first times, and the optimization function includes a loss function and a causal matrix of the graph neural network. The value of each parameter in the causal matrix is ​​determined with the goal of minimizing the optimization function, and the causal relationship between the multivariables is determined based on the value of each parameter. [Brief description of the drawings]

[0014] In order to more clearly explain the technical solutions in the embodiments of the present application or the prior art, the following briefly introduces drawings that need to be used in the embodiments or prior art description. It is obvious that the drawings in the following description are only some embodiments described in the present application, and those skilled in the art can also obtain other drawings based on these drawings without exerting creative efforts. [Figure 1] FIG. 2 is a timeline schematic diagram according to an embodiment of the present application. [Diagram 2] 1 is a flow chart of a method for processing multivariate time series according to an embodiment of the present application; [Figure 3a] FIG. 2 is a schematic diagram of obtaining a first time series segment according to an embodiment of the present application; [Figure 3b] FIG. 13 is a schematic diagram of obtaining another first time series segment according to an embodiment of the present application; [Figure 4] FIG. 2 is a structural schematic diagram of a graph neural network according to an embodiment of the present application; [Diagram 5] 1 is a flowchart of another method for processing multivariate time series according to an embodiment of the present application; [Figure 6] FIG. 13 is a schematic diagram of obtaining a second time series segment according to an embodiment of the present application; [Figure 7] 1 is a schematic diagram of a device for processing multivariate time series according to an embodiment of the present application; [Figure 8] 1 is a structural schematic diagram of an electronic device according to an embodiment of the present application; DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0015] In order to allow those skilled in the art to better understand the solution of the present application, the following clearly and completely describes the technical solutions in the embodiments of the present application in conjunction with the drawings in the embodiments of the present application, obviously, the described embodiments are only some of the embodiments of the present application, not all of the embodiments, and all other embodiments obtained by those skilled in the art on the premise of not exerting creative efforts based on the embodiments in the present application are all within the protection scope of the present application.

[0016] To solve the causal inference problem between each variable in a time series, an embodiment of the present application provides a causal structure mining method based on graph neural network, and analyzes the time lag effect between variables by mining the causal relationship between each variable in a time series.

[0017] For ease of understanding, refer to the application scenario diagram shown in Figure 1. As shown in Figure 1, in this application scenario, a time series includes four variables, x1, x2, x3, and x4, respectively. Figure (a) is a normal time series, (b) is a discretized time series, and the directed graph shown in (c) is obtained by mining. The directed graph (c) shows the influence (outside 1) The factors affecting TIFF2025506528000002.tif11127 can be determined.

[0018] That is, the technical solution of the present application can not only mine the existence of multivariate causal relationships, but also determine the influence degree and maximum influence hysteresis number, such as the value of each edge in Figure (c). (outside 2) Variables that affect TIFF2025506528000003.tif9127 include (Outside 3) If you have TIFF2025506528000004.tif11127, then the maximum effect hysteresis number is 2. (outside 4) Variables that affect TIFF2025506528000005.tif9127 include (outside 5) If you have TIFF2025506528000006.tif11127, then the maximum effect hysteresis number is 2.

[0019] It should be explained that the present application does not currently consider synchronizing interactions between variables. For example, (outside 6) The influence between TIFF2025506528000007.tif11127 is not taken into account.

[0020] Based on the above scenario, the following describes how to process multivariate time series according to an embodiment of the present application with associated figures.

[0021] Referring to FIG. 2, this figure is a flowchart of a method for processing multivariate time series according to an embodiment of the present application, the method may be executed by a processing device, the processing device may be a server, an electronic device, or other devices, not limited here. Here, the server may be a device with storage and calculation capabilities, such as a cloud server or a server cluster. The electronic device may include devices with communication capabilities, such as mobile phones, tablet computers, desktop computers, notebook computers, car terminals, wearable electronic devices, all-in-one, smart home devices, etc., and may be devices simulated by a virtual machine or simulator. As shown in FIG. 2, the method may include the following steps:

[0022] S201: A time series set is obtained, and the time series set includes a plurality of first time series segments.

[0023] In this embodiment, after obtaining a time series, the time series progression may be divided according to a predetermined time length to obtain a plurality of first time series segments. Here, the length of each first time series segment in the plurality of first time series segments is the same, and the first time series segment is a multivariate time series. For example, as shown in FIG. 3a, a time series from tn to t time is obtained, the time series includes variables x1, x2, x3, ..., and the time series is divided into time series segment 1, time series segment 2, time series segment 3, and time series segment 4 with equal lengths. Here, the predetermined time length may be set according to the actual application situation, and this embodiment is not limited here.

[0024] Optionally, an observation window is preset, and the observation window is slid to obtain a plurality of first time series segments. Specifically, an observation window and a sliding step size are obtained, and the observation window is slid in a time series manner by the sliding step size to obtain a plurality of first time series segments. Here, the length of the observation window and the sliding step size may be determined according to the actual application situation. For example, the observation window So=3, and the sliding step size is 1. For example, as shown in FIG. 3b, the solid line observation window is a division of one first time series segment, and is moved to the right by a predetermined sliding step size to obtain a dotted line observation window, which is another first time series segment.

[0025] It should be noted that the length of the observation window is equal to or greater than the maximum time lag value estimated. For example, if the maximum time lag value estimated is 3, i.e., if a variable at time t-3 affects a variable at time t, then the length of the observation window is at least 3.

[0026] S202: For any one first time series segment in the set of time series, input the first time series segment into a graph neural network to predict a multivariate parameter value corresponding to a first time.

[0027] After obtaining the time series set, for each first time series segment in the time series set, the first time series segment is input to a graph neural network to predict a multivariable parameter value corresponding to a first time. Here, the first time is the next time adjacent to the latest time in the first time series segment. For example, if the time corresponding to the first time series segment is from tn to tn-5, the first time is tn-6. That is, the parameter value of each variable at the next time is predicted using the time series segment within the time period from tn to tn-5. That is, the graph neural network can predict the parameter value of each variable at the next time of each first time series segment.

[0028] Specifically, in order to mine the causal relationships between each variable in a time series, this embodiment constructs a structural causal model, and the representation format of this model is as follows:

[0029]

number

[0030] Where X t-τ represents the first time series segment, and A τ represents a causal matrix that includes the influencing factors between each variable. For example, the causal matrix includes a 12 2 and a 32 1 Zt represents the noise corresponding to each variable at time t, and τ=1, ..., p, where p represents the maximum time lag value. For example, in Figure 1, p=3.

[0031] Specifically, the above formula (1) may be expressed as follows by integration:

[0032]

number

[0033] Specifically, the general expression format is as follows:

[0034]

number

[0035] Alternatively, the multivariate parameter value corresponding to the first time may be predicted by the following method: input the first time series segment and the multivariate sequence label at the first time into an encoder of a graph neural network to obtain noise features corresponding to each variable and predicted features of each variable at the first time; input the predicted features of each variable at the first time and the noise features corresponding to each variable into a de-encoder of a graph neural network to obtain the multivariate parameter value corresponding to the first time, where the multivariate sequence label at the first time is the real sequence value corresponding to each variable at the first time.

[0036] Here, the structure of the graph neural network includes a node embedding module (NE), an edge neural network (ENN), and a node neural network (NNN), as shown in FIG. 4. Here, the NE uses a multilayer perceptron (MLP) to map node information to a feature space, and the ENN encodes causal relationships and time lag information represented by edges into edge features and stitches them with node features. (outer 9) The dot product with TIFF2025506528000013.tif9127 is taken, the edge features are updated, and the NNN passes the edge features to the next node via an aggregation function.

[0037] Here, the expression format corresponding to NE is as follows:

[0038]

number

[0039] The representation format corresponding to ENN is as follows:

[0040]

number

[0041] The representation format corresponding to NNN is as follows:

number

[0042] where x represents the original node feature, h represents the embedding of a node or edge, and f emb , f τ represents the MLP, fe and fv may be MLP layers or identity maps, etc., and l is the number of iterations. Here, the calculation result of fv is expressed as (Outside 10) Equivalent to TIFF2025506528000017.tif9127.

[0043] Here, the result of the encoder output is Zt=F1(F2(X t )-G(X)).

[0044] The output corresponding to the decoder is X t ’ =F3(G(X)+F4(Zt)).

[0045] As can be seen from FIG. 4, the input to the model is X+Xt, where X represents the first time series segment and Xt represents the real series value corresponding to the first time for determining the noise Zt. The output from the model is X t ’ and represents the predicted variable parameter value at the first time.

[0046] S203: An optimization function is determined based on multivariate parameter values ​​corresponding to the plurality of first time instants and multivariate sequence labels corresponding to the plurality of first time instants.

[0047] After predicting a multivariate parameter value corresponding to a corresponding first time based on each first time series, determine an optimization function based on the multivariate parameter values ​​corresponding to the plurality of first times and the multivariate sequence labels corresponding to the plurality of first times, the optimization function including a loss function of the graph neural network and a causal matrix, where each parameter in the causal matrix is ​​used to reflect whether there is an influence between the variables.

[0048] Specifically, the optimization function is expressed as follows:

[0049]

number

[0050] Where: (Outside 11) TIFF2025506528000019.tif9127 indicates the loss function of the graph neural network, n represents the number of samples of the input model, i.e., the number of first time series segments in the time series set, (Outside 12) TIFF2025506528000020.tif7127 is the sparsity coefficient.

[0051] Here, the loss function of the graph neural network may be selected according to the actual application, such as mean squared error, cross-entropy cost function, etc.

[0052] S204: Determine the value of each parameter in the causal matrix with the goal of minimizing the optimization function.

[0053] After determining the optimization function, the value of each parameter in the causal mean value is determined by solving the optimization function with the goal of minimizing the optimization function.

[0054] Specifically, for example, it is as shown in formula (8).

[0055]

number

[0056] S205: Determine causal relationships between multiple variables in the multivariate time series based on the values ​​of each parameter in the causal matrix.

[0057] After determining the value of each parameter in the causal matrix, determine the causal relationship between each variable in the multivariate time series based on the value of each parameter. Specifically, a parameter value of zero indicates that the variable corresponding to this parameter does not affect the variable solved at the first time. If a parameter value is not zero, determine that a time-lag causal relationship exists between the multivariables.

[0058] Optionally, when determining the causal relationship between each variable based on the parameter value of the causal matrix, first determine whether the value of the parameter in the causal matrix is ​​smaller than a predetermined threshold, and if so, set the value of the parameter to 0. Specifically, determine a target parameter whose value in the causal matrix is ​​smaller than a predetermined threshold, and set the value of the target parameter to 0.

[0059] As can be seen, in order to mine the causal relationship between each variable in a multivariate time series, a time series set is obtained, and the time series set includes a plurality of first time series segments, where each first time series segment has the same length and is a multivariate time series. For any one of the first time series segments in the time series set, the first time series segment is input into a graph neural network to predict a multivariate parameter value corresponding to a first time. Here, the first time is the next time adjacent to the latest time in the first time series segment. That is, the parameter value of each variable in the first time series segment is used to predict the parameter value of each variable at the next time. An optimization function is determined based on the multivariate parameter values ​​corresponding to the plurality of first times and the multivariate series labels (true values) corresponding to the plurality of first times, and the optimization function includes a loss function and a causal matrix of the graph neural network. The value of each parameter in the causal matrix is ​​determined with the goal of minimizing the optimization function, and the causal relationship between the multivariables is determined based on the value of each parameter.

[0060] As can be seen from the above embodiment, before determining the optimization function, one prediction is made based on the first time series segment, and multiple predictions may be made to further mine the causal relationship between each variable in the multivariate time series, and the optimization function may be determined based on the multiple prediction results. Specifically, the following takes two predictions, i.e., the length of the prediction window is 2, as an example.

[0061] Referring to FIG. 5, this figure is a flowchart of another multivariate time series processing method according to an embodiment of the present application. As shown in FIG. 5, the method includes:

[0062] S501: A time series set is obtained, and the time series set includes a plurality of first time series segments.

[0063] Here, for the specific implementation of S501, reference may be made to the relevant description of S201 above, and the present embodiment is not limited thereto.

[0064] S502: For any one first time series segment in the set of time series, input the first time series segment into a graph neural network to predict a multivariate parameter value corresponding to a first time.

[0065] Here, for the specific implementation of S502, reference may be made to the relevant description of S202 above, and the present embodiment is not limited thereto.

[0066] S503: Obtain a second time series segment for the multivariate parameter value corresponding to the first time.

[0067] In this embodiment, after predicting a multivariate parameter value corresponding to a first time based on a first time series segment, a series corresponding to the earliest time in the first time series segment is deleted, and the predicted multivariate parameter value corresponding to the first time is added to construct a second time series segment, where the second time series segment includes the multivariate parameter value corresponding to the first time, and the length of the second time series segment is equal to the length of the first time series segment.

[0068] For example, in the schematic diagram of different time series segments shown in Fig. 6, the length of the first time series segment is So, and the white box is the multivariate parameter value corresponding to the first time predicted based on the first time series segment. The bottom So-1 values ​​in the first time series segment and the multivariate parameter value corresponding to the first time are configured as a second time series segment.

[0069] S504: The second time series segment is input to the graph neural network to predict a multivariate parameter value corresponding to a second time.

[0070] After obtaining the second time series segment, the second time series segment is input into the graph neural network to predict a multivariate parameter value corresponding to a second time instant, where the second time instant is the next time instant adjacent to the first time instant.

[0071] Specifically, the second time series segment and the multivariate sequence label corresponding to the second time are input into an encoder of a graph neural network to obtain noise features corresponding to each variable and prediction features of each variable at the second time, and the prediction features of each variable at the second time and the noise features corresponding to each variable are input into a de-encoder of a graph neural network to obtain multivariate parameter values ​​corresponding to the second time. Here, for a specific implementation of using a graph neural network to predict multivariate parameter values ​​corresponding to the second time, please refer to the relevant description of S202 above.

[0072] For example, in FIG. 6, the black box is the multivariate parameter value corresponding to the second time obtained based on the second time series segment prediction. It should be noted that when the number of predictions is increased, the bottom So-1 values ​​in the second time series segment and the multivariate parameter value corresponding to the second time are configured as a third time series segment, and the multivariate parameter value corresponding to the third time is predicted based on the third time series segment. By analogy in sequence, multiple predictions can be realized.

[0073] S505: An optimization function is determined based on the multivariate parameter values ​​corresponding to the plurality of second time instants and the multivariate sequence labels corresponding to the second time instants.

[0074] For each first time series segment, after determining the multivariate parameter values ​​corresponding to the second time, an optimization function is determined based on the multivariate parameter values ​​corresponding to the plurality of second times and the multivariate sequence labels corresponding to the plurality of second times, where the optimization function includes a causal matrix and a loss function determined utilizing the multivariate parameter values ​​corresponding to the second time and the multivariate sequence labels corresponding to the second time.

[0075] Here, for a specific implementation of determining the optimization function using the multivariate parameter value corresponding to the second time and the multivariate sequence label corresponding to the second time, reference may be made to the relevant description of S203 above.

[0076] S506: Determine the value of each parameter in the causal matrix with the goal of minimizing the function.

[0077] S507: Determine causal relationships between multiple variables in the multivariate time series based on the values ​​of each parameter in the causal matrix.

[0078] Here, for the specific implementation of S506 and S507, please refer to the relevant description of S204 and S205 above, and this embodiment will not be further described here.

[0079] Based on the above method embodiments, the embodiments of the present application further provide an apparatus for processing multivariate time series, which will be described below in conjunction with the drawings.

[0080] Referring to FIG. 7, this figure is a structural diagram of an apparatus for processing multivariate time series according to an embodiment of the present application, where the apparatus 700 includes an obtaining unit 701, a predicting unit 702 and a determining unit 703.

[0081] The acquiring unit is used to acquire a time series set, the time series set including a plurality of first time series segments, each of the first time series segments in the plurality of first time series segments has the same length, and the first time series segments are multivariate time series; a prediction unit is used for inputting a first time series segment in the set of time series into a graph neural network to predict a multivariate parameter value corresponding to a first time, the first time being a next time adjacent to a latest time in the first time series segment; The determination unit is used to determine an optimization function including a loss function and a causal matrix of the graph neural network based on multivariate parameter values ​​corresponding to the plurality of first times and multivariate sequence labels corresponding to the plurality of first times; The determination unit is further adapted to determine values ​​of each parameter in the causal matrix with the goal of minimizing the optimization function; The determining unit is further used for determining a causal relationship between multiple variables in the multivariate time series based on values ​​of each parameter in the causal matrix.

[0082] In one possible implementation manner, the determining unit 703 is specifically used for obtaining a second time series segment for any one multivariate parameter value corresponding to the first time, where the second time series segment includes a multivariate parameter value corresponding to the first time, and the length of the second time series segment is equal to that of the first time series segment; inputting the second time series segment into the graph neural network to predict a multivariate parameter value corresponding to a second time, which is the next time adjacent to the first time; and determining an optimization function based on a plurality of multivariate parameter values ​​corresponding to the second time and a multivariate sequence label corresponding to the second time.

[0083] In one possible implementation, the determining unit 703 is specifically used to determine that a time-lag causal relationship exists among multiple variables when the value of one parameter in the causal matrix is ​​not zero.

[0084] In one possible implementation manner, the prediction unit 702 is specifically used for inputting the multivariate sequence label corresponding to the first time series segment and the first time into the encoder of the graph neural network to obtain noise features corresponding to each variable and prediction features of each variable at the first time, and inputting the prediction features of each variable at the first time and noise features corresponding to each variable into the de-encoder of the graph neural network to obtain multivariate parameter values ​​corresponding to the first time.

[0085] In one possible implementation, the determining unit 703 is further used for determining a target parameter whose value in the causal matrix is ​​less than a predefined threshold and setting the value of the target parameter to 0 before determining the causal relationship between multiple variables in the multivariate time series based on the value of each parameter in the causal matrix.

[0086] In one possible implementation manner, the acquisition unit 701 is specifically used for acquiring an observation window and a sliding step size, and performing time-series sliding based on the observation window and the sliding step size to obtain a plurality of first time series segments.

[0087] In one possible implementation, the length of the observation window is greater than or equal to the maximum time lag value.

[0088] It should be noted that for the specific implementation of each unit in this embodiment, reference can be made to the relevant descriptions in the above method embodiments.

[0089] The division of the units according to the embodiments of the present application is a schematic and merely a logical functional division, and there may be other division methods when actually realized. Each functional unit in the embodiments of the present application may be integrated into one processing unit, each unit may exist physically alone, or two or more units may be integrated into one unit. For example, in the above embodiment, the processing unit and the transmission unit may be the same unit or different units. The integrated units may be realized in the form of hardware or software functional units.

[0090] Referring to FIG. 8, a structural schematic diagram of an electronic device 800 suitable for implementing the embodiments of the present disclosure is shown. The terminal devices in the embodiments of the present disclosure may include, but are not limited to, mobile terminals such as mobile phones, notebook computers, digital broadcast receivers, PDAs (personal digital assistants), PADs (tablet computers), PMPs (portable multimedia players), in-vehicle terminals (e.g., in-vehicle navigation terminals), and fixed terminals such as digital TVs and desktop computers. The electronic device shown in FIG. 8 is merely an example and should not impose any limitations on the functions and range of use of the embodiments of the present disclosure.

[0091] 8, the electronic device 800 may include a processing unit (e.g., a central processor, a graphics processor, etc.) 801, which can perform various appropriate operations and processes based on programs stored in a read-only memory (ROM) 802 or programs loaded from a storage device 808 into a random access memory (RAM) 803. The RAM 803 further stores various programs and data necessary for the operation of the electronic device 800. The processing unit 801, the ROM 802, and the RAM 803 are connected to each other via a bus 804. An input / output (I / O) interface 805 is also connected to the bus 804.

[0092] In general, input devices 806 including, for example, a touch screen, a touch board, a keyboard, a mouse, a camera, a microphone, an accelerometer, a gyro, etc., output devices 807 including, for example, a liquid crystal display (LCD), a speaker, a vibrator, etc., storage devices 808 including, for example, a magnetic tape, a hard disk, etc., and communication devices 809 can be connected to the I / O interface 805. The communication devices 809 can allow the electronic device 800 and other devices to exchange data by wireless or wired communication. Although FIG. 8 shows the electronic device 800 having various devices, it should be understood that it is not required to implement or include all the devices shown. Alternatively, more or fewer devices can be implemented or included.

[0093] In particular, according to an embodiment of the present disclosure, the process described with reference to the above flowcharts can be implemented as a computer software program. For example, an embodiment of the present disclosure includes a computer program product, which includes a computer program carried on a non-transitory computer-readable medium, the computer program including a program code for executing the method shown in the flowchart. In such an embodiment, the computer program may be downloaded and installed from a network via the communication device 809, or may be installed from the storage device 808, or may be installed from the ROM 802. When the computer program is executed by the processing device 801, the above functions defined in the method of the embodiment of the present disclosure are performed.

[0094] The electronic device according to the embodiments of the present disclosure belongs to the same inventive idea as the method according to the above-mentioned embodiments, and the technical details not described in detail in this embodiment can be referred to the above-mentioned embodiments, and this embodiment has the same beneficial effects as the above-mentioned embodiments.

[0095] An embodiment of the present disclosure provides a computer storage medium, on which a computer program is stored, which, when executed by a processor, implements a method according to the above embodiment.

[0096] It should be explained that the above computer readable medium of the present disclosure may be a computer readable signal medium or a computer readable storage medium or any combination of both. The computer readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination of the above. More specific examples of the computer readable storage medium may include, but are not limited to, an electrical connection having one or more conductors, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In the present disclosure, the computer readable storage medium may be any tangible medium that contains or stores a program, which may be used by or in combination with an instruction execution system, apparatus, or device. However, in this disclosure, a computer-readable signal medium may include a data signal propagated in baseband or as part of a carrier, on which computer-readable program code is carried. Such a propagated data signal may take various forms, including but not limited to an electromagnetic signal, an optical signal, or any suitable combination of the above. A computer-readable signal medium may be any computer-readable medium other than a computer-readable storage medium, which may transmit, propagate, or transmit a program for use by or in combination with an instruction execution system, apparatus, or device. The program code contained on the computer-readable medium may be transmitted by any suitable medium, including but not limited to an electrical wire, an optical cable, RF (radio frequency), or the like, or any suitable combination of the above.

[0097] In some embodiments, the clients and servers may communicate using any network protocol now known or hereafter developed, such as HyperText Transfer Protocol (HTTP), and may interconnect with any form or medium of digital data communication (e.g., a communications network). Examples of communications networks include local area networks ("LANs"), wide area networks ("WANs"), the World Wide Web (e.g., the Internet), and end-to-end networks (e.g., ad hoc end-to-end networks), and any networks now known or hereafter developed.

[0098] The computer-readable medium may be included in the electronic device, or may be a standalone entity not assembled in the electronic device.

[0099] The computer-readable medium has one or more programs stored thereon, which, when executed by the electronic device, can cause the electronic device to perform the method.

[0100] Computer program code for carrying out the operations of the present disclosure can be written in one or a variety of programming languages ​​or combinations thereof, including, but not limited to, object-oriented programming languages ​​such as Java, Smalltalk, C++, and also including general procedural programming languages ​​such as "C" or similar. The program code may be executed entirely on the user computer, partially on the user computer, as a separate software package, partially on the user computer and partially on a remote computer, or entirely on a remote computer or server. In the case of a remote computer, the remote computer may be connected to the user computer via any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computer (e.g., connected via the Internet using an Internet Service Provider).

[0101] The flowcharts and block diagrams in the drawings illustrate possible system architectures, functions and operations of systems, methods and computer program products according to various embodiments of the present disclosure. In this regard, each block in the flowcharts or block diagrams may represent a module, program segment or part of code, which includes one or more executable instructions for implementing the specified logic function. It should also be noted that in some implementations, as a replacement, the functions illustrated in the blocks may occur in a different order than those illustrated in the drawings. For example, two consecutive blocks may actually be essentially executed in parallel, or may be executed in a reverse order depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, may be realized by a dedicated hardware-based system that performs the specified functions or operations, or by a combination of dedicated hardware and computer instructions.

[0102] The units described in relation to the embodiments of the present disclosure may be implemented in a software or hardware manner, and the names of the units / modules are not intended to limit the units themselves in any case.

[0103] The functions described herein above may be performed, at least in part, by one or more hardware logic components. For example, and without limitation, exemplary types of hardware logic components that may be used include field programmable gate arrays (FPGAs), application specific integrated circuits (ASICs), application specific standard products (ASSPs), systems on chips (SOCs), complex programmable logic devices (CPLDs), etc.

[0104] In the context of this disclosure, a machine-readable medium may be a tangible medium that may contain or store a program for use by or in combination with an instruction execution system, device, or apparatus. A machine-readable medium may include, but is not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, device, or apparatus, or any suitable combination of the above. More specific examples of machine-readable storage media include one or more line-based electrical connections, portable computer disks, hard disks, random access memories (RAMs), read-only memories (ROMs), erasable programmable read-only memories (EPROMs or flash memories), optical fibers, portable compact read-only memories (CD-ROMs), optical storage devices, magnetic storage devices, or any suitable combination of the above.

[0105] It should be noted that in this specification, each embodiment is described in a progressive manner, and each embodiment is described by focusing on the differences with other embodiments, and the same and similar parts between each embodiment can be referred to each other. For the system or device disclosed in the embodiment, it corresponds to the method disclosed in the embodiment, so the description is relatively simple, and the relevant points can be referred to the description of the method part.

[0106] It should be understood that in this application, "at least one" means one or more, and "multiple" means two or more. "And / or" is used to describe the relationship between related objects, and indicates that three relationships may exist, for example, "A and / or B" may represent three situations: only A exists, only B exists, and both A and B exist, where A and B may be singular or plural. The character " / " generally indicates that the related objects before and after are in an "or" relationship. "At least one of" or similar expressions refers to any combination of these items, including any combination of a single item or multiple items. For example, at least one of a, b, or c may represent a, b, c, "a and b," "a and c," "b and c," or "a and b and c," where a, b, and c may be singular or plural.

[0107] It should be further explained that, in this specification, relational terms such as first and second are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply that any such actual relationship or order exists between those entities or operations. And, the terms "comprise", "include", or any other variation thereof are intended to cover a non-exclusive "comprise", whereby a process, method, article or device that includes a set of elements not only includes those elements, but also includes other elements not expressly listed, or further includes elements inherent in such process, method, article or device. In the absence of further limitations, an element defined by the phrase "comprises a" does not exclude the presence of other identical elements in the process, method, article or device that includes said element.

[0108] The steps of a method or algorithm described in connection with the embodiments disclosed herein may be embodied directly in hardware, in a software module executed by a processor, or in a combination of both. The software module may be located in a Random Access Memory (RAM), memory, a Read Only Memory (ROM), an Electrically Programmable ROM, an Electrically Erasable Programmable ROM, a register, a hard disk, a removable disk, a CD-ROM, or any other form of storage medium known in the art.

[0109] The above description of the disclosed embodiments enables those skilled in the art to realize or use the present application. Although various modifications of these embodiments will be apparent to those skilled in the art, the general principles defined herein may be embodied in other embodiments without departing from the spirit or scope of the present application. Therefore, the present application is not limited to these embodiments shown herein, but should be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. obtaining a time series set, the time series set including a plurality of first time series segments, each first time series segment in the plurality of first time series segments having the same length, the first time series segments being a multivariate time series; For any one first time series segment in the set of time series, input the first time series segment into a graph neural network to predict a multivariate parameter value corresponding to a first time, the first time being the next time adjacent to a latest time in the first time series segment; determining an optimization function including a loss function and a causal matrix of the graph neural network based on multivariate parameter values ​​corresponding to the plurality of first times and multivariate sequence labels corresponding to the plurality of first times; determining values ​​for each parameter in the causal matrix with the goal of minimizing the optimization function; determining causal relationships between multiple variables in the multivariate time series based on values ​​of each parameter in the causal matrix; 1. A method for processing multivariate time series, comprising:

2. Determining an optimization function based on multivariate parameter values ​​corresponding to the first times and multivariate sequence labels corresponding to the first times includes: obtaining a second time series segment for any one of the multivariate parameter values ​​corresponding to the first time, the second time series segment including the multivariate parameter value corresponding to the first time, and a length of the second time series segment being equal to a length of the first time series segment; inputting the second time series segment into the graph neural network to predict a multivariate parameter value corresponding to a second time instant that is a next time instant adjacent to the first time instant; determining an optimization function based on a plurality of multivariate parameter values ​​corresponding to the second time and a multivariate sequence label corresponding to the second time; 2. The method of claim 1, comprising:

3. Determining causal relationships among multiple variables in the multivariate time series based on values ​​of each parameter in the causal matrix includes:

3. The method of claim 1, further comprising determining that a time-lagged causal relationship exists between multiple variables if the value of one parameter in the causal matrix is ​​not zero.

4. Inputting the first time series segment into a graph neural network to predict a multivariate parameter value corresponding to a first time, comprising: inputting the multivariate sequence label corresponding to the first time series segment and the first time into an encoder of the graph neural network to obtain noise features corresponding to each variable and a predicted feature for each variable at the first time; inputting the predicted feature of each variable at the first time and the noise feature corresponding to each variable into a de-encoder of the graph neural network to obtain a multivariate parameter value corresponding to a first time; 2. The method of claim 1, comprising:

5. prior to determining a causal relationship between multiple variables in the multivariate time series based on values ​​of each parameter in the causal matrix, determining target parameters whose values ​​in the causal matrix are less than a predefined threshold; setting the value of the target parameter to 0; 2. The method of claim 1, further comprising:

6. Obtaining a set of time series is Obtaining an observation window and a slide step size; performing a time-series slide based on the observation window and the slide step size to obtain a plurality of first time-series segments; 2. The method of claim 1, comprising:

7. 7. The method of claim 6, wherein the length of the observation window is greater than or equal to a maximum time lag value.

8. an acquisition unit for acquiring a time series set, the time series set including a plurality of first time series segments, each of the first time series segments in the plurality of first time series segments having the same length, the first time series segments being a multivariate time series; a prediction unit for inputting a first time series segment in the set of time series into a graph neural network and predicting a multivariate parameter value corresponding to a first time, the first time being a next time to a latest time in the first time series segment; a determination unit for determining an optimization function including a loss function and a causal matrix of the graph neural network based on multivariate parameter values ​​corresponding to the plurality of first time instants and multivariate sequence labels corresponding to the plurality of first time instants; The determination unit is further adapted to determine values ​​of each parameter in the causal matrix with the goal of minimizing the optimization function; the determining unit is further adapted to determine a causal relationship between multiple variables in the multivariate time series based on values ​​of each parameter in the causal matrix. A multivariate time series processing device comprising:

9. An electronic device including a processor and a memory, the memory is used to store instructions or computer programs; The processor is adapted to execute the instructions or computer programs in the memory to cause the electronic device to perform the method according to any one of claims 1 to 7.

1. An electronic device comprising:

10. A computer readable storage medium having instructions stored thereon, the instructions, when run on a device, causing the device to perform a method according to any one of claims 1 to 7.

Images