Time sequence prediction method based on time sequence dynamic graph and electronic equipment

Abstract

The invention provides a time sequence prediction method based on a time sequence dynamic graph and electronic equipment. The method comprises the following steps: inputting historical behavior sequence data of a target object into a long short-term memory (LSTM) network to obtain time sequence features; dynamic analysis data are collected, the incidence relation among the multiple objects is recognized according to the dynamic analysis data, an adjacency matrix of the discrete time dynamic graph is adjusted according to the incidence relation among the multiple objects, and the adjacency matrix is used for determining the weight of the incidence relation among the multiple objects; taking a target object as a node in the discrete time dynamic graph, calculating graph embedding momentum of the node according to historical behavior sequence data, and inputting a graph embedding vector and an adjacent matrix into a graph convolution GCN network to obtain a dynamic graph feature output by the GCN network through fusing a relationship between nodes, the graph embedding momentum is used for representing the change condition of the value of the node along with the time sequence interval; and generating a time sequence prediction result of the target object according to the time sequence features and the dynamic graph features.

Description

Technical Field

[0001] The present application mainly relates to the field of dynamic graph neural network technology, and in particular to a time series prediction method and electronic device based on a time series dynamic graph. Background Art

[0002] The traditional approach to time series forecasting is to fit historical time series data, extract feature information from multimodal data such as industry data, financial statements, and news reports, and finally output the desired future time series. However, this approach fails to consider how certain feature information (such as correlations) itself changes over time. This can lead to reduced forecast accuracy in volatile and unstable financial markets. Summary of the Invention

[0003] The present application provides a time series prediction method, electronic device and storage medium based on a time series dynamic graph, which effectively mines the association relationship between multiple objects based on a discrete-time dynamic graph. At the same time, it combines the advantages of an LSTM network and a GCN network to realize the timely extraction of dynamic graph features, thereby improving the prediction accuracy and facilitating risk prediction and management in the financial market.

[0004] In a first aspect, a time series prediction method based on a time series dynamic graph is provided, comprising:

[0005] Inputting the historical behavior sequence data of the target object into the long short-term memory (LSTM) network to obtain the time series features output by the LSTM network through time series encoding;

[0006] Collecting dynamic analysis data, identifying associations between a plurality of objects based on the dynamic analysis data, and adjusting an adjacency matrix of a discrete-time dynamic graph based on the associations between the plurality of objects, wherein the adjacency matrix is used to determine weights of the associations between the plurality of objects;

[0007] The target object is used as a node in the discrete-time dynamic graph, a graph embedding momentum of the node is calculated based on the historical behavior sequence data, and the graph embedding vector and the adjacency matrix are input into a graph convolutional neural network (GCN) to obtain dynamic graph features output by the GCN network by fusing the relationships between nodes, wherein the graph embedding momentum is used to characterize the change of the node value over the time series interval;

[0008] A time series prediction result of the target object is generated according to the time series features and the dynamic graph features.

[0009] In some embodiments, the training process of the LSTM network includes:

[0010] A sliding window method is used to construct a training set and a test set for training the LSTM network;

[0011] The LSTM network is trained based on the training set and the test set.

[0012] In some embodiments, the feature extraction process of the LSTM network includes:

[0013] According to the input vector x at time step t t and the hidden state vector h at the previous time step t-1 t-1 Calculate the forget gate activation vector f at time step t t , input gate activation vector i t , output gate activation vector o t and candidate cell states

[0014] According to the candidate cell state and the cell state C at the previous time step t-1 t-1 Calculate the cell state C at time step t t ;

[0015] According to the cell state C t and the output gate activation vector o t Calculate the current state vector h t , the current state vector h t The time series features as the output of the LSTM network.

[0016] In some embodiments, the dynamic analysis data includes analyst coverage data, where the analyst coverage data is used to represent the research report status of the same analyst on the multiple subjects at a target time point;

[0017] Identifying association relationships among multiple objects according to the dynamic analysis data, and adjusting an adjacency matrix of a discrete-time dynamic graph according to the association relationships among the multiple objects, including:

[0018] In response to the same analyst publishing research reports on at least two of the multiple objects at the target time point, it is confirmed that there is a correlation relationship between the at least two objects, and the weights corresponding to the at least two objects in the adjacency matrix are further increased by one.

[0019] In some embodiments, after adjusting the adjacency matrix of the discrete-time dynamic graph according to the association relationship between the multiple objects, the method further includes:

[0020] Collecting industrial production data of the target object, wherein the industrial production data is used to characterize the production and operation status of the target object;

[0021] The industrial production data is input into the LSTM network, and then the data output by the LSTM based on the industrial production data is input into the GCN network to obtain industrial features.

[0022] In some embodiments, the training process of the GCN network includes:

[0023] Using the time series features output by the LSTM network as initial input data of the GCN network;

[0024] The GCN network is trained according to the initial input data and a target loss function, wherein the GCN network is stacked by multiple layers of graph convolutional networks, and the target loss function adopts a mean square error loss function.

[0025] In some embodiments, the computation process of one layer of a graph convolutional network includes:

[0026]

[0027] Among them, H (l) is the node feature matrix of the lth layer, the initial layer H (0) =X, X is the time series feature output by the LSTM network, is the normalized adjacency matrix, I is the identity matrix, yes The degree matrix, W (l) is the weight matrix of the lth layer, and σ is the activation function.

[0028] In some embodiments, during the training process, the parameters of the GCN network are updated in the following manner:

[0029]

[0030] in, is the weight matrix of layer l at time step t, is the weight matrix of the lth layer at the previous time step t-1, and RNN(.) is the recurrent neural network RNN model.

[0031] In a second aspect, an electronic device is provided. The electronic device includes one or more processors; and one or more memories coupled to the one or more processors and storing instructions thereon, wherein when the one or more processors execute the instructions individually or collectively, the electronic device performs the steps of the time series prediction method based on the time series dynamic graph.

[0032] In a third aspect, a non-transitory computer-readable storage medium storing machine-executable instructions is provided. The machine-executable instructions, when executed by one or more processors of a machine, cause the machine to perform any one of the above methods.

[0033] Compared with the prior art, this application has the following advantages:

[0034] This application first uses the LSTM network to extract the time series features of the target object in the historical stage, and then effectively mines the correlation relationship between multiple objects based on the discrete time dynamic graph. It further combines the GCN network to output dynamic graph features by fusing the relationship between nodes. In this way, in the time series prediction process of the target object, richer and more timely feature information can be obtained, thereby improving the prediction accuracy and contributing to risk prediction and management in the financial market.

[0035] It should be understood that the invention summary is not intended to identify the key or essential features of the embodiments of the present disclosure, nor is it intended to limit the scope of the present disclosure. Other features of the present disclosure will become readily understood through the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] The accompanying drawings are included to provide a further understanding of the present application. They are incorporated into and constitute a part of this application. The accompanying drawings illustrate embodiments of the present application and, together with this specification, serve to explain the principles of the present application. In the accompanying drawings:

[0037] Figure 1 This is a flowchart of a time series prediction method based on a time series dynamic graph proposed by the present application;

[0038] Figure 2 Schematic diagram of a time series prediction model combining an LSTM network and a GCN network, exemplified in this application;

[0039] Figure 3 Schematic diagram of an electronic device exemplified in this application. DETAILED DESCRIPTION

[0040] The principle of the present disclosure will now be described with reference to some embodiments. It should be understood that the description of these embodiments is merely for illustrative purposes and helps those skilled in the art to understand and implement the present disclosure without placing any restriction on the scope of the present disclosure. The disclosure described herein can be implemented in a manner different from that described below.

[0041] In the following description and claims, unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this disclosure belongs.

[0042] References in this disclosure to "one embodiment," "an embodiment," "an exemplary embodiment," etc., indicate that the described embodiment may include a particular feature, structure, or characteristic, but not every embodiment necessarily includes the particular feature, structure, or characteristic. Moreover, such phrases do not necessarily refer to the same embodiment. In addition, when a particular feature, structure, or characteristic is described in conjunction with an exemplary embodiment, whether or not explicitly described, those skilled in the art will recognize that such feature, structure, or characteristic may be combined with other embodiments.

[0043] It should be understood that although the terms "first" and "second" and the like may be used herein to describe various elements, these elements should not be limited by these terms. These terms are merely used to distinguish one element from another. For example, a first element may be referred to as a second element, and similarly, a second element may be referred to as a first element without departing from the scope of the exemplary embodiments. The term "and / or" as used herein includes any and all combinations of one or more of the listed terms.

[0044] This application provides a time series prediction method based on a time series dynamic graph. This application first uses an LSTM network to extract the time series features of the target object in the historical stage, and then uses a discrete time dynamic graph to effectively mine the association relationship between multiple objects. It further combines the GCN network to output dynamic graph features by fusing the relationships between nodes. In this way, in the time series prediction process of the target object, richer and more timely feature information can be obtained, thereby improving the prediction accuracy and contributing to risk prediction and management in the financial market.

[0045] The time series prediction method based on time series dynamic graph uses a neural network model that combines LSTM network and GCN network. Among them:

[0046] LSTM (Long Short-Term Memory) is a special recurrent neural network (RNN) designed to address the vanishing and exploding gradient problems that standard RNNs face when processing long sequences of data. LSTM uses a gating mechanism to control the flow of information, enabling it to better capture and memorize long-term dependencies.

[0047] GCN (Graph Convolutional Network) is a deep learning model for processing graph data. It is primarily used for tasks such as node classification, link prediction, and graph classification. GCN learns node representations by aggregating features of nodes and their neighbors. Key to this is the introduction of a normalized adjacency matrix and the graph convolution operation.

[0048] Figure 1100 is a flowchart of a time series prediction method based on a time series dynamic graph proposed by the present application. Figure 1 , the method comprising:

[0049] S101, inputting the historical behavior sequence data of the target object into the long short-term memory (LSTM) network to obtain the time series features output by the LSTM network through time series encoding.

[0050] The LSTM network consists of four main components: the input gate, the forget gate, the candidate memory cell, and the output gate. These components manage the cell state by controlling the addition and removal of information.

[0051] In some embodiments, the feature extraction process of the LSTM network includes:

[0052] According to the input vector x at time step t t and the hidden state vector h at the previous time step t-1 t-1 Calculate the forget gate activation vector f at time step t t , input gate activation vector i t , output gate activation vector o t and candidate cell states

[0053] Based on the candidate cell state and the cell state C at the previous time step t-1 t-1 Calculate the cell state C at time step t t ;

[0054] According to the cell state C t and the output gate activation vector o t Calculate the current state vector h t , the current state vector h t Time series features as the output of the LSTM network.

[0055] In other words, the feature extraction process of the LSTM model can satisfy the following formulas (1) to (6):

[0056] f t =σ(W f ·[h t-1 ,x t ]+b f )(1)

[0057] i t =σ(W i ·[h t-1,x t ]+b i )(2)

[0058]

[0059] o t =σ(W o ·[h t-1 ,x t ]+b o )(5)

[0060] h t =o t ⊙tanh(C t )(6)

[0061] Among them, f t is the activation vector of the forget gate, i t is the activation vector of the input gate, is the candidate cell state, C t is the cell state, o t is the activation vector of the output gate, h t-1 is the hidden state vector, σ is the sigmoid function, tanh is the hyperbolic tangent function, W f ,W i ,W C ,W o is the weight matrix, b f ,b i ,b C ,b o is the bias vector, x t is the input vector at time step t, h t-1 is the hidden state vector from the previous time step, and ⊙ represents the element-wise multiplication.

[0062] During the training process of the LSTM network, a sliding window method can be used to construct a training set and a test set for training the LSTM network, and the LSTM network can be trained based on the training set and the test set.

[0063] The sliding window method divides the required training set and test set based on the historical behavior data of the financial market. Specifically, assuming that the initial time is t0, in the sliding window process, the historical interval T P =[t0,t0+t past ] is used to input the LSTM network, using the future time interval T F =[t0,t0+t past +w+t future ] behavior sequence data to verify the prediction results of the LSTM network. The former can be used as the training set as described above, and the latter as the corresponding test set.

[0064] S102, collecting dynamic analysis data, identifying association relationships between multiple objects based on the dynamic analysis data, and adjusting the adjacency matrix of the discrete-time dynamic graph based on the association relationships between the multiple objects, wherein the adjacency matrix is used to determine the weights of the association relationships between the multiple objects.

[0065] In some embodiments, the dynamic analysis data includes analyst coverage data, which is used to characterize the research report status of the same analyst on multiple objects at a target time point.

[0066] In the financial field, analyst coverage data can refer to research reports released by securities analysts on company stocks, or news reports, etc., with no restrictions.

[0067] In reality, multiple stocks that are reported by the same analyst at the same time often have certain correlations. Therefore, using analyst coverage data can identify the correlation between multiple objects and then adjust the adjacency matrix of the discrete-time dynamic graph.

[0068] Specifically, the steps of adjusting the adjacency matrix include:

[0069] In response to the same analyst publishing research reports on at least two objects among the multiple objects at the target time point, it is confirmed that there is a correlation relationship between the at least two objects, and the weights corresponding to the at least two objects in the adjacency matrix are further increased by one.

[0070] For example, suppose analyst a released research reports on stocks vi and vj in the fourth quarter of 2024. That is, stocks vi and vj are jointly covered by analyst a in this research report. Then, for the adjacency matrix A associated with the fourth quarter of 2024, the weight of the i-th row and j-th column is increased by 1, and the weight of the j-th row and i-th column is increased by 1.

[0071] It should be noted that, in order to ensure the timeliness of the update of the adjacency matrix, the update frequency of the adjacency matrix can be consistent with the time frequency of analysts releasing research reports or companies releasing performance reports, or it can be flexibly set according to actual conditions.

[0072] In some embodiments, after adjusting the adjacency matrix of the discrete-time dynamic graph according to the association relationship between the multiple objects, the method further includes:

[0073] Collecting industrial production data of the target object, where the industrial production data is used to characterize the production and operation status of the target object;

[0074] The industrial production data is input into the LSTM network, and then the data output by the LSTM based on the industrial production data is input into the GCN network to obtain industrial features.

[0075] In actual situations, analyzing industrial production data can help us understand the company's production and operation status, which can also help us more fully explore the characteristic information of the target object.

[0076] S103, treats the target object as a node in a discrete-time dynamic graph, calculates the graph embedding momentum of the node based on the historical behavior sequence data, and inputs the graph embedding vector and adjacency matrix into the graph convolutional neural network (GCN) to obtain the dynamic graph features output by the GCN network by fusing the relationships between nodes. The graph embedding momentum is used to characterize the changes in the value of the node over the time series interval.

[0077] For discrete-time dynamic graphs, the graph at time t is represented by G t ={V t ,E t ,X t}, where V is the node set, E is the relationship set between nodes, and X is the graph embedding vector of the stage. n ] is divided into n time intervals, and each time point t stores a snapshot of the graph data, forming a set

[0078] In this embodiment, the graph embedding vector can be the price momentum Δp t =p t -p t-1 , which represents the price on day t minus the price on day t-1.

[0079] In some embodiments, the training process of the GCN network includes:

[0080] The time series features output by the LSTM network are used as the initial input data of the GCN network;

[0081] The GCN network is trained based on the initial input data and the target loss function, where the GCN network is composed of a stack of multiple layers of graph convolutional networks, and the target loss function adopts the mean square error loss function.

[0082] The calculation process of one layer of graph convolutional network satisfies formula (7):

[0083]

[0084] Among them, H (l) is the node feature matrix of the lth layer, the initial layer H (0) =X, X is the time series feature output by the LSTM network, is the normalized adjacency matrix, I is the identity matrix, yes The degree matrix, W(l) is the weight matrix of the lth layer, and σ is the activation function.

[0085] During the training process, the following formula (8) is used to update the parameters of the GCN network:

[0086]

[0087] in, is the weight matrix of layer l at time step t, is the weight matrix of the lth layer at the previous time step t-1, and RNN(.) is the recurrent neural network RNN model.

[0088] Based on the above method, the GCN network can fuse the relationships between nodes based on the adjacency matrix and graph embedding vector, thereby outputting dynamic graph features and completing the prediction task based on discrete-time dynamic graphs.

[0089] S104: Generate a time series prediction result of the target object according to the time series features and the dynamic graph features.

[0090] Specifically, by combining the dynamic graph features output by the GCN network and the time series features output by the LSTM network, we can obtain the future time series of the target object, that is, the time series prediction result.

[0091] Figure 2 200 is a schematic diagram of a time series prediction model combining an LSTM network and a GCN network, as exemplified in this application. Figure 2 In this application, the input of the LSTM network is the historical behavior sequence data of the target object. The LSTM network encodes the historical behavior sequence data into a time series and outputs time series features. The input of the GCN network is the adjacency matrix of the discrete-time dynamic graph and the graph embedding momentum of the nodes, where the adjacency matrix comes from dynamic analysis data (such as analyst coverage data). The GCN network outputs dynamic graph features by fusing the relationships between nodes. Finally, the entire time series prediction model combines the dynamic graph features and the time series features to obtain the final time series prediction results.

[0092] In some embodiments, the reliability and authenticity of the time series prediction results can be evaluated using a loss function such as mean square error, where the mean square error satisfies the following formula (9):

[0093]

[0094] Among them, y i The true value of the time series at this moment, This is the time series prediction result at this moment, and n is the sequence length.

[0095] Compared with existing technologies, the above method uses discrete-time dynamic graphs as well as analyst coverage data and industrial production data, which can obtain richer and more timely feature information, thereby improving prediction accuracy and contributing to risk prediction management in the financial market.

[0096] In order to further verify the technical effect, Figure 2 The performance evaluation results of the time series forecasting model (abbreviated as LSTM-EGCN) shown in the figure are compared with those of a traditional LSTM model. The evaluation metrics include IC (information coefficient, which indicates the correlation between the factor value of the selected stock (i.e., the model prediction value) and the stock's next period return), Rank IC (rank information coefficient, which indicates the correlation between the factor value of the selected stock (i.e., the model prediction value) and the ranking of the stock's next period return), AV (annualized volatility), and SR (Sharpe ratio, which indicates the excess return per unit of risk). The comparison results are shown in Table 1 below:

[0097] Table 1

[0098]

[0099] It can be seen that compared with the traditional LSTM model, the above time series prediction model (abbreviated as LSTM-EGCN) has shown better results in all evaluation indicators. Therefore, the LSTM-EGCN can improve the prediction accuracy and contribute to the risk prediction management of the financial market.

[0100] Further, if Figure 3 An exemplary embodiment of the present application provides a schematic diagram 300 of an electronic device, comprising one or more memories 301 and one or more processors 302, wherein the one or more memories 301 are coupled to the one or more processors 302 and store instructions thereon, and the instructions can be executed individually or collectively by the one or more processors 302, so that the electronic device performs any method as in the first aspect.

[0101] It should be understood that the processor mentioned in the embodiments of the present application may be a CPU, or may be other general-purpose processors, DSPs, ASICs, FPGAs or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or any conventional processor.

[0102] It should also be understood that the memory mentioned in the embodiments of the present application may be a volatile memory or a non-volatile memory, or may include both volatile and non-volatile memories. Among them, the non-volatile memory may be a read-only memory (ROM), a programmable read-only memory, an erasable programmable read-only memory, an electrically erasable programmable read-only memory, or a flash memory. The volatile memory may be a random access memory (RAM), which is used as an external cache. By way of example but not limitation, many forms of RAM are available, such as static random access memory, dynamic random access memory, synchronous dynamic random access memory, double data rate synchronous dynamic random access memory, enhanced synchronous dynamic random access memory, synchronously linked dynamic random access memory, and direct memory bus random access memory.

[0103] The present application also provides a non-transitory computer-readable storage medium storing machine-executable instructions, wherein the computer-executable instructions can be executed by one or more processors of a machine. The machine may include the electronic device mentioned above, etc. When the computer-executable instructions are executed by the one or more processors, the machine performs any of the methods mentioned above.

[0104] A computer-readable storage medium may include a propagated data signal embodying computer program code, for example, in baseband or as part of a carrier wave. The propagated signal may be in a variety of forms, including electromagnetic, optical, etc., or a suitable combination thereof. The computer-readable storage medium may be connected to an instruction execution system, device, or apparatus to communicate, propagate, or transmit the program for use. The program code on the computer-readable storage medium may be transmitted via any suitable medium, including radio, cable, fiber optic cable, radio frequency signal, or similar medium, or any combination of the above.

[0105] The basic concepts have been described above. It will be apparent to those skilled in the art that the above disclosures are merely illustrative and do not constitute limitations on this application. Although not explicitly stated herein, those skilled in the art may make various modifications, improvements, and amendments to this application. Such modifications, improvements, and amendments are suggested in this application and remain within the spirit and scope of the exemplary embodiments of this application.

[0106] At the same time, this application uses specific terms to describe the embodiments of this application. For example, "one embodiment," "an embodiment," and / or "some embodiments" refer to a certain feature, structure, or characteristic related to at least one embodiment of this application. Therefore, it should be emphasized and noted that "one embodiment," "an embodiment," or "an alternative embodiment" mentioned twice or multiple times in different locations in this specification does not necessarily refer to the same embodiment. In addition, certain features, structures, or characteristics in one or more embodiments of this application may be appropriately combined.

[0107] Some aspects of the present application can be performed entirely by hardware, entirely by software (including firmware, resident software, microcode, etc.), or by a combination of hardware and software. The above hardware or software can be referred to as "data blocks", "modules", "engines", "units", "components" or "systems". The processor can be one or more application-specific integrated circuits (ASICs), digital signal processors (DSPs), digital signal processing devices (DAPDs), programmable logic devices (PLDs), field programmable gate arrays (FPGAs), processors, controllers, microcontrollers, microprocessors or combinations thereof. In addition, various aspects of the present application may be expressed as computer products located in one or more computer-readable media, which include computer-readable program code. For example, computer-readable media may include, but are not limited to, magnetic storage devices (e.g., hard disks, floppy disks, tapes...), optical disks (e.g., compact disks CDs, digital versatile disks DVDs...), smart cards, and flash memory devices (e.g., cards, sticks, key drives...).

[0108] A computer-readable medium may include a propagated data signal embodying computer program code, for example, in baseband or as part of a carrier wave. The propagated signal may be in a variety of forms, including electromagnetic, optical, etc., or a suitable combination thereof. A computer-readable medium may be any computer-readable medium other than a computer-readable storage medium that can be connected to an instruction execution system, apparatus, or device to communicate, propagate, or transmit the program for use. The program code on the computer-readable medium may be transmitted via any suitable medium, including radio, cable, fiber optic cable, radio frequency signal, or similar medium, or any combination of the above.

[0109] Similarly, it should be noted that, in order to simplify the presentation of this application and thus facilitate understanding of one or more embodiments of the invention, the foregoing descriptions of the embodiments of this application sometimes combine multiple features into a single embodiment, figure, or description thereof. However, this disclosure method does not mean that the subject matter of this application requires more features than those recited in the claims. In fact, an embodiment may have fewer features than all of the features of a single embodiment disclosed above.

[0110] In some embodiments, numbers are used to describe the quantity of components and attributes. It should be understood that such numbers used in the description of the embodiments are modified by the modifiers "about", "approximately" or "substantially" in some examples. Unless otherwise stated, "about", "approximately" or "substantially" indicate that the numbers are allowed to vary by ±20%. Accordingly, in some embodiments, the numerical parameters used in the description and claims are approximate values, which may change according to the required features of individual embodiments. In some embodiments, the numerical parameters should take into account the specified significant digits and adopt the general method of retaining digits. Although the numerical domains and parameters used to confirm the breadth of their range in some embodiments of the present application are approximate values, in specific embodiments, the settings of such numerical values are as accurate as possible within the feasible range.

[0111] Although the present application has been described with reference to the current specific embodiments, ordinary technicians in this technical field should recognize that the above embodiments are only used to illustrate the present application, and various equivalent changes or substitutions can be made without departing from the spirit of the present application. Therefore, as long as the changes and modifications to the above embodiments are within the scope of the essential spirit of the present application, they will fall within the scope of the claims of the present application.

Claims

1. A time series prediction method based on a time series dynamic graph, characterized in that: include: Inputting the historical behavior sequence data of the target object into the long short-term memory (LSTM) network to obtain the time series features output by the LSTM network through time series encoding; Collecting dynamic analysis data, identifying associations between a plurality of objects based on the dynamic analysis data, and adjusting an adjacency matrix of a discrete-time dynamic graph based on the associations between the plurality of objects, wherein the adjacency matrix is used to determine weights of the associations between the plurality of objects; The target object is used as a node in the discrete-time dynamic graph, a graph embedding momentum of the node is calculated based on the historical behavior sequence data, and the graph embedding vector and the adjacency matrix are input into a graph convolutional neural network (GCN) to obtain dynamic graph features output by the GCN network by fusing the relationships between nodes, wherein the graph embedding momentum is used to characterize the change of the node value over the time series interval; A time series prediction result of the target object is generated according to the time series features and the dynamic graph features.

2. The method according to claim 1, wherein The training process of the LSTM network includes: A sliding window method is used to construct a training set and a test set for training the LSTM network; The LSTM network is trained based on the training set and the test set.

3. The method according to claim 1, wherein The feature extraction process of the LSTM network includes: According to the input vector x at time step t t and the hidden state vector h at the previous time step t-1 t-1 Calculate the forget gate activation vector f at time step t t , input gate activation vector i t , output gate activation vector o t and candidate cell states According to the candidate cell state and the cell state C at the previous time step t-1 t-1 Calculate the cell state C at time step t t ; According to the cell state C t and the output gate activation vector o t Calculate the current state vector h t , the current state vector h t The time series features as the output of the LSTM network.

4. The method according to claim 1, wherein The dynamic analysis data includes analyst coverage data, which is used to represent the research report status of the same analyst on the multiple objects at the target time point; Identifying association relationships among multiple objects according to the dynamic analysis data, and adjusting an adjacency matrix of a discrete-time dynamic graph according to the association relationships among the multiple objects, including: In response to the same analyst publishing research reports on at least two of the multiple objects at the target time point, it is confirmed that there is a correlation relationship between the at least two objects, and the weights corresponding to the at least two objects in the adjacency matrix are further increased by one.

5. The method according to claim 1, wherein After adjusting the adjacency matrix of the discrete-time dynamic graph according to the association relationship between the multiple objects, the method further includes: Collecting industrial production data of the target object, wherein the industrial production data is used to characterize the production and operation status of the target object; The industrial production data is input into the LSTM network, and then the data output by the LSTM based on the industrial production data is input into the GCN network to obtain industrial features.

6. The method according to claim 1, wherein The training process of the GCN network includes: Using the time series features output by the LSTM network as initial input data of the GCN network; The GCN network is trained according to the initial input data and a target loss function, wherein the GCN network is stacked by multiple layers of graph convolutional networks, and the target loss function adopts a mean square error loss function.

7. The method according to claim 6, wherein The computational process of one layer of graph convolutional network includes: Among them, H (l) is the node feature matrix of the lth layer, the initial layer H (0) =X, X is the time series feature output by the LSTM network, is the normalized adjacency matrix, I is the identity matrix, yes The degree matrix, W (l) is the weight matrix of the lth layer, and σ is the activation function.

8. The method according to claim 7, wherein During the training process, the parameters of the GCN network are updated in the following way: in, is the weight matrix of layer l at time step t, is the weight matrix of the lth layer at the previous time step t-1, and RNN(.) is the recurrent neural network RNN model.

9. An electronic device comprising: one or more processors; as well as One or more memories coupled to the one or more processors and storing thereon instructions, which, when executed individually or collectively by the one or more processors, cause the electronic device to perform the method according to any one of claims 1 to 8.

10. A non-transitory computer-readable storage medium storing machine-executable instructions, which, when executed by one or more processors of a machine, cause the machine to perform the method of any one of claims 1-8.

Images