Time sequence prediction method and device, equipment and storage medium
By introducing two-way long and short-term memory networks and graph convolution-long and short-term memory networks into the time series prediction model, the problem of poor accuracy of time series prediction is solved, more efficient feature extraction and prediction accuracy is achieved, and the development of financial business and other fields has been promoted.
Patent Information
- Application Number
- CN202411551316.X
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-10-31
- Publication Date
- 2025-03-18
AI Technical Summary
When making time series predictions in the prior art, the problem of poor prediction accuracy is prone to occur, resulting in poor user experience, especially in financial business planning, which may affect business development.
A time series prediction model based on bidirectional long and short-term memory networks and graph convolution-long and short-term memory networks is adopted. Through bidirectional feature extraction and cross-modal feature extraction, feature information under different modes is comprehensively predicted.
It improves the accuracy of time series prediction, and can more effectively capture the long-term dependencies and cross-modal characteristics in the time series, thereby improving the accuracy of prediction results and promoting the development of financial business and other fields.
Smart Images
Figure CN119646476A_ABST
Abstract
Description
Technical Field
[0001] The present application relates to the technical field of time series prediction, and in particular to a time series prediction method, apparatus, device and storage medium. Background Art
[0002] Time series forecasting refers to using historical time series data to forecast future time series corresponding to the field in which the historical time series data is located. For example, time series forecasting can be applied to financial services. For example, the application scenarios of time series forecasting in financial services can include stock price forecasting, stock index forecasting, commodity price forecasting, commodity price volatility forecasting, etc.
[0003] At present, since time series prediction can be applied in daily life, the prediction demand of users for time series prediction has also increased accordingly. However, in the related art, when performing time series prediction, it is easy to have poor prediction accuracy of time series, resulting in poor prediction experience of users for time series prediction. For example, for financial business, if the prediction accuracy of time series related to financial business is poor, if the prediction results of time series related to financial business are used as business planning guidance for financial business, it is easy to have poor business planning accuracy for financial business, thereby adversely affecting the business development of financial business. Summary of the invention
[0004] The main purpose of this application is to provide a time series prediction method, device, equipment and storage medium, aiming to improve the prediction accuracy of time series.
[0005] In a first aspect, the present application provides a time series prediction method, comprising the following steps:
[0006] Get the preset time series of the preset object;
[0007] Based on a bidirectional long short-term memory network of a preset time series prediction model, performing bidirectional feature extraction processing on the preset time series to obtain first feature information corresponding to the preset time series;
[0008] Based on the graph convolution-long short-term memory network of the time series prediction model, cross-modal feature extraction processing is performed on the preset time series to obtain second feature information corresponding to the preset time series;
[0009] Based on the time series prediction network of the time series prediction model, the preset time series is predicted according to the first feature information and the second feature information to obtain a target prediction result corresponding to the preset time series.
[0010] In a second aspect, the present application further provides a time series prediction device, the time series prediction device comprising:
[0011] An acquisition module, used to acquire a preset time series of a preset object;
[0012] A first feature extraction module is used to perform bidirectional feature extraction processing on the preset time series based on a bidirectional long short-term memory network of a preset time series prediction model to obtain first feature information corresponding to the preset time series;
[0013] A second feature extraction module is used to perform cross-modal feature extraction processing on the preset time series based on the graph convolution-long short-term memory network of the time series prediction model to obtain second feature information corresponding to the preset time series;
[0014] A prediction module is used to predict the preset time series based on the time series prediction network of the time series prediction model according to the first feature information and the second feature information to obtain a target prediction result corresponding to the preset time series.
[0015] In a third aspect, the present application further provides a computer device, the computer device comprising a memory and a processor;
[0016] The memory is used to store computer programs;
[0017] The processor is used to execute the computer program and implement the steps of the time series prediction method as described above when executing the computer program.
[0018] In a fourth aspect, the present application further provides a computer-readable storage medium having a computer program stored thereon, and when the computer program is executed by a processor, the steps of the time series prediction method as described above are implemented.
[0019] The present application provides a time series prediction method, apparatus, device and storage medium. The time series prediction method includes: obtaining a preset time series of a preset object; performing bidirectional feature extraction processing on the preset time series based on a bidirectional long short-term memory network of a preset time series prediction model to obtain first feature information corresponding to the preset time series; performing cross-modal feature extraction processing on the preset time series based on a graph convolution-long short-term memory network of the time series prediction model to obtain second feature information corresponding to the preset time series; and predicting the preset time series based on the first feature information and the second feature information using a timing prediction network of the time series prediction model to obtain a target prediction result corresponding to the preset time series.
[0020] In the case of obtaining the preset time series of the preset object, the time series prediction model can perform corresponding feature extraction processing on the preset time series through the bidirectional long short-term memory network and the graph convolution-long short-term memory network respectively. Since the graph convolution-long short-term memory network involves cross-modal feature extraction processing on the preset time series, the first feature information and the second feature information can be the feature information of the preset time series in different modes. Accordingly, the time series prediction network can integrate the feature information of the preset time series in different modes to predict the preset time series, which is conducive to improving the prediction accuracy of the preset time series. For example, for the time series of financial business, the time series prediction model can integrate the feature information of the time series in different modes to predict the time series to obtain the corresponding prediction results, which is conducive to improving the prediction accuracy of the time series of financial business, thereby promoting the development of financial business. BRIEF DESCRIPTION OF THE DRAWINGS
[0021] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.
[0022] Figure 1 It is a flowchart of a time series prediction method provided by an embodiment of the present application;
[0023] Figure 2 is a schematic diagram of a time series prediction method involved in an embodiment of the present application;
[0024] Figure 3 is a schematic block diagram of a time series prediction device provided in an embodiment of the present application;
[0025] Figure 4 It is a schematic block diagram of the structure of a computer device provided in an embodiment of the present application. DETAILED DESCRIPTION
[0026] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.
[0027] The flowcharts shown in the accompanying drawings are only examples and do not necessarily include all the contents and operations / steps, nor must they be executed in the order described. For example, some operations / steps may also be decomposed, combined or partially merged, so the actual execution order may change according to actual conditions.
[0028] The embodiments of the present application provide a time series prediction method, apparatus, device and storage medium. The time series prediction method can be applied to a computer device, which can be a mobile phone, a tablet computer, a laptop computer, a desktop computer and the like. It can also be applied to a server, which can be a separate server or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, content delivery networks (CDNs), and big data and artificial intelligence platforms.
[0029] In conjunction with the accompanying drawings, some embodiments of the present application are described in detail below. In the absence of conflict, the following embodiments and features in the embodiments can be combined with each other.
[0030] See also Figure 1 , Figure 1 It is a flowchart of a time series prediction method provided by an embodiment of the present application. It should be noted that the time series prediction method provided by the embodiment of the present application can be used in a computer device, and of course can also be used in a server. For example, the server can obtain a preset time series of a preset object from a computer device, and perform corresponding processing on the preset time series according to the time series prediction method to obtain a target prediction result corresponding to the preset time series. Exemplarily, the server can, for example, send the target prediction result to a computer device; of course, it is not limited to this, and is not limited here.
[0031] In specific implementation, computer devices include but are not limited to: any one of mobile phones, tablet computers, laptop computers, and desktop computers; the server can be a single server or a server cluster, or a cloud server that provides cloud computing services.
[0032] like Figure 1 As shown, the time series prediction method includes steps S101 to S104.
[0033] Step S101: Obtain a preset time sequence of a preset object.
[0034] For example, when there is a need to predict a time series, the computer device may obtain a preset time series of a preset object. The computer device may input the obtained preset time series into a time series prediction model for subsequent prediction of a target prediction result corresponding to the preset time series.
[0035] The preset objects may include different products in different fields. For example, for financial services, the preset objects may include financial investment products, such as insurance, gold, stocks, etc., which are not limited here.
[0036] The preset time series of the preset object can be used to indicate the preset values corresponding to the preset object at different time points. The preset values include, for example, price, volatility, etc., which are not limited here.
[0037] In an exemplary embodiment, the preset time sequence of the preset object can be expressed as (x t-1 ,x t-2 ,…,x t-p ). Wherein, x is used to indicate a preset object, t-1 to tp are used to indicate different time points, and p is a positive integer.
[0038] In this way, when a preset time series of a preset object is obtained, the preset time series can be used by the computer device to subsequently determine a target prediction result corresponding to the preset time series.
[0039] In some implementations, the preset numerical values corresponding to each time point in the preset time series are normalized to obtain a processed preset time series.
[0040] For example, since the preset time series can be used to indicate the preset values corresponding to different time points, the difference between the preset values at different time points may be large. Taking financial services as an example, the price of gold, the volatility of stocks, etc. are prone to large changes at different time points.
[0041] Accordingly, when the difference between the preset values at different time points is large, the data scale difference between the preset values at different time points will also be large. In order to avoid the data scale difference from adversely affecting the prediction accuracy of the target prediction result corresponding to the subsequent preset time series, after obtaining the preset time series, the computer device can perform normalization processing on the preset values corresponding to each time point in the preset time series to obtain the processed preset time series.
[0042] In an exemplary implementation, the computer device may normalize the preset numerical values corresponding to each time point in the preset time series to a numerical value between 0 and 1, thereby obtaining a processed preset time series.
[0043] Accordingly, when determining the processed preset time series, the computer device can input the processed preset time series into the time series prediction model, so that the time series prediction model can predict the target prediction result corresponding to the processed preset time series based on the processed preset time series.
[0044] Based on this, the data scale differences between the processed preset values at different time points in the processed preset time series can remain unchanged, which can avoid the adverse effects of the data scale differences on the prediction accuracy of the target prediction results corresponding to the subsequently processed preset time series, thereby facilitating the subsequent improvement of the prediction accuracy of the target prediction results corresponding to the processed preset time series.
[0045] Step S102: Based on the bidirectional long short-term memory network of the preset time series prediction model, a bidirectional feature extraction process is performed on the preset time series to obtain first feature information corresponding to the preset time series.
[0046] For example, the time series prediction model may be provided with a bidirectional long short-term memory (BiLSTM) network. When a preset time series is obtained, the computer device may input the preset time series into the BiLSTM network of the time series prediction model, so that the BiLSTM network performs bidirectional feature extraction processing on the preset time series to obtain first feature information corresponding to the preset time series.
[0047] In the process of performing bidirectional feature extraction processing on the preset time series through the BiLSTM network, the BiLSTM network can be used to capture the long-term dependency between the preset numerical values corresponding to different time points in the preset time series. Accordingly, by performing bidirectional feature extraction processing on the preset time series through the BiLSTM network, the BiLSTM network can simultaneously consider the historical change trend and future change trend corresponding to the preset time series, which is conducive to better capturing the change trend and change periodicity between the preset numerical values corresponding to different time points in the preset time series. Based on this, the first feature information corresponding to the preset time series can be used to indicate the long-term dependency between the preset numerical values corresponding to different time points in the preset time series, the change trend between the preset numerical values, and the change periodicity.
[0048] In some embodiments, a forward feature extraction process is performed on a preset time series through a forward long short-term memory layer of a bidirectional long short-term memory network to obtain forward feature information corresponding to the preset time series; a reverse feature extraction process is performed on the preset time series through a reverse long short-term memory layer of a bidirectional long short-term memory network to obtain reverse feature information corresponding to the preset time series; the forward feature information and the reverse feature information are merged to obtain first feature information.
[0049] like Figure 2 As shown, a forward long short-term memory layer and a reverse long short-term memory layer can be set in the BiLSTM network. The forward long short-term memory (LSTM) layer and the reverse long short-term memory layer can be used to memorize and process the temporal relationship of a preset time series.
[0050] The BiLSTM network can traverse the preset values in the preset time series twice to perform bidirectional feature extraction on the preset time series. For example, the BiLSTM network can process the preset time series from left to right through the forward LSTM layer to obtain forward feature information. The BiLSTM network can process the preset time series from right to left through the reverse LSTM layer to obtain reverse feature information.
[0051] Accordingly, the BiLSTM network can determine the first feature information based on the forward feature information and the reverse feature information. For example, the BiLSTM network can also be provided with a fully connected layer. The fully connected layer can merge the forward feature information and the reverse feature information to obtain the first feature information. For example, the fully connected layer can concatenate the forward feature information and the reverse feature information, and perform corresponding classification, regression and other operations to determine the first feature information corresponding to the preset time series, which is not limited here.
[0052] In an exemplary embodiment, the bidirectional feature extraction process of the BiLSTM network for a preset time series can be expressed as:
[0053] h t =f(w1x t +w2h t-1 )
[0054] h′ t =f(w3x t +w5h t+1 )
[0055] o t =g(w4h′ t +w6h t )
[0056] Among them, x t Used to indicate the preset time series, h t It is used to instruct the forward LSTM layer to extract the positive feature information of the preset time series, h′ t It is used to indicate the reverse feature information obtained by the reverse LSTM layer through reverse feature extraction processing on the preset time series. w1 to w6 are used to indicate the weight coefficients of the neurons involved in the BiLSTM network. t Used to indicate the first feature information obtained by merging the forward feature information and the reverse feature information.
[0057] In this way, when the first feature information corresponding to the preset time series is determined, the first feature information can be used by the time series prediction model to subsequently predict the target prediction result corresponding to the preset time series. Based on the fact that the first feature information is obtained by the BiLSTM network performing bidirectional feature extraction processing on the preset time series, the first feature information can be used to indicate the long-term dependency between the preset values corresponding to different time points in the preset time series, the change trend between the preset values, and the change periodicity. Accordingly, the time series prediction model can assist in the subsequent prediction of the target prediction result corresponding to the preset time series through the long-term dependency between the preset values corresponding to different time points in the preset time series indicated by the first feature information, the change trend between the preset values, and the change periodicity, which is conducive to improving the subsequent prediction accuracy of the target prediction result corresponding to the preset time series.
[0058] Step S103: Based on the graph convolution-long short-term memory network of the time series prediction model, cross-modal feature extraction processing is performed on the preset time series to obtain second feature information corresponding to the preset time series.
[0059] For example, the time series prediction model may be provided with a graph convolutional network (GCN)-long short-term memory (LSTM) network. When a preset time series is obtained, the computer device may input the preset time series into the GCN-LSTM network of the time series prediction model, so that the GCN-LSTM network performs cross-modal feature extraction processing on the preset time series to obtain the second feature information corresponding to the preset time series.
[0060] like Figure 2As shown, a preset LSTM layer and a GCN layer can be provided in the GCN-LSTM network. Since the GCN layer can be used to process graph structure data, such as a graph structure matrix, and the preset LSTM layer can be used to process a preset time series, the GCN layer and the preset LSTM layer can respectively perform feature extraction processing on the preset time series under different modes, such as performing cross-modal feature extraction processing on the preset time series to obtain the second feature information corresponding to the preset time series.
[0061] In some embodiments, a preset time series is subjected to unidirectional feature extraction processing through a preset long short-term memory layer of a graph convolution-long short-term memory network to obtain an intermediate feature sequence corresponding to the preset time series; and a cross-modal feature extraction processing is performed on the intermediate feature sequence through a graph convolution layer of a graph convolution-long short-term memory network to obtain second feature information.
[0062] For example, when the GCN-LSTM network obtains a preset time series, the preset time series can be input into the preset LSTM layer, so that the preset LSTM layer performs unidirectional feature extraction processing on the preset time series to obtain an intermediate feature sequence corresponding to the preset time series. Accordingly, when the GCN-LSTM network determines the intermediate feature sequence, the intermediate feature sequence can be input into the GCN layer, so that the GCN layer performs cross-modal feature extraction processing on the intermediate feature sequence to obtain the second feature information.
[0063] Exemplarily, before the intermediate feature sequence is input into the GCN layer, the GCN-LSTM network can perform cross-modal conversion processing on the intermediate feature sequence to determine the graph structure matrix corresponding to the intermediate feature sequence. The graph structure matrix corresponding to the intermediate feature sequence can be used by the GCN-LSTM network to subsequently determine the second feature information.
[0064] In some embodiments, based on a preset architecture size of a preset long short-term memory layer and an intermediate feature sequence, a graph structure matrix corresponding to the intermediate feature sequence is determined; through a graph convolution layer, cross-modal feature extraction processing is performed on the intermediate feature sequence and the graph structure matrix to obtain second feature information.
[0065] Since the output of the preset LSTM layer in the GCN-LSTM network, such as the intermediate feature sequence, can be used as the input of the GCN layer, the GCN-LSTM network can determine the graph structure matrix corresponding to the intermediate feature sequence according to the preset architecture size of the preset LSTM layer and the intermediate feature sequence. The graph structure matrix corresponding to the intermediate feature sequence can be input to the GCN layer so that the GCN layer can perform cross-modal feature extraction processing on the preset time series.
[0066] For example, according to the preset architecture size, the node information corresponding to the graph structure matrix is determined; according to the intermediate feature sequence, the feature dimension corresponding to the graph structure matrix is determined; according to the node information and the feature dimension, the graph structure matrix is determined.
[0067] Take the preset architecture size including the preset LSTM layer length as an example. The GCN-LSTM network can determine the node information corresponding to the graph structure matrix, such as the number of nodes, according to the preset LSTM layer length. Accordingly, the GCN-LSTM network can determine the feature dimension corresponding to the graph structure matrix according to the output of the preset LSTM layer, that is, the intermediate feature sequence. When the node information and feature dimension corresponding to the graph structure matrix are determined, the GCN-LSTM network can obtain the graph structure matrix corresponding to the intermediate feature sequence.
[0068] Accordingly, the GCN-LSTM network can input the intermediate feature sequence and the graph structure matrix into the GCN layer. Since the intermediate feature sequence is obtained by performing unidirectional feature extraction processing on the preset time series through the preset LSTM layer, the intermediate feature sequence can be used to indicate the time dependency between the preset values in the preset time series, and the graph structure matrix belongs to the graph structure data that can be processed by the GCN layer. The graph structure matrix can be used to indicate the spatial relationship between the preset time series corresponding to the intermediate feature sequence. Then, the intermediate feature sequence and the graph structure matrix belong to data under different modes. Based on this, when the GCN layer obtains the intermediate feature sequence and the graph structure matrix corresponding to the intermediate feature sequence, the GCN layer can perform cross-modal feature extraction processing on the intermediate feature sequence and the graph structure matrix, taking into account the spatial relationship between different nodes involved in the graph structure matrix, thereby obtaining the second feature information corresponding to the preset time series.
[0069] Of course, the network architecture of the GCN-LSTM network is not limited to the GCN layer and the preset LSTM layer. The GCN-LSTM network can also be provided with at least one of a random dropout layer and a fully connected layer, which is not limited here.
[0070] Take the preset time series including the time series of financial assets involved in financial business as an example. The GCN-LSTM network can construct the correlation between financial assets into a graph according to the preset architecture size of the preset LSTM layer and the intermediate feature sequence corresponding to the financial asset time series, and obtain the graph structure matrix corresponding to the intermediate feature sequence corresponding to the financial asset time series. Accordingly, the GCN-LSTM network can perform cross-modal feature extraction processing on the intermediate feature sequence and the graph structure matrix through the GCN layer to extract the correlation information between financial assets and obtain the second feature information corresponding to the financial asset time series. Of course, it is not limited to this and is not limited here.
[0071] In an exemplary embodiment, the cross-modal feature extraction process performed by the GCN layer in the GCN-LSTM network on the intermediate feature sequence and the graph structure matrix can be expressed as:
[0072]
[0073] Among them, when l = 0, H l+1 =X,X∈R (N×D) , X is used to indicate the intermediate feature sequence, N is used to indicate the node information corresponding to the graph structure matrix, and D is used to indicate the feature dimension corresponding to the graph structure matrix.
[0074] The GCN-LSTM network can determine the second feature information corresponding to the preset time series based on the output of the GCN layer.
[0075] In this way, when the second feature information corresponding to the preset time series is determined, the second feature information can be used by the time series prediction model to subsequently predict the target prediction result corresponding to the preset time series. Based on the fact that the second feature information is obtained by the GCN-LSTM network performing cross-modal feature extraction processing on the preset time series, the GCN-LSTM network can effectively fuse multimodal feature information. The second feature information can be used to indicate the temporal dependency and spatial relationship between the preset numerical values corresponding to different time points in the preset time series. Accordingly, the time series prediction model can assist in the subsequent prediction of the target prediction result corresponding to the preset time series through the temporal dependency and spatial relationship between the preset numerical values corresponding to different time points in the preset time series indicated by the second feature information, which is conducive to improving the subsequent prediction accuracy of the target prediction result corresponding to the preset time series.
[0076] Step S104: The time series prediction network based on the time series prediction model predicts the preset time series according to the first feature information and the second feature information to obtain a target prediction result corresponding to the preset time series.
[0077] For example, the time series prediction model may be provided with a time series prediction network. When the first feature information and the second feature information corresponding to the preset time series are obtained, the time series prediction model may input the first feature information and the second feature information into the time series prediction network, so that the time series prediction network predicts the preset time series and obtains the corresponding target prediction result.
[0078] In an exemplary embodiment, the network architecture of the time series prediction network may include multiple layers of fully connected layers to perform connection processing on the first feature information and the second feature information to obtain a corresponding target prediction result.
[0079] In this way, the target prediction result is determined by integrating the first feature information and the second feature information. The first feature information and the second feature information can be feature information of the preset time series in different modes. Then, the target prediction result is obtained by predicting the preset time series by the time series prediction network based on the feature information of the preset time series in different modes, which is beneficial to improving the prediction accuracy of the preset time series.
[0080] Exemplarily, based on the time series characteristics of the preset time series, between the preset values of adjacent time points in the preset time series, the previous preset value can be used as the historical value of the next preset value, and the next preset value can be used as the predicted value of the previous preset value. Based on this, the computer device can not only train the time series prediction model through the preset time series, but also verify the time series prediction model through the preset time series, so as to improve the prediction accuracy of the time series prediction model for the target prediction result corresponding to the preset time series.
[0081] In some implementations, the preset time series is divided into a time series to be trained and a time series to be tested.
[0082] For example, the time series to be trained can be used to train a time series prediction model. The time series to be tested can be used to verify the prediction accuracy of the time series prediction model for the time series. For example, the computer device can determine the time series to be trained based on a preset value at a time point in the preset time series where the time interval between the current time point and the time point is greater than or equal to a preset time interval threshold. Correspondingly, the computer device can also determine the time series to be tested based on a preset value at a time point in the preset time series where the time interval between the current time point and the time point is less than a preset time interval threshold. The preset time interval threshold can be pre-set or set by the user, and is not limited here.
[0083] In some implementations, based on a bidirectional long short-term memory network, a bidirectional feature extraction process is performed on the time series to be trained to obtain first feature information corresponding to the time series to be trained.
[0084] For example, when the time series prediction model is trained by the time series to be trained, the time series prediction model needs to learn how to perform bidirectional feature extraction on the time series to be trained. Based on this, the time series prediction model can input the time series to be trained into the BiLSTM network, so that the BiLSTM network performs bidirectional feature extraction on the time series to be trained, and obtains the first feature information corresponding to the time series to be trained.
[0085] The first feature information corresponding to the time series to be trained can be used for subsequent training of the time series prediction model.
[0086] In some embodiments, based on a graph convolution-long short-term memory network, a cross-modal feature extraction process is performed on the time series to be trained to obtain second feature information corresponding to the time series to be trained.
[0087] For example, when training a time series prediction model through a time series to be trained, the time series prediction model needs to learn how to perform cross-modal feature extraction on the time series to be trained. Based on this, the time series prediction model can input the time series to be trained into the GCN-LSTM network, so that the GCN-LSTM network performs cross-modal feature extraction on the time series to be trained, and obtains the second feature information corresponding to the time series to be trained.
[0088] The second feature information corresponding to the time series to be trained can be used for subsequent training of the time series prediction model.
[0089] In some embodiments, based on a time series prediction network, the time series to be trained is predicted according to the first feature information corresponding to the time series to be trained and the second feature information corresponding to the time series to be trained, to obtain a target prediction result corresponding to the time series to be trained.
[0090] For example, when the time series prediction model is trained by the time series to be trained, the time series prediction model needs to learn how to predict the time series to be trained. Based on this, the time series prediction model can input the first feature information and the second feature information corresponding to the time series to be trained into the time series prediction network, so that the time series prediction network predicts the time series to be trained based on the first feature information and the second feature information corresponding to the time series to be trained, and obtains the target prediction result corresponding to the time series to be trained.
[0091] The target prediction results corresponding to the time series to be trained can be used for subsequent training of the time series prediction model.
[0092] In some implementations, model optimization parameters of the time series prediction model are determined based on the target prediction results corresponding to the time series to be trained and the time series to be tested.
[0093] For example, the target prediction result corresponding to the time series to be trained is equivalent to the prediction result of the time series prediction model for the time series to be trained. The time series to be tested is equivalent to the actual result of the time series to be trained. According to the target prediction result corresponding to the time series to be trained and the time series to be tested, the loss value between the actual result and the prediction result of the time series prediction model when determining the target prediction result corresponding to the time series to be trained can be determined.
[0094] The time series prediction model can determine the model optimization parameters of the time series prediction model according to the loss value. The model optimization parameters can be used to subsequently adjust the model parameters of the time series prediction model so that the time series prediction model can more accurately bring the prediction results of the target prediction results corresponding to the training time series or other preset time series of the time series prediction model closer to the actual results, thereby achieving the purpose of training.
[0095] In some implementations, model parameters of the time series prediction model are adjusted according to the model optimization parameters.
[0096] For example, the time series prediction model can adjust the model parameters of the time series prediction model according to the model optimization parameters. For example, the time series prediction model can adjust the network parameters of at least one of the bidirectional long short-term memory network, the graph convolution-long short-term memory network, and the time series prediction network according to the model optimization parameters, which is not limited here.
[0097] By adjusting the model parameters of the time series prediction model, the time series prediction model can more accurately bring the prediction results of the target prediction results corresponding to the time series to be trained or other preset time series closer to the actual results, thereby achieving the purpose of training, which is conducive to improving the prediction accuracy of the time series.
[0098] The time series prediction method provided in the above embodiment obtains a preset time series of a preset object; based on a bidirectional long short-term memory network of a preset time series prediction model, performs bidirectional feature extraction processing on the preset time series to obtain first feature information corresponding to the preset time series; based on a graph convolution-long short-term memory network of the time series prediction model, performs cross-modal feature extraction processing on the preset time series to obtain second feature information corresponding to the preset time series; based on a time series prediction network of the time series prediction model, predicts the preset time series according to the first feature information and the second feature information, and obtains a target prediction result corresponding to the preset time series, so as to improve the prediction accuracy of the time series.
[0099] In the case of obtaining the preset time series of the preset object, the time series prediction model can perform corresponding feature extraction processing on the preset time series through the bidirectional long short-term memory network and the graph convolution-long short-term memory network respectively. Since the graph convolution-long short-term memory network involves cross-modal feature extraction processing on the preset time series, the first feature information and the second feature information can be the feature information of the preset time series in different modes. Accordingly, the time series prediction network can integrate the feature information of the preset time series in different modes to predict the preset time series, which is conducive to improving the prediction accuracy of the preset time series. For example, for the time series of financial business, the time series prediction model can integrate the feature information of the time series in different modes to predict the time series to obtain the corresponding prediction results, which is conducive to improving the prediction accuracy of the time series of financial business, thereby promoting the development of financial business.
[0100] See also Figure 3 , Figure 3 The schematic block diagram of a time series prediction device provided in an embodiment of the present application is shown in FIG. The time series prediction device can be configured in a server or an electronic device to execute the above-mentioned time series prediction method.
[0101] like Figure 3 As shown, the time series prediction device includes: an acquisition module 110, a first feature extraction module 120, a second feature extraction module 130 and a prediction module 140.
[0102] The acquisition module 110 is used to acquire a preset time series of a preset object.
[0103] The first feature extraction module 120 is used to perform bidirectional feature extraction processing on the preset time series based on the bidirectional long short-term memory network of the preset time series prediction model to obtain first feature information corresponding to the preset time series.
[0104] The second feature extraction module 130 is used to perform cross-modal feature extraction processing on the preset time series based on the graph convolution-long short-term memory network of the time series prediction model to obtain second feature information corresponding to the preset time series.
[0105] The prediction module 140 is used to predict the preset time series based on the time series prediction network of the time series prediction model according to the first feature information and the second feature information to obtain a target prediction result corresponding to the preset time series.
[0106] It should be noted that technicians in the relevant field can clearly understand that for the convenience and simplicity of description, the specific working processes of the above-described devices and modules and units can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.
[0107] The method of the present application can be used in many general or special computer system environments or configurations. For example: personal computers, server computers, handheld devices or portable devices, tablet devices, multiprocessor systems, microprocessor-based systems, set-top boxes, programmable consumer electronic devices, network PCs, minicomputers, mainframe computers, distributed computing environments including any of the above systems or devices, etc. The present application can be described in the general context of computer-executable instructions executed by a computer, such as program modules. Generally, program modules include routines, programs, objects, components, data structures, etc. that perform specific tasks or implement specific abstract data types. The present application can also be practiced in distributed computing environments, in which tasks are performed by remote processing devices connected through a communication network. In a distributed computing environment, program modules can be located in local and remote computer storage media including storage devices.
[0108] Exemplarily, the above method and apparatus may be implemented in the form of a computer program, which may be run on a computer device.
[0109] See also Figure 4 , Figure 4 1 is a schematic block diagram of a computer device provided in an embodiment of the present application. The computer device may be a server or an electronic device.
[0110] like Figure 4 As shown, the computer device includes a processor, a memory, and a network interface connected via a system bus, wherein the memory may include a storage medium and an internal memory.
[0111] The storage medium can store an operating system and a computer program. The computer program includes program instructions, and when the program instructions are executed, the processor can execute any time series prediction method.
[0112] The processor is used to provide computing and control capabilities and support the operation of the entire computer equipment.
[0113] The internal memory provides an environment for the operation of the computer program in the storage medium. When the computer program is executed by the processor, the processor can execute any time series prediction method.
[0114] The network interface is used for network communication, such as sending assigned tasks, etc. Those skilled in the art will understand that Figure 4The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.
[0115] It should be understood that the processor may be a central processing unit (CPU), and the processor may also be other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. Among them, the general-purpose processor may be a microprocessor or the processor may also be any conventional processor, etc.
[0116] In one embodiment, the processor is used to execute a computer program and implement the following steps when executing the computer program:
[0117] Get the preset time series of the preset object;
[0118] Based on a bidirectional long short-term memory network of a preset time series prediction model, performing bidirectional feature extraction processing on the preset time series to obtain first feature information corresponding to the preset time series;
[0119] Based on the graph convolution-long short-term memory network of the time series prediction model, cross-modal feature extraction processing is performed on the preset time series to obtain second feature information corresponding to the preset time series;
[0120] Based on the time series prediction network of the time series prediction model, the preset time series is predicted according to the first feature information and the second feature information to obtain a target prediction result corresponding to the preset time series.
[0121] It should be noted that those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working process of the time series prediction described above can refer to the corresponding process in the aforementioned time series prediction method embodiment, and will not be repeated here.
[0122] An embodiment of the present application also provides a computer-readable storage medium, on which a computer program is stored. The method implemented when the computer program is executed by a processor can refer to the various embodiments of the time series prediction method of the present application.
[0123] The computer-readable storage medium may be an internal storage unit of the computer device described in the foregoing embodiment, such as a hard disk or memory of the computer device. The computer-readable storage medium may also be an external storage device of the computer device, such as a plug-in hard disk, a smart memory card (SmartMedia Card, SMC), a secure digital (Secure Digital, SD) card, a flash card (Flash Card), etc., equipped on the computer device.
[0124] It should be understood that the terms used in this application specification are only for the purpose of describing specific embodiments and are not intended to limit the application. As used in this application specification and the appended claims, unless the context clearly indicates otherwise, the singular forms "a", "an" and "the" are intended to include plural forms.
[0125] It should also be understood that the term "and / or" used in the specification of this application and the appended claims refers to any combination of one or more of the associated listed items and all possible combinations, including these combinations. It should be noted that, in this article, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or system including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or system. In the absence of further restrictions, an element defined by the sentence "including a..." does not exclude the presence of other identical elements in the process, method, article or system including the element.
[0126] It should also be understood that the non-Company software tools or components appearing in the embodiments of the present application are merely examples and do not represent actual use.
[0127] The serial numbers of the embodiments of the present application are for description only and do not represent the advantages and disadvantages of the embodiments. The above description is only a specific implementation mode of the present application, but the protection scope of the present application is not limited thereto. Any technician familiar with the technical field can easily think of various equivalent modifications or replacements within the technical scope disclosed in the present application, and these modifications or replacements should be included in the protection scope of the present application. Therefore, the protection scope of the present application shall be based on the protection scope of the claims.
Claims
1. A time series prediction method, characterized in that: include: Get the preset time series of the preset object; Based on a bidirectional long short-term memory network of a preset time series prediction model, performing bidirectional feature extraction processing on the preset time series to obtain first feature information corresponding to the preset time series; Based on the graph convolution-long short-term memory network of the time series prediction model, cross-modal feature extraction processing is performed on the preset time series to obtain second feature information corresponding to the preset time series; Based on the time series prediction network of the time series prediction model, the preset time series is predicted according to the first feature information and the second feature information to obtain a target prediction result corresponding to the preset time series.
2. The time series prediction method according to claim 1, characterized in that: The graph convolution-long short-term memory network based on the time series prediction model performs cross-modal feature extraction processing on the preset time series to obtain second feature information corresponding to the preset time series, including: Performing unidirectional feature extraction processing on the preset time series through a preset long short-term memory layer of the graph convolution-long short-term memory network to obtain an intermediate feature sequence corresponding to the preset time series; The intermediate feature sequence is subjected to cross-modal feature extraction processing through the graph convolution layer of the graph convolution-long short-term memory network to obtain the second feature information.
3. The time series prediction method according to claim 2, characterized in that: The step of performing cross-modal feature extraction processing on the intermediate feature sequence through the graph convolution layer of the graph convolution-long short-term memory network to obtain the second feature information includes: Determining a graph structure matrix corresponding to the intermediate feature sequence according to a preset architecture size of the preset long short-term memory layer and the intermediate feature sequence; Through the graph convolution layer, cross-modal feature extraction processing is performed on the intermediate feature sequence and the graph structure matrix to obtain the second feature information.
4. The time series prediction method according to claim 3, characterized in that: The step of determining the graph structure matrix corresponding to the intermediate feature sequence according to the preset architecture size of the preset long short-term memory layer and the intermediate feature sequence includes: Determining node information corresponding to the graph structure matrix according to the preset architecture size; Determining a feature dimension corresponding to the graph structure matrix according to the intermediate feature sequence; The graph structure matrix is determined according to the node information and the feature dimension.
5. The time series prediction method according to claim 1, characterized in that: The bidirectional long short-term memory network based on the preset time series prediction model performs bidirectional feature extraction processing on the preset time series to obtain first feature information corresponding to the preset time series, including: Performing forward feature extraction processing on the preset time series through the forward long short-term memory layer of the bidirectional long short-term memory network to obtain forward feature information corresponding to the preset time series; Performing reverse feature extraction processing on the preset time series through the reverse long short-term memory layer of the bidirectional long short-term memory network to obtain reverse feature information corresponding to the preset time series; The forward feature information and the reverse feature information are combined to obtain the first feature information.
6. The time series prediction method according to any one of claims 1 to 5, characterized in that: After obtaining the preset time sequence of the preset object, the method further includes: Dividing the preset time series into a time series to be trained and a time series to be tested; The bidirectional long short-term memory network based on the preset time series prediction model performs bidirectional feature extraction processing on the preset time series to obtain first feature information corresponding to the preset time series, including: Based on the bidirectional long short-term memory network, bidirectional feature extraction processing is performed on the time series to be trained to obtain first feature information corresponding to the time series to be trained; The graph convolution-long short-term memory network based on the time series prediction model performs cross-modal feature extraction processing on the preset time series to obtain second feature information corresponding to the preset time series, including: Based on the graph convolution-long short-term memory network, cross-modal feature extraction processing is performed on the time series to be trained to obtain second feature information corresponding to the time series to be trained; The time series prediction network based on the time series prediction model predicts the preset time series according to the first feature information and the second feature information to obtain a target prediction result corresponding to the preset time series, including: Based on the time series prediction network, predicting the time series to be trained according to the first feature information corresponding to the time series to be trained and the second feature information corresponding to the time series to be trained, to obtain a target prediction result corresponding to the time series to be trained; After the time series prediction network based on the time series prediction model predicts the preset time series according to the first feature information and the second feature information to obtain a target prediction result corresponding to the preset time series, the method further includes: Determining the model optimization parameters of the time series prediction model according to the target prediction result corresponding to the time series to be trained and the time series to be tested; According to the model optimization parameters, the model parameters of the time series prediction model are adjusted.
7. The time series prediction method according to any one of claims 1 to 5, characterized in that: After obtaining the preset time sequence of the preset object, the method further includes: The preset numerical values corresponding to each time point in the preset time series are normalized to obtain a processed preset time series.
8. A time series prediction device, characterized in that: The time series prediction device comprises: An acquisition module, used to acquire a preset time series of a preset object; A first feature extraction module is used to perform bidirectional feature extraction processing on the preset time series based on a bidirectional long short-term memory network of a preset time series prediction model to obtain first feature information corresponding to the preset time series; A second feature extraction module is used to perform cross-modal feature extraction processing on the preset time series based on the graph convolution-long short-term memory network of the time series prediction model to obtain second feature information corresponding to the preset time series; A prediction module is used to predict the preset time series based on the time series prediction network of the time series prediction model according to the first feature information and the second feature information to obtain a target prediction result corresponding to the preset time series.
9. A computer device, characterized in that: The computer device includes a memory and a processor; The memory is used to store computer programs; The processor is used to execute the computer program and implement the time series prediction method according to any one of claims 1 to 7 when executing the computer program.
10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the time series prediction method according to any one of claims 1 to 7 are implemented.