Method and device for generating hypergraph structure of time series data

Abstract

The present application relates to the field of network inference technology, and in particular to a method and device for generating a hypergraph structure of time series data, wherein the method comprises: obtaining time series data of a target system; obtaining a joint probability distribution of the interaction of variables in the target system according to the time series data, and obtaining a hypergraph structure based on sampling of the joint probability distribution; inputting the state and hypergraph structure of the target system at the previous moment of operation into a preset hypergraph neural network, generating a hypergraph structure of time series data to predict the operating state of the target system at the next moment. Thus, the problem that the related technology fails to filter out noise unrelated to the network structure while retaining the rich information of the time series data collected by multiple observations when inferring the interaction of variables in the system through time series data, and further increases the difficulty of time series data inference when the system has nonlinear characteristics, thereby reducing the performance of time series data inference, etc.

Description

Technical Field

[0001] The present application relates to the field of network inference technology, and in particular to a method and device for generating a hypergraph structure of time series data. Background Art

[0002] In related technologies, the research method of complex systems is to summarize prior knowledge based on observed actual phenomena, establish a set of network dynamics equations, and generate a set of time series through simulation to reflect the state of the system at different time points.

[0003] However, in related technologies, due to the complexity and diversity of the system, it is difficult to observe the accurate and complete system structure in many cases, and there is an information underdetermination problem. At the same time, with the development of technologies such as big data, related technologies can record a large amount of time series data generated by the operation of a complex system. Due to the large amount of noise in the data, especially in specific scenarios, the system itself is composed of a set of mutually coupled random variables and does not follow fixed system dynamics, which makes it difficult to combine traditional system modeling methods with large amounts of micro data.

[0004] Therefore, it is necessary to automatically build a dynamic model of the system operation through the performance of the system operation, that is, to infer the interaction structure of the variables in the system through the time series data of the system operation, which is also called network inference. However, network inference methods include the use of statistics, Bayesian inference, and probabilistic graphical models. When dealing with large-scale noisy data, they fail to filter out noise that is irrelevant to the network structure while retaining the rich information of the time series data collected by multiple observations. Moreover, when the system has nonlinear characteristics, it will further increase the difficulty of time series data inference, reducing the performance of time series data inference, which needs to be solved urgently. Summary of the invention

[0005] The embodiments of the present application are made based on the following objectives:

[0006] The hypergraph is introduced to mathematically model the interaction structure of variables in the system, and a variational autoencoder is designed. The variational autoencoder takes the time series data of the system operation as input, and includes three parts: encoder, generator and decoder. The encoder obtains the joint probability distribution of the interaction of variables in the system by introducing RNN (Recurrent Neural Network). The generator samples the hypergraph structure from the joint probability distribution. Each node in the hypergraph structure represents a variable in the system, and the hyperedge in the hypergraph structure represents the interaction structure of the variables. Finally, the decoder inputs the state of the system at the previous moment and the hypergraph structure into HGNN (Hypergraph Neural Nerworks) to predict the state of the system at the next moment. In the learning process, the entire variational autoencoder is trained by the back propagation algorithm, and finally the interaction structure of the variables in the system is inferred, which makes up for the lack of information underdetermination caused by the difficulty of directly observing the interaction of variables, and realizes the approximate fitting of the dynamic model of the system operation.

[0007] The present application provides a method and device for generating a hypergraph structure of time series data to solve the problems in the related art of inferring the interaction of variables in a system through time series data, failing to filter out noise unrelated to the network structure while retaining the rich information of time series data collected through multiple observations, and further increasing the difficulty of time series data inference when the system has nonlinear characteristics, reducing the performance of time series data inference, and failing to meet the user's usage needs.

[0008] The first aspect of the present application provides a method for generating a hypergraph structure of time series data, comprising the following steps: acquiring time series data of a target system; obtaining a joint probability distribution of interactions between variables in the target system based on the time series data, and obtaining a hypergraph structure based on sampling of the joint probability distribution; inputting the state of the target system at a previous moment of operation and the hypergraph structure into a preset hypergraph neural network to generate a hypergraph structure of time series data to predict the operating state of the target system at the next moment.

[0009] Optionally, in one embodiment of the present application, each node in the hypergraph structure identifies a variable of the target system, and the hyperedges of the hypergraph structure represent the interaction structure of the variables.

[0010] Optionally, in one embodiment of the present application, the joint probability distribution of the interaction of variables in the target system is obtained according to the time series data, and a hypergraph structure is obtained based on the sampling of the joint probability distribution, including: calculating the probability distribution of the interaction of variables in the target system according to the hidden state at the data moment; and obtaining an association matrix based on the probability distribution sampling to represent the hypergraph structure.

[0011] Optionally, in one embodiment of the present application, the state of the target system at a previous moment of operation and the hypergraph structure are input into a preset hypergraph neural network to generate a time series data hypergraph structure to predict the operating state of the target system at the next moment, including: calculating the hyperedge weight diagonal matrix and the node degree diagonal matrix of the hypergraph structure; calculating the variable state prediction value at the next moment based on the association matrix, the state of at least one variable in the system at the current moment, the hyperedge weight diagonal matrix, the node degree diagonal matrix, the parameters of the first hypergraph convolution layer of the preset hypergraph neural network, the parameters of the second hypergraph convolution layer and the parameters of the fully connected output layer; calculating the current prediction error according to the variable state prediction value and the actual observation value, and calculating the total loss function value according to the smoothness constraint penalty term of the hypergraph structure and the current prediction error; performing back propagation calculation according to the total loss function value to calculate the parameters of the variational autoencoder until the convergence condition is met.

[0012] Optionally, in one embodiment of the present application, the association matrix is:

[0013]

[0014] Among them, H ij is whether the i-th node interacts with the j-th node, ∈ v = -log(-log(s)), where s is randomly drawn from Uniform(0;1) and τ is a hyperparameter called temperature that controls the interpolation between the discrete distribution and the continuous categorical density.

[0015] The second aspect of the present application provides a device for generating a hypergraph structure of time series data, including: an acquisition module for acquiring time series data of a target system; a collection module for obtaining a joint probability distribution of interactions between variables in the target system based on the time series data, and obtaining a hypergraph structure based on sampling of the joint probability distribution; a generation module for inputting the state of the target system at a previous moment of operation and the hypergraph structure into a preset hypergraph neural network to generate a hypergraph structure of time series data to predict the operating state of the target system at the next moment.

[0016] Optionally, in one embodiment of the present application, each node in the hypergraph structure identifies a variable of the target system, and the hyperedges of the hypergraph structure represent the interaction structure of the variables.

[0017] Optionally, in one embodiment of the present application, the acquisition module includes: a calculation unit, used to calculate the probability distribution of variable interactions in the target system according to the hidden state at the data moment; and a sampling unit, used to obtain an association matrix according to the probability distribution sampling to represent the hypergraph structure.

[0018] Optionally, in one embodiment of the present application, the generation module includes: a first calculation unit, used to calculate the hyperedge weight diagonal matrix and the node degree diagonal matrix of the hypergraph structure; a second calculation unit, used to calculate the variable state prediction value at the next moment based on the association matrix, the state of at least one variable in the system at the current moment, the hyperedge weight diagonal matrix, the node degree diagonal matrix, the parameters of the first hypergraph convolution layer of the preset hypergraph neural network, the parameters of the second hypergraph convolution layer and the parameters of the fully connected output layer; a third calculation unit, used to calculate the current prediction error based on the variable state prediction value and the actual observation value, and calculate the total loss function value based on the smoothness constraint penalty term of the hypergraph structure and the current prediction error; a fourth calculation unit, used to perform back propagation calculation based on the total loss function value, and calculate the parameters of the variational autoencoder until the convergence conditions are met.

[0019] Optionally, in one embodiment of the present application, the association matrix is:

[0020]

[0021] Among them, H ij is whether the i-th node interacts with the j-th node, ∈ v = -log(-log(s)), where s is randomly drawn from Uniform(0;1) and τ is a hyperparameter called temperature that controls the interpolation between the discrete distribution and the continuous categorical density.

[0022] The third aspect of the present application provides an electronic device, comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the method for generating a hypergraph structure of time series data as described in the above embodiment.

[0023] The fourth aspect of the present application provides a computer-readable storage medium, which stores computer instructions, and the computer instructions are used to enable the computer to execute the method for generating a hypergraph structure of time series data as described in the above embodiment.

[0024] The embodiment of the present application can obtain the joint probability distribution of the interaction of variables in the target system based on the time series data, and obtain the hypergraph structure based on the sampling of the joint probability distribution, so as to input the state and hypergraph structure of the target system at the previous moment of operation into the preset hypergraph neural network, generate the time series data hypergraph structure, and then predict the operation state of the target system at the next moment, and automatically generate the hypergraph structure through the time series of the variable state in the system, so as to model the high-order interaction of the variables in the system, and realize the inference of the high-order interaction structure of the variables in the system through the time series data generated by the operation of the system, and effectively improve the performance of network inference in complex systems. Thus, the problem that when the interaction of variables in the system is inferred through time series data in the related art, the noise unrelated to the network structure cannot be filtered while retaining the rich information of the time series data collected by multiple observations, and when the system has nonlinear characteristics, the difficulty of time series data inference will be further increased, the performance of time series data inference will be reduced, and the user's usage needs cannot be met.

[0025] Additional aspects and advantages of the present application will be given in part in the description below, and in part will become apparent from the description below, or will be learned through the practice of the present application. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] The above and / or additional aspects and advantages of the present application will become apparent and easily understood from the following description of the embodiments in conjunction with the accompanying drawings, in which:

[0027] Figure 1 A flowchart of a method for generating a hypergraph structure of time series data provided according to an embodiment of the present application;

[0028] Figure 2 A schematic diagram of a method for generating a hypergraph structure of time series data according to a specific embodiment of the present application;

[0029] Figure 3 It is a structural schematic diagram of a device for generating a hypergraph structure of time series data according to an embodiment of the present application;

[0030] Figure 4 It is a schematic diagram of the structure of an electronic device provided according to an embodiment of the present application. DETAILED DESCRIPTION

[0031] Embodiments of the present application are described in detail below, and examples of the embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are intended to be used to explain the present application, and should not be construed as limiting the present application.

[0032] The following describes the method and device for generating the hypergraph structure of time series data in the embodiment of the present application with reference to the accompanying drawings. In view of the problem that when the related technology mentioned in the background technology center mentioned above infers the interaction of variables in the system through time series data, it fails to filter out the noise unrelated to the network structure while retaining the rich information of the time series data collected by multiple observations, and when the system has nonlinear characteristics, it will further increase the difficulty of time series data inference and reduce the performance of time series data inference, the present application provides a method for generating a hypergraph structure of time series data, in which the embodiment of the present application obtains the joint probability distribution of the interaction of variables in the target system according to the time series data, and obtains the hypergraph structure based on the sampling of the joint probability distribution, so as to input the state and hypergraph structure of the target system at the previous moment of operation into the preset hypergraph neural network, generate the hypergraph structure of time series data, and then predict the operation state of the target system at the next moment, and automatically generate the hypergraph structure through the time series of the variable state in the system to model the high-order interaction of the variables in the system, so as to realize the inference of the high-order interaction structure of the variables in the system through the time series data generated by the operation of the system, and effectively improve the performance of time series data inference in complex systems. This solves the problem that when inferring the interaction of variables in the system through time series data in the related technology, it fails to filter out noise unrelated to the network structure while retaining the rich information of the time series data collected by multiple observations, and when the system has nonlinear characteristics, it will further increase the difficulty of time series data inference, reduce the performance of time series data inference, and cannot meet the user's usage needs.

[0033] Specifically, Figure 1 A flowchart of a method for generating a hypergraph structure of time series data provided in an embodiment of the present application.

[0034] like Figure 1 As shown, the method for generating the time series data hypergraph structure includes the following steps:

[0035] In step S101 , time series data of a target system is acquired.

[0036] That is, the embodiment of the present application introduces a hypergraph to mathematically model the interaction structure of variables in the system. First, time series data is used as input, and there are many ways to obtain time series data, which will not be described in detail here.

[0037] In step S102, the joint probability distribution of the interaction of variables in the target system is obtained according to the time series data, and the hypergraph structure is obtained based on the sampling of the joint probability distribution.

[0038] After taking time series data as input, the embodiment of the present application can utilize a variational autoencoder designed to perform modeling, which includes three parts: an encoder, a generator, and a decoder.

[0039] Specifically, the embodiment of the present application inputs the acquired time series data of the target system into a variational autoencoder, and the encoder obtains the joint probability distribution of the interaction of variables in the system by introducing RNN, so that the generator samples the hypergraph structure from the joint probability distribution.

[0040] In one embodiment of the present application, each node in the hypergraph structure identifies a variable of the target system, and the hyperedges of the hypergraph structure represent the interaction structure of the variables.

[0041] Furthermore, in one embodiment of the present application, a joint probability distribution of interactions between variables in a target system is obtained based on time series data, and a hypergraph structure is obtained based on sampling of the joint probability distribution, including: calculating the probability distribution of interactions between variables in the target system based on hidden states at the data moment; and obtaining an association matrix based on sampling of the probability distribution to represent the hypergraph structure.

[0042] As a possible way to achieve this, Figure 2 As shown, the embodiment of the present application may include the following steps:

[0043] Step S1: Consider a system with N variables, and obtain the time series data X∈R of the system running at T moments through observation N×T×D , let x t =X[:,t,:] represents the state of N variables in the system at the tth moment.

[0044] For example, the present application embodiment uses a directed communication network dynamic system to illustrate a directed communication network including five individuals. The individuals exchange information under the directed communication network topology. The information exchange between members is defined as follows: If the hypergraph H ij =1, then individual i can receive information from individual j, and vice versa. Each individual determines the position to move to at the next moment based on the information received at the current moment. Due to the lack of information, the topological structure of the directed communication network is unknown. Only the position information of the five individuals at the 50 consecutive moments before and after measured by the sensor is used, and the position information of the five individuals at the first 49 moments is used as the time series data X∈R for the system operation. 5×49×2 .

[0045] Step S2: After obtaining the time series data of the system operation, input it into the encoder. The encoder consists of two parts: the forward encoder and the reverse encoder. The forward encoder RNN f Input x in the order of time 1 to T 1 to x T , cyclically update the hidden state from time 1 to T Specifically, we first randomly initialize the forward encoder RNN f The hidden state of the initial hidden state is Then the calculation begins in a loop. In each cycle, input x at time t. t and the hidden state passed down at time t-1 Forward Encoder RNN f The hidden state at time t is obtained by the following formula

[0046]

[0047]

[0048]

[0049]

[0050]

[0051]

[0052] in, is the hidden state at time t, is the initial hidden state, is the hidden state, and For the forward encoder RNN f The parameters that need to be learned finally get the forward encoding of the data

[0053] Backward Encoder RNN b Input x in the order of time T to 1 T to x 1 , cyclic update to obtain the hidden state from time T to 1 Specifically, we first randomly initialize the backward encoder RNN b The hidden state of the initial hidden state is Then the calculation begins in a loop. In each cycle, input x at time t. t and the hidden state passed down at time t+1 Forward Encoder RNN b The hidden state at time t is obtained by the following formula

[0054]

[0055]

[0056]

[0057]

[0058]

[0059]

[0060] in, and For the backward encoder RNN b The parameters that need to be learned finally get the backward encoding of the data

[0061] The forward encoding of the data time T and the backward encoding at time 1 Connect them together to get the final output e of the encoder, that is:

[0062]

[0063] in, is the forward encoding, For backward coding, let e i =e[i,:] represents the encoding of the i-th variable in the system.

[0064] Step S3: Use a hypergraph structure to model the interaction of variables in the system. Let the N variables in the system be represented as N nodes. For each node, all nodes that interact with it form a hyperedge with the node. The hypergraph structure can be represented by the association matrix H∈R N×N Indicates that H ij Indicates whether the i-th node interacts with the j-th node. If so, H ij =1, otherwise H ij = 0. Due to the lack of information, H cannot be obtained directly. It is necessary to calculate the probability distribution P(H) of the interaction of variables in the system based on the encoder output e, and sample H from P(H).

[0065] Step S4: Calculate the probability distribution P(H) of the interaction between variables in the system. To reduce the computational complexity, assume that H ij (1≤i≤N,1≤j≤N) conforms to independent and identical distribution and is a Bernoulli distribution, that is:

[0066] P(H)=∏ 1≤i≤N,1≤j≤N P(H ij ),

[0067] P(H ij )=Bern(H ij ∣ρ ij ),ρ ij >0,

[0068] Among them, H ijIt indicates whether the i-th node interacts with the j-th node, and P(H) represents the probability distribution of the interaction between variables in the system.

[0069] where ρ ij Calculated by the following formula:

[0070] ρ ij =σ(e i e j ),

[0071] Step S5: Use the Gumbel-Softmax function to perform a differentiable sampling process to obtain the correlation matrix H.

[0072] Optionally, in one embodiment of the present application, the association matrix is:

[0073]

[0074] Among them, H ij is whether the i-th node interacts with the j-th node, ∈ v = -log(-log(s)), where s is randomly drawn from Uniform(0;1) and τ is a hyperparameter called temperature that controls the interpolation between the discrete distribution and the continuous categorical density.

[0075] Optionally, in one embodiment of the present application, each node in the hypergraph structure identifies a variable of the target system, and the hyperedges of the hypergraph structure represent the interaction structure of the variables.

[0076] It can be understood that the generator samples the hypergraph structure from the joint probability distribution. Each node in the hypergraph structure represents a variable in the target system, and the hyperedges in the hypergraph structure represent the interaction structure of the variables, thereby inferring the interaction structure of the variables in the system, making up for the lack of information uncertainty caused by the difficulty in directly observing the interaction between variables, and effectively improving the performance of time series data inference in complex systems.

[0077] In step S103, the state and hypergraph structure of the target system at the previous moment of operation are input into a preset hypergraph neural network to generate a time series data hypergraph structure to predict the operating state of the target system at the next moment.

[0078] That is, the decoder finally inputs the state of the system at the previous moment and the hypergraph structure into the hypergraph neural network (HGNN) to predict the state of the system at the next moment. During the learning process, the entire variational autoencoder is trained through the back-propagation algorithm, and finally the interaction structure of the variables in the system is inferred, which makes up for the lack of information due to the difficulty of directly observing the interaction of variables, and realizes the approximate fitting of the dynamic model of the system operation.

[0079] Optionally, in one embodiment of the present application, the state and hypergraph structure of the target system at a previous moment of operation are input into a preset hypergraph neural network to generate a time series data hypergraph structure to predict the operating state of the target system at the next moment, including: calculating the hyperedge weight diagonal matrix and the node degree diagonal matrix of the hypergraph structure; calculating the variable state prediction value at the next moment based on the association matrix, the state of at least one variable in the system at the current moment, the hyperedge weight diagonal matrix, the node degree diagonal matrix, the parameters of the first hypergraph convolution layer of the preset hypergraph neural network, the parameters of the second hypergraph convolution layer and the parameters of the fully connected output layer; calculating the current prediction error according to the variable state prediction value and the actual observation value, and calculating the total loss function value according to the smoothness constraint penalty term of the hypergraph structure and the current prediction error; performing back propagation calculation according to the total loss function value to calculate the parameters of the variational autoencoder until the convergence conditions are met.

[0080] In the actual implementation process, Figure 2 As shown, the embodiment of the present application uses a time series data hypergraph structure to predict the operating state of the target system at the next moment, which is described in detail below. The specific steps are as follows:

[0081] Step S6: H and x T The input is sent to the decoder based on the hypergraph neural network to predict the state of the variables in the system at the next moment. Specifically, let 1 represent a vector with all elements set to 1. First, the hyperedge weight diagonal matrix D of the hypergraph is calculated by the following formula: e and the node degree diagonal matrix D v :

[0082] D e =diag(1 T H),

[0083] D v =diag(HW1),

[0084] Among them, D e is the hyperedge weight diagonal matrix, D v is the node degree diagonal matrix.

[0085] The decoder based on the hypergraph neural network consists of two hypergraph convolutional layers plus a fully connected output layer. The parameters of the first hypergraph convolutional layer are Θ (1) , the parameter of the second hypergraph convolutional layer is Θ (2) , the parameters of the fully connected output layer are W and b, based on the above H, x T , D e , D v , Θ (1) , Θ (2) , W and b, calculate the predicted value x' of the variable state in the system at time T+1 by the following formula T+1 :

[0086]

[0087]

[0088]

[0089]

[0090] Among them, Θ (1) is the parameter of the first hypergraph convolutional layer, Θ (2) are the parameters of the second hypergraph convolutional layer, Predict values ​​for the states of variables in the system.

[0091] Step S7: Based on the predicted value of the variable state in the system at time T+1 and the actual observed value x T+1 Calculate the prediction error L emp :

[0092] in, is the predicted value of the variable state in the system, x T+1 is the actual observed value, L emp To calculate the prediction error.

[0093] Step S8: Let tr(·) function represent the sum of the main diagonal elements of the matrix. Based on the above H and x t (t=1,…,T) Calculate the penalty term of the hypergraph structure smoothness constraint:

[0094]

[0095] Among them, L reg is the penalty term for the smoothness constraint of the hypergraph structure.

[0096] Step S9: Based on the above L emp and L reg Calculate the total loss function value:

[0097] L=L emp +L reg ,

[0098] Among them, L is the total loss function value, L reg is the penalty term for the hypergraph structure smoothness constraint, L emp To calculate the prediction error.

[0099] Step S10: Calculate encoder parameters based on L through back propagation algorithm and and the decoder parameters Θ (1) , Θ (2), W and b’s gradients, and update the parameter values ​​through the Adam algorithm.

[0100] Step S11: Repeat steps S2 to S10 until the iteration converges and the optimization solution is completed.

[0101] Specifically, after solving, the final generated hypergraph structure H is obtained. In this way, the topological structure of the directed communication network can be inferred.

[0102] In summary, the embodiments of the present application automatically generate a hypergraph structure through the time series of the variable states in the target system, thereby modeling the high-order interactions of the variables in the system, and inferring the high-order interaction structure of the variables in the system through the time series data generated by the operation of the system, solving the information underdetermination problem caused by the difficulty in directly observing the interactions between variables in complex systems. In the process of optimizing and updating the encoder and decoder, the embodiments of the present application simultaneously consider the prediction accuracy of the variable states and the hypergraph structure constraints of the interactions between the variables. By learning and optimizing the parameters of the encoder and decoder, the system dynamics model can be fit more accurately.

[0103] According to the method for generating a hypergraph structure of time series data proposed in the embodiment of the present application, the joint probability distribution of the interaction of variables in the target system can be obtained based on the time series data, and the hypergraph structure can be obtained based on the sampling of the joint probability distribution, so that the state and hypergraph structure of the target system at the previous moment of operation are input into the preset hypergraph neural network to generate a hypergraph structure of time series data, and then the operating state of the target system at the next moment is predicted. The hypergraph structure is automatically generated through the time series of the variable states in the system to model the high-order interaction of the variables in the system, and the high-order interaction structure of the variables in the system can be inferred through the time series data generated by the operation of the system, which effectively improves the performance of time series data inference in complex systems.

[0104] Next, a device for generating a hypergraph structure of temporal data according to an embodiment of the present application will be described with reference to the accompanying drawings.

[0105] Figure 3 It is a block diagram of a device for generating a hypergraph structure of time series data according to an embodiment of the present application.

[0106] like Figure 3 As shown, the device 10 for generating a hypergraph structure of time series data includes: an acquisition module 100 , a collection module 200 and a generation module 300 .

[0107] Specifically, the acquisition module 100 is used to acquire time series data of the target system.

[0108] The acquisition module 200 is used to obtain the joint probability distribution of the interaction of variables in the target system according to the time series data, and obtain the hypergraph structure based on the joint probability distribution sampling.

[0109] The generation module 300 is used to input the state and hypergraph structure of the target system at the previous moment of operation into a preset hypergraph neural network, generate a time series data hypergraph structure, and predict the operating state of the target system at the next moment.

[0110] Optionally, in one embodiment of the present application, each node in the hypergraph structure identifies a variable of the target system, and the hyperedges of the hypergraph structure represent the interaction structure of the variables.

[0111] Optionally, in one embodiment of the present application, the acquisition module 200 includes: a calculation unit and a sampling unit.

[0112] Among them, the computing unit is used to calculate the probability distribution of the interaction of variables in the target system according to the hidden state at the data moment.

[0113] The sampling unit is used to obtain the association matrix according to the probability distribution sampling to represent the hypergraph structure.

[0114] Optionally, in one embodiment of the present application, the generation module 300 includes: a first calculation unit, a second calculation unit, a third calculation unit and a fourth calculation unit.

[0115] Among them, the first calculation unit is used to calculate the hyperedge weight diagonal matrix and the node degree diagonal matrix of the hypergraph structure.

[0116] The second calculation unit is used to calculate the predicted value of the variable state at the next moment based on the association matrix, the state of at least one variable in the system at the current moment, the hyperedge weight diagonal matrix, the node degree diagonal matrix, the parameters of the first hypergraph convolution layer of the preset hypergraph neural network, the parameters of the second hypergraph convolution layer and the parameters of the fully connected output layer.

[0117] The third calculation unit is used to calculate the current prediction error according to the variable state prediction value and the actual observation value, and calculate the total loss function value according to the smoothness constraint penalty term of the hypergraph structure and the current prediction error.

[0118] The fourth computing unit is used to perform back propagation calculation according to the total loss function value to calculate the parameters of the variational autoencoder until the convergence condition is met.

[0119] Optionally, in one embodiment of the present application, the association matrix is:

[0120]

[0121] Among them, H ij is whether the i-th node interacts with the j-th node, ∈ v= -log(-log(s)), where s is randomly drawn from Uniform(0;1) and τ is a hyperparameter called temperature that controls the interpolation between the discrete distribution and the continuous categorical density.

[0122] It should be noted that the aforementioned explanation of the embodiment of the method for generating a hypergraph structure of time series data is also applicable to the device for generating a hypergraph structure of time series data of this embodiment, and will not be repeated here.

[0123] According to the device for generating a hypergraph structure for time series data proposed in the embodiment of the present application, the joint probability distribution of the interaction of variables in the target system can be obtained according to the time series data, and the hypergraph structure can be obtained based on the sampling of the joint probability distribution, so that the state and hypergraph structure of the target system at the previous moment of operation are input into the preset hypergraph neural network to generate a hypergraph structure for time series data, and then the operation state of the target system at the next moment is predicted. The hypergraph structure is automatically generated through the time series of the variable states in the system to model the high-order interaction of the variables in the system, and the high-order interaction structure of the variables in the system can be inferred through the time series data generated by the operation of the system, which effectively improves the performance of time series data inference in complex systems.

[0124] Figure 4 A schematic diagram of the structure of an electronic device provided in an embodiment of the present application. The electronic device may include:

[0125] Memory 401 , processor 402 , and a computer program stored in the memory 401 and executable on the processor 402 .

[0126] When the processor 402 executes the program, the method for generating a hypergraph structure of time series data provided in the above embodiment is implemented.

[0127] Furthermore, the electronic device further comprises:

[0128] The communication interface 403 is used for communication between the memory 401 and the processor 402 .

[0129] The memory 401 is used to store computer programs that can be executed on the processor 402 .

[0130] The memory 401 may include a high-speed RAM memory, and may also include a non-volatile memory (non-volatile memory), such as at least one disk memory.

[0131] If the memory 401, the processor 402 and the communication interface 403 are implemented independently, the communication interface 403, the memory 401 and the processor 402 can be connected to each other through a bus and communicate with each other. The bus can be an Industry Standard Architecture (ISA) bus, a Peripheral Component (PCI) bus or an Extended Industry Standard Architecture (EISA) bus. The bus can be divided into an address bus, a data bus, a control bus, etc. For ease of representation, Figure 4 Only one thick line is used in the diagram, but this does not mean that there is only one bus or only one type of bus.

[0132] Optionally, in a specific implementation, if the memory 401, the processor 402 and the communication interface 403 are integrated on a chip, the memory 401, the processor 402 and the communication interface 403 can communicate with each other through an internal interface.

[0133] The processor 402 may be a central processing unit (CPU), or an application specific integrated circuit (ASIC), or one or more integrated circuits configured to implement the embodiments of the present application.

[0134] This embodiment also provides a computer-readable storage medium on which a computer program is stored. When the program is executed by a processor, the method for generating a hypergraph structure of temporal data as described above is implemented.

[0135] In the description of this specification, the description with reference to the terms "one embodiment", "some embodiments", "example", "specific example", or "some examples" etc. means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present application. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described may be combined in any one or N embodiments or examples in a suitable manner. In addition, those skilled in the art may combine and combine the different embodiments or examples described in this specification and the features of the different embodiments or examples, without contradiction.

[0136] In addition, the terms "first" and "second" are used for descriptive purposes only and should not be understood as indicating or implying relative importance or implicitly indicating the number of technical features indicated. Therefore, a feature defined as "first" or "second" may explicitly or implicitly include at least one of the features. In the description of this application, "N" means at least two, such as two, three, etc., unless otherwise clearly and specifically defined.

[0137] Any process or method description in a flowchart or otherwise described herein may be understood to represent a module, fragment or portion of code comprising one or more executable instructions for implementing the steps of a custom logical function or process, and the scope of the preferred embodiments of the present application includes alternative implementations in which functions may not be performed in the order shown or discussed, including performing functions in a substantially simultaneous manner or in reverse order depending on the functions involved, which should be understood by technicians in the technical field to which the embodiments of the present application belong.

[0138] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as an ordered list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by an instruction execution system, device or apparatus (such as a computer-based system, a system including a processor, or other system that can fetch instructions from an instruction execution system, device or apparatus and execute instructions), or in combination with these instruction execution systems, devices or apparatuses. For the purpose of this specification, "computer-readable medium" can be any device that can contain, store, communicate, propagate or transmit a program for use by an instruction execution system, device or apparatus, or in combination with these instruction execution systems, devices or apparatuses. More specific examples of computer-readable media (a non-exhaustive list) include the following: an electrical connection with one or N wirings (electronic devices), a portable computer disk box (magnetic device), a random access memory (RAM), a read-only memory (ROM), an erasable and programmable read-only memory (EPROM or flash memory), a fiber optic device, and a portable compact disk read-only memory (CDROM). In addition, the computer-readable medium may even be paper or other suitable medium on which the program is printed, since the program may be obtained electronically, for example, by optically scanning the paper or other medium and then editing, interpreting or processing in other suitable ways if necessary, and then stored in a computer memory.

[0139] It should be understood that the various parts of the present application can be implemented by hardware, software, firmware or a combination thereof. In the above-mentioned embodiment, the N steps or methods can be implemented by software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented by hardware, as in another embodiment, it can be implemented by any one of the following technologies known in the art or their combination: a discrete logic circuit having a logic gate circuit for implementing a logic function for a data signal, a dedicated integrated circuit having a suitable combination of logic gate circuits, a programmable gate array (PGA), a field programmable gate array (FPGA), etc.

[0140] A person skilled in the art may understand that all or part of the steps in the method for implementing the above-mentioned embodiment may be completed by instructing related hardware through a program, and the program may be stored in a computer-readable storage medium, which, when executed, includes one or a combination of the steps of the method embodiment.

[0141] In addition, each functional unit in each embodiment of the present application may be integrated into a processing module, or each unit may exist physically separately, or two or more units may be integrated into one module. The above-mentioned integrated module may be implemented in the form of hardware or in the form of a software functional module. If the integrated module is implemented in the form of a software functional module and sold or used as an independent product, it may also be stored in a computer-readable storage medium.

[0142] The storage medium mentioned above may be a read-only memory, a magnetic disk or an optical disk, etc. Although the embodiments of the present application have been shown and described above, it can be understood that the above embodiments are exemplary and cannot be understood as limiting the present application. A person of ordinary skill in the art may change, modify, replace and modify the above embodiments within the scope of the present application.

Claims

1. A method for generating a hypergraph structure of time series data, characterized in that: The following steps are involved: Obtain time series data of the target system; A joint probability distribution of the interaction of variables in the target system is obtained according to the time series data, and a hypergraph structure is obtained based on the sampling of the joint probability distribution, wherein each node in the hypergraph structure identifies a variable of the target system, and the hyperedges of the hypergraph structure represent the interaction structure of the variables, wherein the joint probability distribution of the interaction of variables in the target system is obtained according to the time series data, and the hypergraph structure is obtained based on the sampling of the joint probability distribution comprises: calculating the probability distribution of the interaction of variables in the target system according to the hidden state at the data moment; obtaining an association matrix based on the sampling of the probability distribution to represent the hypergraph structure; and The state of the target system at a previous moment of operation and the hypergraph structure are input into a preset hypergraph neural network to generate a time series data hypergraph structure to predict the operating state of the target system at the next moment.

2. The method according to claim 1, characterized in that The step of inputting the state of the target system at a previous moment and the hypergraph structure into a preset hypergraph neural network to generate a time series data hypergraph structure to predict the operating state of the target system at a next moment includes: Calculate the hyperedge weight diagonal matrix and node degree diagonal matrix of the hypergraph structure; Calculate the predicted value of the variable state at the next moment based on the association matrix, the state of at least one variable in the system at the current moment, the hyperedge weight diagonal matrix, the node degree diagonal matrix, the parameters of the first hypergraph convolution layer of the preset hypergraph neural network, the parameters of the second hypergraph convolution layer and the parameters of the fully connected output layer; Calculating a current prediction error according to the variable state prediction value and the actual observation value, and calculating a total loss function value according to the smoothness constraint penalty term of the hypergraph structure and the current prediction error; Back propagation calculation is performed according to the total loss function value to calculate the parameters of the variational autoencoder until the convergence condition is met.

3. The method according to claim 1, characterized in that The association matrix is: Among them, H ij is whether the i-th node interacts with the j-th node, ∈ v = -log(-log(s)), where s is randomly drawn from Uniform(0;1) and τ is a hyperparameter called temperature that controls the interpolation between the discrete distribution and the continuous categorical density.

4. A device for generating a hypergraph structure of time series data, characterized in that: include: The acquisition module is used to obtain the time series data of the target system; A collection module, used for obtaining a joint probability distribution of the interaction of variables in the target system according to the time series data, and obtaining a hypergraph structure based on the joint probability distribution sampling, wherein each node in the hypergraph structure identifies a variable of the target system, and the hyperedges of the hypergraph structure represent the interaction structure of the variables, wherein obtaining the joint probability distribution of the interaction of variables in the target system according to the time series data, and obtaining the hypergraph structure based on the joint probability distribution sampling comprises: calculating the probability distribution of the interaction of variables in the target system according to the hidden state at the data moment; obtaining an association matrix based on the probability distribution sampling to represent the hypergraph structure; and A generation module is used to input the state of the target system at a previous moment and the hypergraph structure into a preset hypergraph neural network to generate a time series data hypergraph structure to predict the operating state of the target system at the next moment.

5. An electronic device, characterized in that: include: A memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the method for generating a hypergraph structure of temporal data as described in any one of claims 1 to 3.

6. A computer-readable storage medium having a computer program stored thereon, characterized in that: The program is executed by a processor to implement the method for generating a hypergraph structure of temporal data as described in any one of claims 1 to 3.

Images