Data analysis and prediction methods, devices, servers, storage media, and program products

Abstract

The present application relates to the field of artificial intelligence technology, and particularly to a data analysis and prediction method, device, server, storage medium, and program product. The method includes: obtaining graph structure data of the historical time series data of each variable according to the historical time series data of multiple variables of a target entity; inputting the graph structure data into a preset graph convolutional neural network model, and after aggregating the graph structure data in the time dimension through each convolutional layer in the graph convolutional neural network model, obtaining predicted time series data of multiple variables; the number of convolutional layers in the graph convolutional neural network model is determined based on the number of sampling moments in the time dimension. Using this method can improve the prediction accuracy of time series.

Description

Technical Field

[0001] This application relates to the field of artificial intelligence technology, and particularly to a data analysis and prediction method, apparatus, server, storage medium, and program product. Background Art

[0002] With the continuous development of data collection technology, rich dynamic data has been collected in the financial and other fields, such as stock data, personal customer asset data, program operation index data, etc. These data are usually modeled as time series data. Analyzing and predicting time series data can assist enterprises in achieving precise marketing and financial risk prediction, helping to avoid business risks and financial risks, etc.

[0003] In related technologies, time series data prediction algorithms include prediction algorithms based on statistical methods and prediction algorithms based on neural networks. Among them, the prediction algorithm based on statistical methods can be the autoregressive integrated moving average algorithm. After using the difference operator to eliminate the local level or trend of non-stationary time series data, assuming that there is similarity between parts of the time series data, an existing model is then selected to predict the time series data. The prediction algorithm based on neural networks can be a multivariate time series data prediction model based on a dual-window mechanism, which uses two neural network windows to extract the short-term stable sequence features and periodic and seasonal long-term sequence features in the time series data respectively, and at the same time uses the aggregation of the two features to predict the time series data.

[0004] However, the methods in related technologies have low prediction accuracy for time series. Summary of the Invention

[0005] Based on this, it is necessary to provide a data analysis and prediction method, apparatus, server, storage medium, and program product that can improve the prediction accuracy of time series for the above technical problems.

[0006] In a first aspect, this application provides a data analysis and prediction method, and the method includes:

[0007] Obtain the graph structure data of the historical time series data of each variable according to the historical time series data of multiple variables of the target entity;

[0008] Input the graph structure data into a preset graph convolutional neural network model, and after aggregating the graph structure data in the time dimension through each convolutional layer in the graph convolutional neural network model, obtain the predicted time series data of multiple variables; the number of convolutional layers in the graph convolutional neural network model is determined based on the number of sampling moments in the time dimension.

[0009] In one embodiment, according to the historical time series data of multiple variables of the target subject, graph structure data of the historical time series data of each variable is obtained, including:

[0010] All variable nodes in the historical time series data of each variable are obtained;

[0011] Edge connection processing is performed on each variable node to obtain graph structure data of the historical time series data of each variable.

[0012] In one embodiment, edge connection processing is performed on each variable node to obtain graph structure data of the historical time series data of each variable, including:

[0013] Edge connection operations are performed on each variable node a preset number of times to obtain graph structure data of the historical time series data of each variable;

[0014] Among them, the edge connection operation includes:

[0015] Each variable node is randomly grouped to obtain multiple variable node sets;

[0016] The similarity of variable nodes in each variable node set is obtained;

[0017] Variable nodes corresponding to similarities greater than a preset threshold in each variable node set are connected.

[0018] In one embodiment, if the number of convolutional layers and the number of sampling times are both N, where N is a positive integer;

[0019] Then, the graph structure data is aggregated in the time dimension through each convolutional layer in the graph convolutional neural network model to obtain the predicted time series data of multiple variables, including:

[0020] The features of each sampling time in the first convolutional layer of the graph structure data are obtained;

[0021] The features of each sampling time in the first convolutional layer are input into the second convolutional layer, and the features of N sampling times and the features of N - 1 sampling times in the first convolutional layer are aggregated to form the features of N sampling times in the second convolutional layer, so as to obtain the features of each sampling time in the second convolutional layer;

[0022] The features of each sampling time in the second convolutional layer are input into the third convolutional layer, and the features of N sampling times, the features of N - 1 sampling times, and the features of N - 2 sampling times in the second convolutional layer are aggregated to form the features of N sampling times in the third convolutional layer, so as to obtain the features of each sampling time in the third convolutional layer;

[0023] By analogy, the features at each sampling moment in the Nth convolutional layer are obtained, and the feature at the last sampling moment in the Nth convolutional layer is determined as the multi-variable aggregated feature;

[0024] Based on the multi-variable aggregated feature, multi-variable predicted time series data is generated.

[0025] In one embodiment, the preset graph convolutional neural network model further includes a prediction layer; then generating multi-variable predicted time series data according to the multi-variable aggregated feature includes:

[0026] Inputting the multi-variable aggregated feature into the prediction layer, and analyzing and predicting the multi-variable aggregated feature through the prediction layer to obtain multi-variable predicted time series data.

[0027] In one embodiment, the method further includes:

[0028] Obtaining the real time series data corresponding to the predicted time series data of each variable;

[0029] Updating the model parameters in the graph convolutional neural network model according to the difference between the predicted time series data of each variable and the corresponding real time series data.

[0030] In one embodiment, the construction process of the graph convolutional neural network model includes:

[0031] Obtaining the historical time series sample data of multiple sample variables;

[0032] According to the historical time series sample data of each sample variable, obtaining the sample graph structure data corresponding to the historical time series sample data of each sample variable;

[0033] Training the initial graph convolutional neural network model through the sample graph structure data until the preset convergence condition is met, determining that the graph convolutional neural network model converges, and obtaining the preset graph convolutional neural network model.

[0034] In a second aspect, the present application also provides a data analysis and prediction device, and the device includes:

[0035] A first acquisition module, configured to obtain the graph structure data of the historical time series data of each multi-variable according to the historical time series data of multiple variables of the target entity;

[0036] A first determination module, configured to input the graph structure data into the preset graph convolutional neural network model, and after aggregating the graph structure data in the time dimension through each convolutional layer in the graph convolutional neural network model, obtaining multi-variable predicted time series data; the number of convolutional layers in the graph convolutional neural network model is determined based on the number of sampling moments in the time dimension.

[0037] In a third aspect, the present application also provides a server, which includes a memory and a processor. The memory stores a computer program, and when the processor executes the computer program, all the contents in the above method embodiments are implemented.

[0038] In a fourth aspect, the present application also provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, all the contents in the above method embodiments are implemented.

[0039] In a fifth aspect, the present application also provides a computer program product, which includes a computer program. When the computer program is executed by a processor, all the contents in the above method embodiments are implemented.

[0040] For the above data analysis and prediction method, device, server, storage medium, and program product, the method obtains the graph structure data of the historical time series data of each variable according to the historical time series data of multiple variables of the target entity, inputs the graph structure data into a preset graph convolutional neural network model, and after aggregating the graph structure data in the time dimension through each convolutional layer in the graph convolutional neural network model, obtains the predicted time series data of multiple variables. The number of convolutional layers in the graph convolutional neural network model in this method is determined based on the number of sampling moments in the time dimension. For different historical time series data of multiple variables, different graph convolutional neural network models can be selected for prediction, making the prediction process of the historical time series data of each variable more flexible; at the same time, by predicting the historical time series data of multiple variables, compared with only predicting the historical time series data of a single variable, the prediction result is more accurate; through the convolutional layer in the graph convolutional neural network, the historical time series data of each variable can be aggregated, and through the aggregated historical time series data of each variable, the time series data of multiple variables in a future period can be accurately predicted. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] Figure 1 It is an application environment diagram of the data analysis and prediction method in an embodiment;

[0042] Figure 2 It is a flowchart of the data analysis and prediction method in an embodiment;

[0043] Figure 3 It is a flowchart of the data analysis and prediction method in an embodiment;

[0044] Figure 4 It is a flowchart of the data analysis and prediction method in an embodiment;

[0045] Figure 5 It is a flowchart of the data analysis and prediction method in an embodiment;

[0046] Figure 6 It is a schematic flow chart of a data analysis and prediction method in an embodiment;

[0047] Figure 7 It is a schematic flow chart of a data analysis and prediction method in an embodiment;

[0048] Figure 8 It is a schematic flow chart of a data analysis and prediction method in an embodiment;

[0049] Figure 9 It is a schematic flow chart of a data analysis and prediction method in an embodiment;

[0050] Figure 10 It is a structural block diagram of a data analysis and prediction device in an embodiment. Detailed implementation manners

[0051] In order to make the objectives, technical solutions and advantages of the present application clearer, the present application will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application, and are not used to limit the present application.

[0052] It should be noted that the data analysis and prediction method, device, server, storage medium and program product of the present disclosure can be applied in the field of artificial intelligence technology, and can also be used in other technical fields except artificial intelligence technology. The present disclosure does not limit the application fields of the data analysis and prediction method, device, server, storage medium and program product.

[0053] First, before specifically introducing the technical solutions of the embodiments of the present application, the technical background on which the embodiments of the present application are based will be introduced.

[0054] In real life, most time series data are jointly affected by multiple variables, and there are also mutual connections between multiple variables. By analyzing multi-variable time series data, multi-variable predicted time series data for a certain period in the future can be predicted. This process can assist enterprises in achieving precise marketing and financial risk prediction, etc., and help avoid business risks and financial risks, etc. For example, when the user's deposit amount is used as time series data, the deposit interest rate, the fund purchase amount, and the fund historical return rate can be used as multiple variables affecting the user's deposit amount.

[0055] However, the current prediction method only analyzes single-variable time series data to predict the predicted time series data for a certain period in the future. This process ignores the interaction between different variables at the same time, as well as the non-linear interaction between multiple variables at different times, and cannot extract the dynamic dependence and periodic dependence between multiple variables, resulting in low prediction accuracy.

[0056] The data analysis and prediction method provided by the embodiment of the present application can be applied to, for example, Figure 1 the application environment shown. The server includes a processor, a memory, and a network interface connected through a system bus. Among them, the processor of the server is used to provide computing and control capabilities. The memory of the server includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program, and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the server is used to store data analysis and prediction data. The network interface of the server is used to communicate with an external terminal through a network connection. Among them, the server can be implemented by an independent server or a server cluster composed of multiple servers.

[0057] In one embodiment, as Figure 2 shown, a data analysis and prediction method is provided. Taking the server in Figure 1 as an example, the method includes the following steps:

[0058] S201, obtain the graph structure data of the historical time series data of each variable according to the historical time series data of multiple variables of the target entity.

[0059] Among them, the time series data of multiple variables refers to data that changes continuously over time. Multiple variables mean that there are multiple variables simultaneously in the same system, and the values of multiple variables change continuously over time. For example, multiple variables can include the user's deposit amount and deposit interest rate, etc. The time series data of multiple variables can be the user's deposit amount, the price of stocks or futures, etc. The above graph structure data refers to the graph structure data generated by modeling with variables as nodes and the relationships between nodes as edges. For example, when generating corresponding graph structure data through social network data modeling, users are used as nodes, and interactions between users such as likes, forwards, and comments are used as edges between nodes.

[0060] Optionally, the server can obtain the historical time series data of multiple variables consistent with the electronic tag of the target entity in the corresponding database, or the server can also obtain the historical time series data of multiple variables corresponding to the keyword information in the corresponding database according to the keyword information of the target entity. This embodiment does not limit the method of obtaining the historical time series data of multiple variables of the target entity. Further, after the server obtains the historical time series data of multiple variables, it can determine the graph structure data of the historical time series data of each variable according to the dependency relationship between variables and / or the similarity between variables.

[0061] S202. Input the graph-structured data into a preset graph convolutional neural network model. After aggregating the graph-structured data in the time dimension through each convolutional layer in the graph convolutional neural network model, multivariate predicted time series data is obtained. The number of convolutional layers in the graph convolutional neural network model is determined based on the number of sampling moments in the time dimension.

[0062] Among them, the graph convolutional neural network is a new neural network algorithm for graph-structured data. Through the graph convolutional neural network model, the topological structure features between nodes and the respective attribute features of nodes in the graph-structured data are obtained. Traditional neural networks only consider the features of nodes and ignore the structural relationships between nodes in the graph-structured data, while graph neural networks can utilize the structural relationships to obtain more accurate features in the analysis of graph-structured data.

[0063] Specifically, the preset graph convolutional neural network model is trained with a large amount of sample data, and the accuracy of the prediction results output by the preset graph convolutional neural network model is relatively high. The server takes the graph-structured data in the time dimension as the input of the preset graph convolutional neural network model, aggregates all the graph-structured data in the time dimension at the last sampling moment through the convolutional layers in the graph convolutional neural network model, and predicts the time series data of multiple variables in a future period of time based on the aggregated data at the last sampling moment, and takes this data as the multivariate predicted time series data. Among them, the number of convolutional layers can be equal to the number of sampling moments in the time dimension, or can have a linear relationship with the number of sampling moments in the time dimension.

[0064] In the above data analysis and prediction method, the method obtains the graph-structured data of the historical time series data of each variable according to the historical time series data of multiple variables of the target entity, inputs the graph-structured data into a preset graph convolutional neural network model, and after aggregating the graph-structured data in the time dimension through each convolutional layer in the graph convolutional neural network model, multivariate predicted time series data is obtained. The number of convolutional layers in the graph convolutional neural network model in this method is determined based on the number of sampling moments in the time dimension. For different historical time series data of multiple variables, different graph convolutional neural network models can be selected for prediction, making the prediction process of the historical time series data of each variable more flexible; at the same time, by predicting the historical time series data of multiple variables, compared with only predicting the historical time series data of a single variable, the prediction results are more accurate; through the convolutional layers in the graph convolutional neural network, the historical time series data of each variable can be aggregated, and based on the aggregated historical time series data of each variable, the time series data of multiple variables in a future period of time can be accurately predicted.

[0065] Figure 3It is a schematic flowchart of the data analysis and prediction method provided by the embodiments of the present application. The embodiments of the present application relate to an optional implementation manner of obtaining the graph structure data of the historical time series data of each variable according to the historical time series data of multiple variables of a target entity. In Figure 2 On the basis of the embodiment shown, as Figure 3 shown, the above S201 may include the following steps:

[0066] S301, obtain all variable nodes in the historical time series data of each variable.

[0067] Specifically, the server may use each variable affecting the historical time series data as a variable node to obtain all variable nodes. For example, the historical time series data of a computer is affected by factors such as computer configuration, computer usage years, the occupancy of the central processing unit (CPU) in the computer, memory occupancy, and response time. The computer configuration, computer usage years, the occupancy of the central processing unit (CPU) in the computer, memory occupancy, and response time, etc. are used as all variable nodes in the historical time series data.

[0068] S302, perform edge connection processing on each variable node to obtain the graph structure data of the historical time series data of each variable.

[0069] Optionally, the larger the quantification value of the dependency relationship between two variables, the greater the relationship between the two variables, and edge connection operation needs to be performed between the two variables. The server may call a variable dependency relationship extractor to extract the quantification values of the dependency relationships between multiple variables, and determine the graph structure data corresponding to the historical time series data of the multiple variables according to the quantification values of the dependency relationships between the multiple variables. Optionally, the greater the similarity between two variables, the greater the relationship between the two variables, and edge connection operation needs to be performed between the two variables. The server may calculate the similarity between each variable. If the similarity between variables is greater than a preset similarity threshold, edge connection operation is performed on the variable. If the similarity between variables is less than the preset similarity threshold, no edge connection operation is performed on the variable. The graph structure data corresponding to the historical time series data of the multiple variables is determined according to the edge connection results of the variables. This embodiment does not make a limitation on this.

[0070] Figure 4 It is a schematic flowchart of the data analysis and prediction method provided by the embodiments of the present application. The embodiments of the present application relate to an optional implementation manner of performing edge connection processing on each variable node to obtain the graph structure data of the historical time series data of each variable. On the basis of the embodiment shown in Figure 3 as Figure 4As shown, the above S302 may include the following steps:

[0071] Perform edge connection operations on each variable node a preset number of times to obtain the graph structure data of the historical time series data of each variable;

[0072] Specifically, directly calculating the similarity relationship between each pair of nodes will result in the time complexity of the algorithm being O(N) 2 , in order to reduce the time complexity, the server can randomly divide multiple variable nodes into g groups, calculate the similarity between variable nodes within each group using the cosine similarity algorithm, and connect the k node pairs with the largest similarity, thereby reducing the time complexity of the graph construction algorithm to Repeating the process of random grouping and calculating similarity m times can obtain more accurate graph structure data, which can also be called an adjacency matrix.

[0073] Among them, the edge connection operation includes:

[0074] S401, randomly group each variable node to obtain multiple variable node sets.

[0075] Specifically, the server can randomly group each variable node through a random grouping algorithm, and determine the variable nodes in each group as multiple variable node sets. Among them, the random grouping algorithm can be the Krap algorithm, etc. For example, there are N variable nodes in the historical time series data, and the N variable nodes are randomly divided into m groups. The number of variable nodes in each group may be the same or different.

[0076] S402, obtain the similarity of variable nodes in each variable node set.

[0077] Specifically, the server can calculate the similarity of variable nodes in each variable node set through the cosine similarity algorithm. Among them, the cosine similarity measures the similarity between two vectors by measuring the cosine value of the angle between them. The cosine value of a 0-degree angle is 1, and the cosine value of any other angle is not greater than 1; and its minimum value is -1. Since the cosine similarity is usually used in the positive space, the value given is between -1 and 1. For example, the greater the similarity between variable 1 and variable 2, the closer the cosine value is to 1.

[0078] S403, connect the variable nodes corresponding to the similarity greater than the preset threshold in each variable node set.

[0079] Specifically, the server can compare the similarity of variable nodes in each set of variable nodes with a preset threshold. When the similarity of variable nodes in each set of variable nodes is greater than the preset threshold, the variable nodes are connected; when the similarity of variable nodes in each set of variable nodes is less than or equal to the preset threshold, there is no need to connect between the variable nodes. For example, when the preset threshold is 0.3 and the similarity between variable 1 and variable 2 is 0.5, the similarity between variable 1 and variable 2 is greater than the preset threshold, and variable 1 and variable 2 are connected.

[0080] In the above data analysis and prediction method, all variable nodes in the historical time series data of each variable are obtained, and a preset number of edge connection operations are performed on each variable node to obtain the graph structure data of the historical time series data of each variable. The edge connection operation in this method includes randomly grouping each variable node to obtain multiple sets of variable nodes, obtaining the similarity of variable nodes in each set of variable nodes, and connecting the variable nodes corresponding to the similarity greater than the preset threshold in each set of variable nodes. This method can reduce the time complexity in the operation process through the random grouping process, and can more accurately determine whether to connect between each variable by comparing the similarity of variable nodes with the preset threshold, making the obtained graph structure data more accurate.

[0081] Figure 5 It is a schematic flowchart of the data analysis and prediction method provided by the embodiments of the present application. The embodiments of the present application relate to an optional implementation manner of aggregating graph structure data in the time dimension through each convolutional layer in a graph convolutional neural network model to obtain multi-variable predicted time series data. Figure 2 On the basis of the embodiment shown, as Figure 5 shown, the above S202 may include the following steps:

[0082] S501, obtain the features of each sampling moment in the first convolutional layer of the graph structure data.

[0083] Specifically, the server inputs the graph structure data corresponding to each sampling moment into the graph convolutional neural network model, extracts the features in the graph structure data corresponding to each sampling moment through the first convolutional layer, and outputs the features of each sampling moment.

[0084] S502, input the features of each sampling moment in the first convolutional layer into the second convolutional layer, and aggregate the features of the Nth sampling moment and the (N - 1)th sampling moment in the first convolutional layer to form the features of the Nth sampling moment in the second convolutional layer, and obtain the features of each sampling moment in the second convolutional layer.

[0085] Specifically, the features at each sampling moment output by the first convolutional layer are used as the input information for the second convolutional layer, and the features at the sampling moments of two adjacent moments in the first convolutional layer are aggregated to obtain the features at each sampling moment in the second convolutional layer. For example, the features at the first sampling moment and the second sampling moment are aggregated, the features at the second sampling moment and the third sampling moment are aggregated, and the features at the (N - 1)-th sampling moment and the N-th sampling moment are aggregated.

[0086] S503. Input the features at each sampling moment in the second convolutional layer into the third convolutional layer, and aggregate the features at the N-th sampling moment, the (N - 1)-th sampling moment, and the (N - 2)-th sampling moment in the second convolutional layer to form the features at the N-th sampling moment in the third convolutional layer, thereby obtaining the features at each sampling moment in the third convolutional layer.

[0087] Specifically, the features at each sampling moment output by the second convolutional layer are used as the input information for the third convolutional layer, and the features at the sampling moments of two adjacent moments in the second convolutional layer are aggregated to obtain the features at each sampling moment in the third convolutional layer. For example, the features at the first sampling moment, the second sampling moment, and the third sampling moment are aggregated, the features at the second sampling moment, the third sampling moment, and the fourth sampling moment are aggregated, and the features at the (N - 1)-th sampling moment, the N-th sampling moment, and the (N - 2)-th sampling moment are aggregated.

[0088] S504. And so on, obtain the features at each sampling moment in the N-th convolutional layer, and determine the feature at the last sampling moment in the N-th convolutional layer as the multi-variable aggregated feature.

[0089] Specifically, the features at each sampling moment output by the (N - 1)-th convolutional layer are used as the input information for the N-th convolutional layer, and the features at the sampling moments of two adjacent moments in the (N - 1)-th convolutional layer are aggregated to obtain the features at each sampling moment in the N-th convolutional layer. At this time, all the features are aggregated at the N-th sampling moment, and the feature at the N-th sampling moment is used as the multi-variable aggregated feature. The process of each convolutional layer aggregating the features of the graph structure data can be expressed as:

[0090]

[0091]

[0092]

[0093]

[0094]

[0095] Among them, is the feature of the variable node in the first convolutional layer at the t-th moment; X t is the initial feature of the variable node; W is the trainable weight matrix of each convolutional layer in the graph convolutional neural network model; (·) -1 is the time shift operation, which shifts the latent feature vector of the previous moment to the current moment; dropout(·) randomly discards the learned latent features with a certain probability to avoid overfitting when the graph convolutional neural network model is learning; is A t corresponding degree matrix.

[0096] S505, generate the predicted time series data of multiple variables according to the multi-variable aggregated features.

[0097] Specifically, the server inputs the multi-variable aggregated features into the prediction layer, and the prediction layer predicts the historical time series data of multiple variables to predict the time series data of multiple variables at a certain future moment or a certain time period, and determines this data as the predicted time series data of multiple variables.

[0098] In the above data analysis and prediction method, the features of each sampling moment in the first convolutional layer of the graph structure data are obtained, the features of each sampling moment in the first convolutional layer are input into the second convolutional layer, and the features of the N sampling moment and the N-1 sampling moment in the first convolutional layer are aggregated to form the features of the N sampling moment in the second convolutional layer, and the features of each sampling moment in the second convolutional layer are obtained. The features of each sampling moment in the second convolutional layer are input into the third convolutional layer, and the features of the N sampling moment, the N-1 sampling moment, and the N-2 sampling moment in the second convolutional layer are aggregated to form the features of the N sampling moment in the third convolutional layer, and the features of each sampling moment in the third convolutional layer are obtained, and so on, until the features of each sampling moment in the Nth convolutional layer are obtained, and the feature of the last sampling moment in the Nth convolutional layer is determined as the multi-variable aggregated feature. According to the multi-variable aggregated feature, the predicted time series data of multiple variables is generated. In this method, through the graph convolutional neural network model, not only the interaction relationship between multiple variables at the same sampling moment can be obtained, but also the dynamic changes between multiple variables at different moments can be obtained. By aggregating the variable features at different moments into the last sampling moment through each convolutional layer, the calculation amount of generating the predicted time series data of multiple variables is relatively small.

[0099] In another embodiment, the embodiment of the present application relates to an optional implementation manner of generating the predicted time series data of multiple variables according to the multi-variable aggregated features. In Figure 5Based on the illustrated embodiment, the above process may further include the following steps: inputting the multi-variable aggregated features into the prediction layer, and analyzing and predicting the multi-variable aggregated features through the prediction layer to obtain the predicted time series data of the multi-variables.

[0100] Specifically, the prediction layer may be a Multilayer Perceptron (MLP). Input the multi-variable aggregated features into the multi-layer perceptron, and analyze and predict the multi-variable aggregated features through the multi-layer perceptron to output the predicted time series data of the multi-variables. The process of analyzing and predicting through the multi-layer perceptron can be expressed as:

[0101]

[0102] Where, is the predicted value of the multi-layer perceptron; σ is the activation function; represents the multi-variable aggregated features; s1:s p represents p features in the multi-variables; W pre and b pre are the parameters in the graph convolutional neural network model.

[0103] In the above data analysis and prediction method, the multi-variable aggregated features are input into the prediction layer, and the multi-variable aggregated features are analyzed and predicted through the prediction layer to obtain the predicted time series data of the multi-variables. In this method, the prediction layer in the graph convolutional neural network model is used to predict the multi-variable aggregated features, making the obtained predicted time series data of the multi-variables more accurate.

[0104] Figure 6 This is a schematic flowchart of the data analysis and prediction method provided by the embodiments of the present application. The embodiments of the present application relate to an optional implementation manner of updating the parameters of the graph convolutional neural network model. On the basis of Figure 2 the illustrated embodiment, as Figure 6 shown, the above process may further include the following steps:

[0105] S601, obtain the true time series data corresponding to the predicted time series data of each variable.

[0106] Specifically, the server may obtain the true time series data at that moment according to the moment corresponding to the predicted time series data of each variable and the electronic tag of the target entity. For example, if the moment corresponding to the predicted time series data of each variable is 9:00 on March 24, 2022, after the time of 9:00 on March 24, 2022, obtain the true time series data corresponding to that moment.

[0107] S602, update the model parameters in the graph convolutional neural network model according to the difference between the predicted time series data of each variable and the corresponding true time series data.

[0108] Specifically, the server can calculate the difference between the predicted time series data of each variable and the corresponding real time series data. The smaller the difference, the more accurate the graph convolutional neural network model; the larger the difference, the less accurate the graph convolutional neural network model. According to this difference, the model parameters in the graph convolutional neural network model are updated, so that the difference between the predicted time series data of the graph convolutional neural network model and the corresponding real time series data approaches zero. This process can be expressed as:

[0109]

[0110] Among them, Θ is the trainable parameter in the graph convolutional neural network model, and MSE is the mean square error. is the predicted time series data of each variable, is the real time series data. The graph convolutional neural network model parameters are iteratively optimized through the gradient descent algorithm until the graph convolutional neural network model converges.

[0111] In the above data analysis and prediction method, the real time series data corresponding to the predicted time series data of each variable is obtained, and the model parameters in the graph convolutional neural network model are updated according to the difference between the predicted time series data of each variable and the corresponding real time series data. This method retrains the graph convolutional neural network model by the difference between the predicted time series data and the real time series data, updates the model parameters in the graph convolutional neural network model, increases the training data of the graph convolutional neural network model, and makes the obtained graph convolutional neural network model more accurate.

[0112] Figure 7 This is a schematic flowchart of the data analysis and prediction method provided by the embodiments of the present application. The embodiments of the present application relate to an optional implementation manner of the construction process of the graph convolutional neural network model. On the basis of the embodiment shown in Figure 2 As shown in Figure 7 the above process may further include the following steps:

[0113] S701, obtain the historical time series sample data of multiple sample variables.

[0114] For the specific obtaining steps, refer to step S201.

[0115] S702, according to the historical time series sample data of each sample variable, obtain the sample graph structure data corresponding to the historical time series sample data of each sample variable.

[0116] Optionally, the server may call a variable dependency extractor to extract the dependency quantification values between each sample variable, and determine the sample graph structure data corresponding to the historical time series sample data of each sample variable according to the dependency quantification values between each sample variable. Optionally, the server may calculate the similarity between each sample variable. If the similarity between each sample variable is greater than a preset similarity threshold, an edge connection operation is performed on each sample variable. If the similarity between each sample variable is less than the preset similarity threshold, no edge connection operation is performed on the sample variable, and the sample graph structure data corresponding to the historical time series sample data of each sample variable is determined according to the edge connection result of each sample variable.

[0117] S703. Train the initial graph convolutional neural network model with the sample graph structure data until the preset convergence condition is met, determine that the graph convolutional neural network model converges, and obtain the preset graph convolutional neural network model.

[0118] Optionally, input the sample graph structure data into the graph convolutional neural network for training. The graph convolutional neural network model extracts the features of the sample graph structure data through several convolutional layers, then inputs the features of the graph structure data into the transfer learning model to obtain the transfer loss function, and uses the stochastic gradient descent algorithm to optimize the transfer loss function until the transfer loss function converges. Optimize and update the parameters of each layer of the graph convolutional neural network model according to the converged transfer loss function to obtain the graph convolutional neural network model, and use this graph convolutional neural network model as the preset graph convolutional neural network model.

[0119] In the above data analysis and prediction method, obtain the historical time series sample data of multiple sample variables, and according to the historical time series sample data of each sample variable, obtain the sample graph structure data corresponding to the historical time series sample data of each sample variable, and train the initial graph convolutional neural network model with the sample graph structure data until the preset convergence condition is met, determine that the graph convolutional neural network model converges, and obtain the preset graph convolutional neural network model. This method trains the initial graph convolutional neural network model with the historical time series sample data of multiple sample variables. Compared with the prediction results obtained through a single sample, the prediction accuracy and prediction efficiency of this method are higher.

[0120] In one embodiment, for the convenience of understanding by those skilled in the art, the data analysis and prediction method is introduced in detail below. As Figure 8 shown, the method may include:

[0121] S801. Obtain all variable nodes in the historical time series data of each variable;

[0122] S802. Randomly group each variable node to obtain multiple variable node sets;

[0123] S803, obtain the features of the graph structure data at each sampling moment in the first convolutional layer;

[0124] S804, input the features of each sampling moment in the first convolutional layer into the second convolutional layer, and aggregate the features of the Nth sampling moment and the (N - 1)th sampling moment in the first convolutional layer to form the features of the Nth sampling moment in the second convolutional layer, obtaining the features of each sampling moment in the second convolutional layer;

[0125] S805, input the features of each sampling moment in the second convolutional layer into the third convolutional layer, and aggregate the features of the Nth sampling moment, the (N - 1)th sampling moment, and the (N - 2)th sampling moment in the second convolutional layer to form the features of the Nth sampling moment in the third convolutional layer, obtaining the features of each sampling moment in the third convolutional layer;

[0126] S806, and so on, obtaining the features of each sampling moment in the Nth convolutional layer, and determining the feature of the last sampling moment in the Nth convolutional layer as the multi-variable aggregated feature;

[0127] S807, input the multi-variable aggregated feature into the prediction layer, and analyze and predict the multi-variable aggregated feature through the prediction layer to obtain the predicted time series data of the multi-variable;

[0128] S808, obtain the true time series data corresponding to the predicted time series data of each variable;

[0129] S809, update the model parameters in the graph convolutional neural network model according to the difference between the predicted time series data of each variable and the corresponding true time series data.

[0130] It should be noted that for the descriptions in S801 - S809 above, reference can be made to the relevant descriptions in the above embodiments, and their effects are similar. This embodiment will not be elaborated here.

[0131] Furthermore, it can be understood that Figure 9 represents the flow schematic diagram of the data analysis and prediction method. Figure 9 The graph below the first convolutional layer represents the graph structure data at different sampling moments. Input this graph structure data into the first convolutional layer in the graph convolutional neural network model. The first convolutional layer aggregates the features of the Nth sampling moment and the (N - 1)th sampling moment. The second convolutional layer aggregates the features of the Nth sampling moment, the (N - 1)th sampling moment, and the (N - 2)th sampling moment, and so on. The last convolutional layer aggregates all the features of the sampling moments into the Nth sampling moment, obtaining the multi-variable aggregated feature. Input this multi-variable aggregated feature into the prediction layer, and predict the multi-variable aggregated feature through the prediction layer to obtain the predicted time series data of the multi-variable.

[0132] In the above data analysis and prediction method, all variable nodes in the historical time series data of each variable are obtained, the variable nodes are randomly grouped to obtain multiple variable node sets, a preset number of edge connection operations are performed on each variable node to obtain the graph structure data of the historical time series data of each variable, the features at each sampling moment in the first convolutional layer of the graph structure data are obtained, the features at each sampling moment in the first convolutional layer are input into the second convolutional layer, and the features at the Nth sampling moment and the features at the (N - 1)th sampling moment in the first convolutional layer are aggregated to form the features at the Nth sampling moment in the second convolutional layer, so as to obtain the features at each sampling moment in the second convolutional layer. The features at each sampling moment in the second convolutional layer are input into the third convolutional layer, and the features at the Nth sampling moment, the features at the (N - 1)th sampling moment, and the features at the (N - 2)th sampling moment in the second convolutional layer are aggregated to form the features at the Nth sampling moment in the third convolutional layer, so as to obtain the features at each sampling moment in the third convolutional layer. And so on, the features at each sampling moment in the Nth convolutional layer are obtained, and the features at the last sampling moment in the Nth convolutional layer are determined as the multi-variable aggregation features. The multi-variable aggregation features are input into the prediction layer, and the prediction layer analyzes and predicts the multi-variable aggregation features to obtain the predicted time series data of the multi-variables. The real time series data corresponding to the predicted time series data of each variable is obtained, and the model parameters in the graph convolutional neural network model are updated according to the difference between the predicted time series data of each variable and the corresponding real time series data. The number of convolutional layers in the graph convolutional neural network model in this method is determined based on the number of sampling moments in the time dimension. For different historical time series data of multi-variables, different graph convolutional neural network models can be selected for prediction, making the prediction process of the historical time series data of each variable more flexible. At the same time, by predicting the historical time series data of multi-variables, compared with only predicting the historical time series data of a single variable, the prediction result is more accurate. Through the convolutional layer in the graph convolutional neural network, the historical time series data of each variable can be aggregated, and through the aggregated historical time series data of each variable, the multi-variable time series data in the future period can be accurately predicted.

[0133] It should be understood that although the steps in the flowcharts involved in the above embodiments are sequentially shown according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless there is a clear description in this article, the execution of these steps has no strict order limit, and these steps can be executed in other orders. Moreover, at least a part of the steps in the flowcharts involved in the above embodiments may include multiple steps or multiple stages. These steps or stages are not necessarily executed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be executed alternately or in turn with at least a part of other steps or steps or stages in other steps.

[0134] Based on the same inventive concept, an embodiment of the present application further provides a data analysis and prediction device for implementing the data analysis and prediction method described above. The solution provided by this device to solve the problem is similar to the solution described in the above method. Therefore, the specific limitations in one or more embodiments of the data analysis and prediction device provided below can refer to the limitations on the data analysis and prediction method in the above text, and will not be repeated here.

[0135] In one embodiment, as Figure 10 shown, a data analysis and prediction device is provided, including: a first acquisition module 11 and a first determination module 12, where:

[0136] The first acquisition module 11 is used to obtain the graph structure data of the historical time series data of each multi-variable according to the historical time series data of multiple variables of the target subject;

[0137] The first determination module 12 is used to input the graph structure data into a preset graph convolutional neural network model, and after aggregating the graph structure data in the time dimension through each convolutional layer in the graph convolutional neural network model, obtain the predicted time series data of the multi-variable; the number of convolutional layers in the graph convolutional neural network model is determined based on the number of sampling moments in the time dimension.

[0138] The data analysis and prediction device provided in this embodiment can execute the above method embodiment, and its implementation principle and technical effect are similar, and will not be repeated here.

[0139] In one embodiment, the above first acquisition module includes: an acquisition unit and a processing unit, where:

[0140] The first acquisition unit is used to acquire all variable nodes in the historical time series data of each variable;

[0141] The first processing unit is used to perform edge connection processing on each variable node to obtain the graph structure data of the historical time series data of each variable.

[0142] The data analysis and prediction device provided in this embodiment can execute the above method embodiment, and its implementation principle and technical effects are similar, which will not be elaborated here.

[0143] Optionally, the above first processing unit is specifically configured to perform edge connection operations on each variable node for a preset number of times to obtain graph structure data of historical time series data of each variable; wherein, the edge connection operation includes: randomly grouping each variable node to obtain a plurality of variable node sets; obtaining the similarity of variable nodes in each variable node set; and connecting variable nodes corresponding to similarities greater than a preset threshold in each variable node set.

[0144] The data analysis and prediction device provided in this embodiment can execute the above method embodiment, and its implementation principle and technical effects are similar, which will not be elaborated here.

[0145] In one embodiment, the above first determination module includes: a second acquisition unit, a second processing unit, a third processing unit, an Nth processing unit, and a generation unit, where:

[0146] The second acquisition unit is configured to acquire the features of each sampling moment in the first convolutional layer of the graph structure data.

[0147] The second processing unit is configured to input the features of each sampling moment in the first convolutional layer into the second convolutional layer, and aggregate the features of the Nth sampling moment and the (N - 1)th sampling moment in the first convolutional layer to form the features of the Nth sampling moment in the second convolutional layer, so as to obtain the features of each sampling moment in the second convolutional layer.

[0148] The third processing unit is configured to input the features of each sampling moment in the second convolutional layer into the third convolutional layer, and aggregate the features of the Nth sampling moment, the (N - 1)th sampling moment, and the (N - 2)th sampling moment in the second convolutional layer to form the features of the Nth sampling moment in the third convolutional layer, so as to obtain the features of each sampling moment in the third convolutional layer.

[0149] The Nth processing unit is configured to, by analogy, obtain the features of each sampling moment in the Nth convolutional layer, and determine the features of the last sampling moment in the Nth convolutional layer as the multi-variable aggregation features.

[0150] The generation unit is configured to generate predicted time series data of multi-variables according to the multi-variable aggregation features.

[0151] The data analysis and prediction device provided in this embodiment can execute the above method embodiment, and its implementation principle and technical effects are similar, which will not be elaborated here.

[0152] Optionally, the above-mentioned generation unit is specifically configured to input the multi-variable aggregation feature into a prediction layer, and analyze and predict the multi-variable aggregation feature through the prediction layer to obtain the predicted time series data of the multi-variable.

[0153] The data analysis and prediction device provided in this embodiment can execute the above method embodiment, and its implementation principle and technical effect are similar, which will not be elaborated here.

[0154] In one embodiment, the above-mentioned data analysis and prediction device further includes: a second acquisition module and an update module, where:

[0155] The second acquisition module is configured to acquire the real time series data corresponding to the predicted time series data of each variable;

[0156] The update module is configured to update the model parameters in the graph convolutional neural network model according to the difference between the predicted time series data of each variable and the corresponding real time series data.

[0157] The data analysis and prediction device provided in this embodiment can execute the above method embodiment, and its implementation principle and technical effect are similar, which will not be elaborated here.

[0158] In one embodiment, the above-mentioned data analysis and prediction device further includes: a third acquisition module, a fourth acquisition module, and a second determination module, where:

[0159] The third acquisition module is configured to acquire the historical time series sample data of multiple sample variables;

[0160] The fourth acquisition module is configured to acquire the sample graph structure data corresponding to the historical time series sample data of each sample variable according to the historical time series sample data of each sample variable;

[0161] The second determination module is configured to train the initial graph convolutional neural network model through the sample graph structure data until a preset convergence condition is met, determine that the graph convolutional neural network model converges, and obtain a preset graph convolutional neural network model.

[0162] The data analysis and prediction device provided in this embodiment can execute the above method embodiment, and its implementation principle and technical effect are similar, which will not be elaborated here.

[0163] Each module in the above-mentioned data analysis and prediction device can be implemented in whole or in part by software, hardware, and their combination. The above-mentioned modules can be embedded in or independent of the processor in the server in the form of hardware, or stored in the memory in the server in the form of software, so that the processor can call and execute the operations corresponding to the above-mentioned modules.

[0164] In one embodiment, a server is provided, including a memory and a processor. A computer program is stored in the memory, and when the processor executes the computer program, all the contents in the above method embodiments are implemented.

[0165] In one embodiment, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, all the contents in the above method embodiments are implemented.

[0166] In one embodiment, a computer program product is provided, including a computer program. When the computer program is executed by a processor, all the contents in the above method embodiments are implemented.

[0167] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data for analysis, stored data, displayed data, etc.) involved in this application are all information and data that have been authorized by the user or fully authorized by all parties.

[0168] Those of ordinary skill in the art can understand that all or part of the processes in the methods of the above embodiments can be completed by instructing relevant hardware through a computer program. The computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above methods. Among them, any reference to a memory, database, or other medium used in the embodiments provided in the present application can include at least one of non-volatile and volatile memories. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetoresistive random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM), etc. The databases involved in the embodiments provided in the present application can include at least one of relational databases and non-relational databases. Non-relational databases can include distributed databases based on blockchain, etc., without limitation. The processors involved in the embodiments provided in the present application can be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, data processing logics based on quantum computing, etc., without limitation.

[0169] The technical features of the above embodiments can be combined arbitrarily. For the sake of concise description, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, it should be considered as the scope recorded in this specification.

[0170] The above embodiments only represent several implementation manners of the present application. The description is relatively specific and detailed, but it should not be construed as a limitation on the patent scope of the present application. It should be noted that for those of ordinary skill in the art, without departing from the concept of the present application, several modifications and improvements can still be made, and these all belong to the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the appended claims.

Claims

1. A data analysis and prediction method, characterized in that, The method includes: Obtaining the graph structure data of the historical time series data of each of the variables according to the historical time series data of multiple variables of the target entity; Inputting the graph structure data into a preset graph convolutional neural network model, and aggregating the graph structure data in the time dimension through each convolutional layer in the graph convolutional neural network model to obtain the predicted time series data of multiple variables; the number of convolutional layers in the graph convolutional neural network model is determined based on the number of sampling moments in the time dimension; The obtaining the graph structure data of the historical time series data of each of the variables according to the historical time series data of multiple variables of the target entity includes: Obtaining all variable nodes in the historical time series data of each of the variables; Performing edge connection processing on each of the variable nodes to obtain the graph structure data of the historical time series data of each of the variables; If both the number of convolutional layers and the number of sampling moments are N, where N is a positive integer; Then the aggregating the graph structure data in the time dimension through each convolutional layer in the graph convolutional neural network model to obtain the predicted time series data of multiple variables includes: Obtaining the features of each sampling moment in the first convolutional layer of the graph structure data; Inputting the features of each sampling moment in the first convolutional layer into the second convolutional layer, and aggregating the features of N sampling moments and the features of N - 1 sampling moments in the first convolutional layer to form the features of N sampling moments in the second convolutional layer, to obtain the features of each sampling moment in the second convolutional layer; Inputting the features of each sampling moment in the second convolutional layer into the third convolutional layer, and aggregating the features of N sampling moments, the features of N - 1 sampling moments, and the features of N - 2 sampling moments in the second convolutional layer to form the features of N sampling moments in the third convolutional layer, to obtain the features of each sampling moment in the third convolutional layer; By analogy, obtaining the features of each sampling moment in the Nth convolutional layer, and determining the feature of the last sampling moment in the Nth convolutional layer as the multi-variable aggregation feature; Generating the predicted time series data of multiple variables according to the multi-variable aggregation feature.

2. The method according to claim 1, wherein The performing edge connection processing on each of the variable nodes to obtain the graph structure data of the historical time series data of each of the variables includes: Performing edge connection operations on each of the variable nodes for a preset number of times to obtain the graph structure data of the historical time series data of each of the variables; Wherein, the edge connection operation includes: Randomly grouping each of the variable nodes to obtain multiple variable node sets; Obtaining the similarity of the variable nodes in each of the variable node sets; Connecting the variable nodes corresponding to the similarities greater than a preset threshold in each of the variable node sets.

3. The method according to claim 1, characterized in that, The preset graph convolutional neural network model further includes a prediction layer; then the generating the predicted time series data of multiple variables according to the multi-variable aggregation feature includes: Inputting the multi-variable aggregation feature into the prediction layer, and analyzing and predicting the multi-variable aggregation feature through the prediction layer to obtain the predicted time series data of multiple variables.

4. The method according to claim 1, characterized in that, The method further includes: Obtain the real time series data corresponding to the predicted time series data of each of the variables; Update the model parameters in the graph convolutional neural network model according to the difference between the predicted time series data and the corresponding real time series data of each of the variables.

5. The method according to claim 1, wherein The construction process of the graph convolutional neural network model includes: Obtain the historical time series sample data of multiple sample variables; According to the historical time series sample data of each of the sample variables, obtain the sample graph structure data corresponding to the historical time series sample data of each of the sample variables; Train the initial graph convolutional neural network model through the sample graph structure data until the preset convergence condition is met, determine that the graph convolutional neural network model converges, and obtain the preset graph convolutional neural network model.

6. A data analysis and prediction device, characterized in that, The device includes: A first acquisition module, configured to obtain the graph structure data of the historical time series data of each of the variables according to the historical time series data of multiple variables of the target subject; the obtaining the graph structure data of the historical time series data of each of the variables according to the historical time series data of multiple variables of the target subject includes: obtaining all variable nodes in the historical time series data of each of the variables; performing edge connection processing on each of the variable nodes to obtain the graph structure data of the historical time series data of each of the variables; A first determination module, configured to input the graph structure data into a preset graph convolutional neural network model, and after aggregating the graph structure data in the time dimension through each convolutional layer in the graph convolutional neural network model, obtain the predicted time series data of multiple variables; the number of convolutional layers in the graph convolutional neural network model is determined based on the number of sampling moments in the time dimension; if both the number of convolutional layers and the number of sampling moments are N, where N is a positive integer; then the aggregating the graph structure data in the time dimension through each convolutional layer in the graph convolutional neural network model to obtain the predicted time series data of multiple variables includes: Obtain the features of each sampling moment in the first convolutional layer of the graph structure data; Input the features of each sampling moment in the first convolutional layer into the second convolutional layer, and aggregate the features of N sampling moments and the features of N - 1 sampling moments in the first convolutional layer to form the features of N sampling moments in the second convolutional layer, and obtain the features of each sampling moment in the second convolutional layer; Input the features of each sampling moment in the second convolutional layer into the third convolutional layer, and aggregate the features of N sampling moments, the features of N - 1 sampling moments, and the features of N - 2 sampling moments in the second convolutional layer to form the features of N sampling moments in the third convolutional layer, and obtain the features of each sampling moment in the third convolutional layer; And so on, obtain the features of each sampling moment in the Nth convolutional layer, and determine the feature of the last sampling moment in the Nth convolutional layer as the multi-variable aggregation feature; Generate the predicted time series data of multiple variables according to the multi-variable aggregation feature.

7. A server, comprising a memory and a processor, the memory storing a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 5.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 5.

9. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 5.

Images