Time sequence prediction method and device for business data, equipment, medium and program product

Abstract

The invention provides a time sequence prediction method for business data, which can be applied to the field of artificial intelligence. The time sequence prediction method comprises the following steps: performing variational mode decomposition on service time sequence data to obtain a mode feature sequence corresponding to the service time sequence data; the business time sequence data comprises business data before a date to be tested; performing feature extraction on the modal feature sequence corresponding to the service time sequence data to obtain principal component data corresponding to the service time sequence data; constructing to-be-tested feature data based on a service index and principal component data corresponding to the service time sequence data; and on the basis of the to-be-tested feature data, predicting a service data result of the to-be-tested date by using a pre-trained long-short-term memory network model. The invention further provides a time sequence prediction device and equipment for the business data, a storage medium and a program product.

Description

Technical Field

[0001] The present application relates to the field of artificial intelligence, specifically to the field of machine learning and the field of financial technology, and more specifically to a method, device, equipment, medium and program product for time series prediction of business data. Background Art

[0002] In financial scenarios, there are many types of business data that need to be predicted, such as market data, institutional business data, and customer behavior data. The prediction of business data is crucial for financial institutions' risk management, business decision-making, and market analysis. In the process of digital transformation in the financial industry, financial institutions have an increasingly urgent need to accurately predict business data. Taking the dynamic prediction of bank deposits and loans in core business scenarios as an example, accurately predicting bank deposit and loan business data is of strategic significance to bank fund management and risk prevention and control: on the one hand, accurate deposit and loan business forecasts can support banks in scientifically allocating funds, dynamically balancing the supply and demand of funds, and avoiding liquidity gaps or redundancies; on the other hand, this forecasting mechanism can help banks proactively manage credit risk, market risk, and liquidity risk, formulate response strategies in advance through quantitative analysis, and build a solid and stable operating defense line.

[0003] However, current time series prediction models for business data in the financial sector generally suffer from insufficient accuracy, making it difficult to meet the needs of refined management in complex market environments. There is an urgent need to improve the accuracy and timeliness of predictions. Summary of the Invention

[0004] In view of the above problems, the present application provides a method, apparatus, device, medium and program product for time series prediction of business data to improve prediction accuracy.

[0005] According to the first aspect of the present application, a time series prediction method for business data is provided, comprising: performing variational modal decomposition on the business time series data to obtain a modal feature sequence corresponding to the business time series data; the business time series data includes business data before a date to be tested; performing feature extraction on the modal feature sequence corresponding to the business time series data to obtain principal component data corresponding to the business time series data; constructing feature data to be tested based on a business index and the principal component data corresponding to the business time series data; and predicting the business data results of the date to be tested using a pre-trained long short-term memory network model based on the feature data to be tested.

[0006] According to an embodiment of the present application, the characteristic data to be measured is constructed based on the business index and the principal component data corresponding to the business time series data, including: using the business index and the principal component data corresponding to the business time series data as the characteristic data stream, and based on the prediction time step, using a sliding window to process the characteristic data stream to construct the characteristic data to be measured; wherein the size of the sliding window is determined based on the prediction time step.

[0007] According to an embodiment of the present application, based on the feature data to be tested, a pre-trained long short-term memory network model is used to predict the business data results of the date to be tested, including: standardizing the feature data to be tested; and inputting the standardized feature data to be tested into the pre-trained long short-term memory network model for prediction, and outputting the business data results of the date to be tested.

[0008] According to an embodiment of the present application, the standardization of the feature data to be measured includes: calculating the maximum value and the minimum value corresponding to each column of features based on the feature data to be measured; and mapping each feature data value in the feature data to be measured to a preset numerical range using a linear transformation based on the maximum value and the minimum value.

[0009] According to an embodiment of the present application, the feature extraction of the modal feature sequence corresponding to the business time series data to obtain the principal component data corresponding to the business time series data includes: mean-normalizing the modal feature sequence corresponding to the business time series data; calculating the covariance matrix based on the mean-normalized modal feature sequence; calculating the cumulative variance contribution rate according to the eigenvalues ​​of the covariance matrix; and determining the principal component data corresponding to the business time series data according to the cumulative variance contribution rate and the mean-normalized modal feature sequence.

[0010] According to an embodiment of the present application, the long short-term memory network model includes a long short-term memory network unit based on the prediction time step; the long short-term memory network unit includes a forget gate, an input gate and an output gate; the forget gate is used to receive historical hidden states and control the retention ratio of current input data; the input gate is used to filter the valid information of the current input data; and the output gate is used to generate the current hidden state.

[0011] The second aspect of the present application provides a time series prediction device for business data, including: a modal decomposition module, used to perform variational modal decomposition on the business time series data to obtain a modal feature sequence corresponding to the business time series data; the business time series data includes business data before a date to be tested; a feature extraction module, used to perform feature extraction on the modal feature sequence corresponding to the business time series data to obtain principal component data corresponding to the business time series data; a sample construction module, used to construct the feature data to be tested based on the business index and the principal component data corresponding to the business time series data; and a prediction module, used to predict the business data results of the date to be tested based on the feature data to be tested using a pre-trained long short-term memory network model.

[0012] According to an embodiment of the present application, the sample construction module includes: a sliding window unit, which is used to use the principal component data corresponding to the business index and the business time series data as a feature data stream, and based on the prediction time step, use a sliding window to process the feature data stream to construct the feature data to be tested; wherein, the size of the sliding window is determined based on the prediction time step.

[0013] According to an embodiment of the present application, the prediction module includes: a standardization unit for standardizing the feature data to be tested; and a prediction unit for inputting the standardized feature data to be tested into the pre-trained long short-term memory network model for prediction, and outputting the business data results of the date to be tested.

[0014] According to an embodiment of the present application, the standardization unit includes: a maximum value calculation subunit, which is used to calculate the maximum value and the corresponding minimum value corresponding to each column of features based on the feature data to be measured; and a mapping subunit, which is used to map each feature data value in the feature data to be measured to a preset numerical range based on the maximum value and the minimum value using linear transformation.

[0015] According to an embodiment of the present application, the feature extraction module includes: a mean processing unit for performing mean standardization on the modal feature sequence corresponding to the business time series data; a covariance calculation unit for calculating the covariance matrix based on the modal feature sequence after mean standardization; a contribution rate calculation unit for calculating the cumulative variance contribution rate according to the eigenvalues ​​of the covariance matrix; and a principal component determination unit for determining the principal component data corresponding to the business time series data based on the cumulative variance contribution rate and the modal feature sequence after mean standardization.

[0016] According to an embodiment of the present application, the device also includes a model processing module, the model processing module includes the long short-term memory network model, the long short-term memory network model includes a long short-term memory network unit based on the prediction time step; the long short-term memory network unit includes a forget gate, an input gate and an output gate; the forget gate is used to receive historical hidden states and control the retention ratio of current input data; the input gate is used to filter the valid information of the current input data; and the output gate is used to generate the current hidden state.

[0017] The third aspect of the present application provides an electronic device, comprising: one or more processors; a memory for storing one or more computer programs, wherein the one or more processors execute the one or more computer programs to implement the steps of the above method.

[0018] The fourth aspect of the present application further provides a computer-readable storage medium having a computer program or instructions stored thereon, which implements the steps of the above method when the computer program or instructions are executed by a processor.

[0019] The fifth aspect of the present application further provides a computer program product, comprising a computer program or instructions, which implement the steps of the above method when executed by a processor. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] The above contents and other objects, features and advantages of the present application will become more apparent through the following description of the embodiments of the present application with reference to the accompanying drawings, in which:

[0021] Figure 1 A diagram schematically illustrates an application scenario of a method, apparatus, device, medium, and program product for predicting time series of business data according to an embodiment of the present application;

[0022] Figure 2 A flowchart of a method for time series prediction of business data according to an embodiment of the present application is schematically shown;

[0023] Figure 3 A diagram schematically illustrates a time series prediction model of a method for predicting time series of business data according to an embodiment of the present application;

[0024] Figure 4 A sliding window diagram of deposit data according to a time series prediction method for business data according to an embodiment of the present application is schematically shown;

[0025] Figure 5 Another flowchart of the method for time series prediction of business data according to an embodiment of the present application is schematically shown;

[0026] Figure 6A block diagram schematically illustrates a structure of a device for predicting time series of business data according to an embodiment of the present application; and

[0027] Figure 7 A block diagram of an electronic device suitable for implementing a method for time series prediction of business data according to an embodiment of the present application is schematically shown. DETAILED DESCRIPTION

[0028] Hereinafter, embodiments of the present application will be described with reference to the accompanying drawings. However, it should be understood that these descriptions are exemplary only and are not intended to limit the scope of the present application. In the detailed description below, for ease of explanation, many specific details are set forth to provide a comprehensive understanding of the embodiments of the present application. However, it is obvious that one or more embodiments may also be implemented without these specific details. In addition, in the following description, descriptions of known structures and technologies are omitted to avoid unnecessarily confusing the concepts of the present application.

[0029] The terms used herein are only for describing specific embodiments and are not intended to limit the present application. The terms "comprise," "include," etc. used herein indicate the presence of features, steps, operations, and / or components, but do not exclude the presence or addition of one or more other features, steps, operations, or components.

[0030] All terms used herein (including technical and scientific terms) have the meanings commonly understood by those skilled in the art unless otherwise defined. It should be noted that the terms used herein should be interpreted as having a meaning consistent with the context of this specification and should not be interpreted in an idealized or overly rigid manner.

[0031] When expressions such as "at least one of A, B, and C, etc." are used, they should generally be interpreted in accordance with the meaning commonly understood by those skilled in the art (for example, "a system having at least one of A, B, and C" should include but is not limited to a system having A alone, B alone, C alone, A and B, A and C, B and C, and / or A, B, C, etc.).

[0032] The embodiment of the present application provides a time series prediction method for business data, which performs variational modal decomposition on the business time series data to obtain a modal feature sequence corresponding to the business time series data; the business time series data includes business data before a date to be measured; the modal feature sequence corresponding to the business time series data is subjected to feature extraction to obtain principal component data corresponding to the business time series data; based on the business index and the principal component data corresponding to the business time series data, the feature data to be measured is constructed; and based on the feature data to be measured, a pre-trained long short-term memory network model is used to predict the business data results of the date to be measured. After the variational modal decomposition, the principal component data is extracted, and the feature data to be measured is constructed in combination with the business index, the business data features are accurately extracted, and the long short-term memory network model is used to predict the business data results. The business data results can be accurately predicted, the accuracy and timeliness of the business data prediction are improved, and strong data support is provided for finance. Reasonable business policies are formulated based on accurate prediction data to effectively promote business development.

[0033] Figure 1 The application scenario diagram of the time series prediction method, device, equipment, medium and program product of business data according to the embodiments of the present application is schematically shown.

[0034] like Figure 1 As shown, the application scenario 100 according to this embodiment may include a first terminal device 101, a second terminal device 102, a third terminal device 103, and a server 105. A network 104 is used as a medium for providing a communication link between the first terminal device 101, the second terminal device 102, the third terminal device 103, and the server 105. The network 104 may include various connection types, such as wired or wireless communication links or optical fiber cables.

[0035] A user may use a first terminal device 101, a second terminal device 102, or a third terminal device 103 to interact with a server 105 via a network 104 to receive or send messages, etc. Various communication client applications may be installed on the first terminal device 101, the second terminal device 102, or the third terminal device 103, such as shopping applications, web browser applications, search applications, instant messaging tools, email clients, social platform software, etc. (for example only).

[0036] The first terminal device 101 , the second terminal device 102 , and the third terminal device 103 may be various electronic devices having display screens and supporting web browsing, including but not limited to smart phones, tablet computers, laptop computers, desktop computers, and the like.

[0037] The server 105 may be a server that provides various services, such as a background management server (for example only) that supports websites browsed by users using the first terminal device 101, the second terminal device 102, and the third terminal device 103. The background management server may analyze and process received data such as user requests, and feed back processing results (e.g., web pages, information, or data obtained or generated based on user requests) to the terminal devices.

[0038] It should be noted that the time series prediction method for business data provided in the embodiment of the present application can generally be executed by the server 105. Accordingly, the time series prediction device for business data provided in the embodiment of the present application can generally be set in the server 105. The time series prediction method for business data provided in the embodiment of the present application can also be executed by a server or server cluster that is different from the server 105 and can communicate with the first terminal device 101, the second terminal device 102, the third terminal device 103 and / or the server 105. Accordingly, the time series prediction device for business data provided in the embodiment of the present application can also be set in a server or server cluster that is different from the server 105 and can communicate with the first terminal device 101, the second terminal device 102, the third terminal device 103 and / or the server 105.

[0039] It should be understood that Figure 1 The number of terminal devices, networks and servers in the embodiment is merely illustrative. Any number of terminal devices, networks and servers may be provided as required.

[0040] The following will be based on Figure 1 The scene described by Figures 2 to 5 A method for predicting the time series of business data according to an embodiment of the present application is described in detail.

[0041] Figure 2 The flowchart of the time series prediction method of business data according to an embodiment of the present application is schematically shown.

[0042] like Figure 2 As shown in FIG. , the time series prediction method for business data of this embodiment includes operations S210 to S240. The time series prediction method is not limited to a specific execution subject. The execution subject can be any electronic device, such as a terminal device or a server device. The execution subject can also be any software application or client.

[0043] In operation S210 , variational modal decomposition is performed on the business time series data to obtain a modal feature sequence corresponding to the business time series data; the business time series data includes business data before the date to be tested.

[0044] Business data refers to structured or semi-structured data related to business activities, recording the status, changes, or results of the business at different points in time. Business data includes, but is not limited to, market transaction data (such as real-time historical data such as trading volume and order flow), institutional business data (such as credit records, deposits and loans, premium income and expenditure, fund net value, and other business flow data), and user and risk data (such as asset changes, credit scores, user transaction behavior data, etc.).

[0045] Business time series data represents time-stamped business data arranged in chronological order, including business data prior to the test date, i.e., historical business data recorded before the test date. The specific business data required can be determined based on the prediction time step. For example, if the prediction time step is set to t, then at least business data from t days ago must be obtained. Using the prediction method in this application, the business data results for day t can be predicted. The choice of the prediction time step (e.g., daily, weekly, monthly, quarterly, etc.) can be determined based on a comprehensive consideration of business needs, data characteristics, and model features.

[0046] Performing variational modal decomposition on business time series data can obtain n modal feature sequences corresponding to the business time series data (such as modal feature sequences IMF1, IMF2, ...IMF n ), n is at least greater than 1, and variational mode decomposition (VMD) is introduced to achieve feature reconstruction of business time series data. By decomposing the features of business time series data and extracting the main components, it can adaptively process time series, achieve data stabilization and extract deep-level fluctuation characteristics, thereby improving the prediction accuracy of the model. It should be noted that VMD is a signal decomposition algorithm that transforms the original business data into Decomposed into n IMF components , that is, the goal of the variational problem is to find n central frequencies The modal components of The basic assumption is that each modal component With finite bandwidth, the variational problem is solved by finding the modal function with the smallest sum of n estimated bandwidths. VMD can effectively solve the modal aliasing problem of traditional empirical mode decomposition (EMD) and more effectively reduce the noise of data features.

[0047] It should be noted that in the technical solution of this application, the user information (including but not limited to user personal information, user image information, user device information, such as location information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in the business data are all information and data authorized by the user or fully authorized by all parties, and the collection, storage, use, processing, transmission, provision, disclosure and application of the relevant data comply with relevant laws, regulations and standards, take necessary confidentiality measures, do not violate public order and good morals, and provide corresponding operation entrances for users to choose to authorize or refuse.

[0048] In the scenario of using personal information for automated decision-making, the methods, devices, and systems provided in the embodiments of the present application all provide users with corresponding operation portals for users to choose to agree or reject the automated decision-making results; if the user chooses to reject, the expert decision-making process will be entered. The expression "automated decision-making" here refers to the activity of automatically analyzing and evaluating an individual's behavioral habits, interests and hobbies, or economic, health, credit status, etc. through computer programs and making decisions. The expression "expert decision-making" here refers to the activity of making decisions by people who specialize in a certain field, have specialized experience, knowledge and skills, and have reached a certain level of professionalism.

[0049] In operation S220 , feature extraction is performed on the modal feature sequence corresponding to the business time series data to obtain principal component data corresponding to the business time series data.

[0050] After VMD decomposition, business time series data has a high feature dimension. Each modal feature sequence includes the modal component functions of each feature. If the modal feature sequence is used directly for model prediction, when the feature dimension of the input data is too high, the model will face a heavy computational burden and may encounter problems such as dimensionality explosion, which will hinder the training process and slow prediction. Therefore, to improve the model's computational efficiency, it is necessary to perform feature dimensionality reduction on the modal feature sequence, eliminating redundant information and retaining only a small number of principal components that carry the key information.

[0051] Modal feature sequences are used as features corresponding to business time series data, but not all features are relevant to the prediction target, so further extraction of key features is necessary. Principal component analysis (PCA) can be used to extract key features. PCA can filter out features that carry key information, effectively removing most noise in time series data and features that have little impact on the output. It also maintains the independence of each feature, reducing the computational complexity of the long-short-term memory network model and improving prediction accuracy.

[0052] In operation S230 , feature data to be measured is constructed based on the business index and the principal component data corresponding to the business time series data.

[0053] Business index refers to macro-characteristic index data related to business data. For example, for user transaction behavior (such as activity, retention, conversion, etc.), you can choose business indices related to transaction scale, such as the total merchandise transaction volume index and the average order value index. For asset change data, you can choose business indices related to consumption scale, such as the consumer confidence index and the inflation rate.

[0054] Using the business index and the extracted principal component data as the feature data of the long short-term memory network model, the business index features and business data features are integrated, the feature dimensions are enriched, and the long short-term memory network model can enhance the accuracy of capturing and predicting business logic.

[0055] In operation S240 , based on the feature data to be tested, the business data result of the date to be tested is predicted using the pre-trained long short-term memory network model.

[0056] According to an embodiment of the present application, the long short-term memory network model includes a long short-term memory network unit based on a prediction time step; the long short-term memory network unit includes a forget gate, an input gate, and an output gate; the forget gate is used to receive historical hidden states and control the retention ratio of current input data; the input gate is used to filter valid information of the current input data; and the output gate is used to generate the current hidden state.

[0057] The Long Short-Term Memory (LSTM) network model is composed of a long short-term memory unit that transmits information, including a forget gate, an input gate, and an output gate.

[0058] The forget gate controls the retention ratio of the current input data, determining which information should be discarded from the cell state. A sigmoid layer (a neuron layer using the upper sigmoid activation function) is used to output a value between 0 and 1, where 0 represents "completely forgotten" and 1 represents "completely retained." In this way, the LSTM model discards old, no longer important information to make room for new information. Its implementation formula is as follows:

[0059] ;

[0060] in, Expand to , the weight in the above formula Not shared, is the sigmoid activation function, Represents the weight between the forget gate of this layer and the previous hidden layer, Represents the weight between the forget gate of this layer and the input layer of this layer, is the bias of the forget gate, Represents the historical hidden state, Indicates the current input data, Represents the forgetting coefficient.

[0061] The input gate filters the valid information of the current input data, and its function is to control the current input data should be added to the cell state It consists of two parts: a sigmoid neural network layer and a dot product operation. The sigmoid layer outputs a value between 0 and 1, indicating how much of each state value should be updated, and the dot product operation applies this proportion to the current state. Its implementation formula is as follows:

[0062] ;

[0063] ;

[0064] in, is the weight of the input gate. Like the forget gate, the weight is not shared. is the bias of the input gate, is the input weight, is the candidate cell state, and tanh is the tanh activation function.

[0065] After calculation of the forget gate and input gate, the old cell state Multiply the forgetting coefficient , discard the information that needs to be forgotten, and then add , add information that needs to be remembered, and use the cell state updater to obtain the cell state , the formula is:

[0066] ;

[0067] Output gate determines cell state Which information in will be used to calculate the output of the current time step, that is, output the current hidden state First, the cell state passes through a tanh layer (a neuron layer that uses the hyperbolic tangent function tanh as the activation function) to compress the value to between -1 and 1 and multiply it with the output of the sigmoid layer so that only important information is output. The implementation formula is as follows:

[0068] ;

[0069] ;

[0070] in, Denotes the weight parameter. Similar to the forget gate and the input gate, the weights are not shared. Denotes the output gate bias parameter. Denotes the output weight.

[0071] By introducing a gating mechanism and multiple computational units, the LSTM model significantly enhances its ability to retain, forget, and process memory of information, giving it an obvious advantage in processing long time series data. At each time step, the LSTM model updates its cell state by controlling the way information flows through these gates and generates new outputs, enabling the LSTM model to still maintain the memory of early inputs when the sequence is long.

[0072] In the embodiments of the present application, the long short-term memory network model significantly enhances its ability to retain, forget, and process memory of information by introducing a gating mechanism, has an obvious advantage in processing long time series data, and is highly suitable for processing business time series data.

[0073] The time series prediction method of the present application is implemented through a time series prediction model. Figure 3 Schematically shows a time series prediction model diagram of the time series prediction method for business data according to the embodiments of the present application.

[0074] As Figure 3 shown, the business time series data is decomposed into n modal feature sequences IMF (IMF 1… IMFn) by the VMD algorithm, and feature extraction is performed by the PCA algorithm to obtain m extracted principal component data ( ) where m < n. Based on the principal component data and the business index, feature data is determined, and a sliding window is used to construct the待测 feature data (X 1… X t ). X1 represents all the feature data on the first day, and Xt contains all the feature data in the past t days. The待测 feature data is input into the trained LSTM model to obtain the business data prediction result output.

[0075] The training process of the time series prediction model is as follows: By obtaining long-term historical business data, the historical business data is subjected to variational mode decomposition to obtain the historical modal feature sequences corresponding to the historical business data; feature extraction is performed on the historical modal feature sequences to obtain the historical principal component data corresponding to the historical business data; based on the historical business index and the principal component data corresponding to the historical business data, a historical待测 feature data set is constructed; based on the historical待测 feature data set and the actual business data on the corresponding date, the long short-term memory network model is trained.

[0076] Taking business data as an example of bank deposit and loan data, the training process of the time series prediction model is as follows: Obtain the bank's historical deposit and loan data in recent years, construct the original dataset of the bank's historical deposits as historical deposit data. Reconstruct the historical deposit data into n intrinsic mode functions (modal feature sequences IMF) through the VMD algorithm, and then perform feature extraction through the PCA algorithm to extract m principal component data, where m < n. The deposit and loan data is related to the current market price situation, and the consumer price index is selected as the business index. Use the principal component data and the business index as the input features of the sliding window, and use the sliding window technology to extract feature data and construct the historical dataset. Taking the deposit data as an example, Figure 4 Schematically shows a sliding window diagram of deposit data for the time series prediction method of business data according to an embodiment of the present application. When constructing the historical dataset, assume that in the input data stream, the subscript represents the time point when the sample point enters the sliding window. Let the window size be t + 1, and a sample set consists of t + 1 data points. The first t data points in the sample set are used as the input features (X_train) for training the LSTM model, and the value of the (t + 1)-th data point is used as the prediction target (y_train) of the sample set. By using the feature data of the previous t days to predict the deposit and loan amounts on the (t + 1)-th day, where t represents the selected time step. The historical dataset can be divided into a training set and a test set in a ratio of 8:2. Use the training set to train the LSTM model and use the validation set to test the LSTM model. The mean squared error (MSE) is used as the evaluation index for training the LSTM model. MSE refers to the mean of the squared errors between the model's predicted values and the target values, and its formula is as follows:

[0077] ;

[0078] where, is the th true value in the sample, is the th predicted value of the model, is the number of samples. The training ends when MSE is less than the empirical value (such as 0.01).

[0079] In the embodiment of the present application, after variational mode decomposition, principal component data is extracted, combined with the business index to construct the待测特征数据 (to-be-tested feature data), accurately extract the business data features, and use the long short-term memory network model to predict the business data results, which can accurately predict the business data results, improve the accuracy and timeliness of business data prediction, provide strong data support for finance, formulate reasonable business policies based on accurate prediction data, and effectively promote business development.

[0080] According to an embodiment of the present application, in operation S220 , feature extraction is performed on the modal feature sequence corresponding to the business time series data to obtain the principal component data corresponding to the business time series data, including operations S2201 to S2204 .

[0081] In operation S2201, the modal feature sequence corresponding to the business time series data is mean-normalized.

[0082] The decomposed business time series data includes For a modal feature sequence, mean normalization can calculate the mean of each feature (feature sequence) and then remove the mean. The formula is as follows:

[0083] ;

[0084] in, is the i-th feature, that is, the modal feature sequence after mean normalization; is the i-th original feature, i.e., the modal feature sequence; is the mean of the i-th feature, that is, the mean of the modal feature sequence, .

[0085] In operation S2202 , a covariance matrix is ​​calculated based on the mean-normalized modal feature sequence.

[0086] Covariance matrix The calculation formula is as follows:

[0087] ;

[0088] in, For the The variance of the features, For the Features and The correlation coefficient of the features, For the Features and The standard deviation of the features, For the Features and The covariance of the features.

[0089] In operation S2203 , the cumulative variance contribution rate is calculated according to the eigenvalues ​​of the covariance matrix.

[0090] Pair covariance matrix Performing eigendecomposition, we can calculate the eigenvalues ​​of the covariance matrix Σ, eigenvalues Greater than 1, The eigenvalues ​​are arranged from large to small. , and the corresponding eigenvectors v1, v2, …, v n , obtaining principal components. The variance contribution rate of the i-th principal component = . Calculate the sum of the variance contribution rates of the first m principal components, and then obtain the cumulative variance contribution rate of the m-th principal component . The formula is:

[0091] ;

[0092] where is the i-th eigenvalue.

[0093] In operation S2204, based on the cumulative variance contribution rate and the mode feature sequence after mean normalization, determine the principal component data corresponding to the business time series data.

[0094] Select the first m principal components with high contribution rates from large to small according to the cumulative variance contribution rate, where m < k, and the selection conditions are as follows:

[0095] ;

[0096] Each eigenvalue corresponds to an eigenvector v1, v2, …, v n . Based on the selected first m principal components , for the corresponding first k eigenvectors, arrange the eigenvectors by column to form an eigenvector matrix V, and project the mode feature sequence matrix E into the eigenvector space to obtain the principal component matrix X. The formula is:

[0097] ;

[0098] where is the mode feature sequence matrix after mean normalization. Each column of the principal component matrix X represents a principal component, and its variance is equal to the corresponding eigenvalue .

[0099] In the embodiments of the present application, principal component analysis is used to perform feature dimensionality reduction on the mode feature sequence, eliminate redundant information, and only retain a small number of principal components carrying the main information, further improving the calculation efficiency.

[0100] According to the embodiments of the present application, in operation S230, based on the business index and the principal component data corresponding to the business time series data, construct the待测特征数据 (to-be-detected feature data), including: using the business index and the principal component data corresponding to the business time series data as the feature data stream, and based on the prediction time step, using a sliding window to process the feature data stream to construct the to-be-detected feature data; where the size of the sliding window is determined based on the prediction time step.

[0101] The principal component data corresponding to the business index and business time series data is used as the feature data stream. That is, the business index and principal component data are used as features. When constructing the feature data to be tested, based on the input feature data stream, the feature data represents the sample point, and the subscript represents the time point when the sample point enters the sliding window. The prediction time step is t, and the window size is t+1. The window is moved successively by the window size, each time including the current sample point and the previous t historical sample points (a total of t+1 data). The previous t points can be used to predict the t+1 point (i.e., t steps in the future), realizing recursive prediction of time series data. The constructed feature data X to be tested is as follows:

[0102] ;

[0103] in, represents the jth feature of the i-th day, Represents the feature vector of the i-th day, j=1,2,3…q, represents the feature; i=1,2,3…t, represents the time.

[0104] In an embodiment of the present application, a sliding window technology is used to process the features corresponding to the time series data, and the data in the window is continuously refreshed in the feature data stream, and the feature data is updated in real time to ensure the time series of the feature data to be tested, so as to facilitate the subsequent entry into the long short-term memory network model for prediction.

[0105] Figure 5 Another flowchart of the method for time series prediction of business data according to an embodiment of the present application is schematically shown.

[0106] like Figure 5 As shown, in operation S240, based on the feature data to be tested, the business data result of the date to be tested is predicted using the pre-trained long short-term memory network model, including operations S510 to S520.

[0107] In operation S510 , the feature data to be measured is normalized.

[0108] Before the feature data is input into the LSTM model, it is necessary to standardize the feature data to ensure that all feature sequences are compared and learned at the same scale.

[0109] In the embodiment of the present application, the feature data to be measured is standardized so that the feature data are distributed on the same scale, thereby avoiding the model being biased towards large-value features and improving the prediction effect.

[0110] According to an embodiment of the present application, the operation S510 of normalizing the characteristic data to be measured includes operations S5101 and S5102.

[0111] In operation S5101 , the maximum value and the minimum value corresponding to each column of features are calculated based on the feature data to be measured.

[0112] In operation S5102 , each feature data value in the feature data to be measured is mapped to a preset value range using linear transformation based on the maximum value and the minimum value.

[0113] The Min-Max normalization method is used to map each eigenvalue to a preset value range (such as the range between 0 and 1) through linear transformation. The formula for the Min-Max normalization is as follows:

[0114] ;

[0115] in, and is the maximum and minimum value in each column feature. is the original eigenvalue to be transformed, is the characteristic data value after normalization.

[0116] In an embodiment of the present application, based on the maximum and minimum values ​​of each column of features, a linear transformation is used to map each feature data value in the feature data to be measured to a preset numerical range, so that the feature data is scaled to a uniform range, eliminating the dimensionality effect, accelerating model convergence, and improving prediction accuracy and stability.

[0117] In operation S520 , the standardized feature data to be tested is input into a pre-trained long short-term memory network model for prediction, and the business data result of the date to be tested is output.

[0118] LSTM models can effectively predict business data results for a specified date. Accurately predicting business data can provide strong data support for financial institutions. For example, accurately predicting deposit and loan data for a specific date can help banks formulate appropriate loan policies, such as determining loan size, loan structure, and loan interest rates, to better meet market demand and promote business development. Furthermore, these predictions can guide banks in adjusting deposit interest rates to attract and retain depositors and strengthen their financial strength. Through these precise prediction-based decisions, banks can significantly improve capital utilization efficiency, optimize asset allocation, reduce operating costs, and enhance asset quality, thereby gaining a competitive advantage in the market and achieving a virtuous cycle of cost reduction and efficiency improvement.

[0119] Based on the above-mentioned time series prediction method for business data, this application also provides a time series prediction device for business data. Figure 6 The device is described in detail.

[0120] Figure 6 The structural block diagram of the time series prediction device for business data according to an embodiment of the present application is schematically shown.

[0121] like Figure 6 As shown, the time series prediction device 600 for business data of this embodiment includes a modal decomposition module 610 , a feature extraction module 620 , a sample construction module 630 and a prediction module 640 .

[0122] The modal decomposition module 610 is configured to perform variational modal decomposition on the business time series data to obtain a modal feature sequence corresponding to the business time series data; the business time series data includes business data before the date to be measured. In one embodiment, the modal decomposition module 610 can be configured to perform operation S210 described above, which will not be further described here.

[0123] The feature extraction module 620 is used to extract features from the modal feature sequence corresponding to the business time series data to obtain the principal component data corresponding to the business time series data. In one embodiment, the feature extraction module 620 can be used to perform the operation S220 described above, which will not be repeated here.

[0124] The sample construction module 630 is used to construct the feature data to be measured based on the business index and the principal component data corresponding to the business time series data. In one embodiment, the sample construction module 630 can be used to perform the operation S230 described above, which will not be repeated here.

[0125] The prediction module 640 is used to predict the business data results of the test date based on the feature data to be tested and using the pre-trained long short-term memory network model. In one embodiment, the prediction module 640 can be used to perform the operation S240 described above, which will not be repeated here.

[0126] According to an embodiment of the present application, the sample construction module 630 includes: a sliding window unit, which is used to use the principal component data corresponding to the business index and the business time series data as the feature data stream, and based on the prediction time step, use the sliding window to process the feature data stream to construct the feature data to be tested; wherein, the size of the sliding window is determined based on the prediction time step.

[0127] According to an embodiment of the present application, the prediction module 640 includes: a standardization unit for standardizing the feature data to be tested; and a prediction unit for inputting the standardized feature data to be tested into a pre-trained long short-term memory network model for prediction, and outputting the business data results of the date to be tested.

[0128] According to an embodiment of the present application, the standardization unit includes: a maximum value calculation subunit, which is used to calculate the maximum value and the minimum value corresponding to each column of features based on the feature data to be measured; and a mapping subunit, which is used to map each feature data value in the feature data to be measured to a preset numerical range based on the maximum value and the minimum value and using linear transformation.

[0129] According to an embodiment of the present application, the feature extraction module 620 includes: a mean processing unit for performing mean standardization on the modal feature sequence corresponding to the business time series data; a covariance calculation unit for calculating the covariance matrix based on the modal feature sequence after mean standardization; a contribution rate calculation unit for calculating the cumulative variance contribution rate according to the eigenvalue of the covariance matrix; and a principal component determination unit for determining the principal component data corresponding to the business time series data based on the cumulative variance contribution rate and the modal feature sequence after mean standardization.

[0130] According to an embodiment of the present application, the device also includes a model processing module, the model processing module includes a long short-term memory network model, the long short-term memory network model includes a long short-term memory network unit based on the prediction time step; the long short-term memory network unit includes a forget gate, an input gate and an output gate; the forget gate is used to receive historical hidden states and control the retention ratio of current input data; the input gate is used to filter valid information of the current input data; and the output gate is used to generate the current hidden state.

[0131] According to an embodiment of the present application, any multiple modules among the modal decomposition module 610, feature extraction module 620, sample construction module 630, and prediction module 640 can be combined into a single module, or any one of these modules can be split into multiple modules. Alternatively, at least part of the functionality of one or more of these modules can be combined with at least part of the functionality of other modules and implemented in a single module. According to an embodiment of the present application, at least one of the modal decomposition module 610, feature extraction module 620, sample construction module 630, and prediction module 640 can be at least partially implemented as a hardware circuit, such as a field programmable gate array (FPGA), a programmable logic array (PLA), a system on a chip, a system on a substrate, a system on a package, an application-specific integrated circuit (ASIC), or can be implemented in hardware or firmware by any other reasonable means of integrating or packaging circuits, or can be implemented in any one of the three implementation methods of software, hardware, and firmware, or in any appropriate combination of any of them. Alternatively, at least one of the modal decomposition module 610 , the feature extraction module 620 , the sample construction module 630 and the prediction module 640 may be at least partially implemented as a computer program module, which may perform corresponding functions when executed.

[0132] Figure 7 A block diagram of an electronic device suitable for implementing a method for time series prediction of business data according to an embodiment of the present application is schematically shown.

[0133] like Figure 7As shown, an electronic device 900 according to an embodiment of the present application includes a processor 901, which can perform various appropriate actions and processes based on programs stored in a read-only memory (ROM) 902 or programs loaded from a storage unit 908 into a random access memory (RAM) 903. The processor 901 may include, for example, a general-purpose microprocessor (e.g., a CPU), an instruction set processor and / or related chipsets and / or a dedicated microprocessor (e.g., an application-specific integrated circuit (ASIC)), etc. The processor 901 may also include onboard memory for caching purposes. The processor 901 may include a single processing unit or multiple processing units for performing different actions of the method flow according to the embodiment of the present application.

[0134] Various programs and data required for the operation of the electronic device 900 are stored in the RAM 903. The processor 901, the ROM 902, and the RAM 903 are connected to each other via a bus 904. The processor 901 performs various operations of the method flow according to the embodiment of the present application by executing the programs in the ROM 902 and / or the RAM 903. It should be noted that the programs may also be stored in one or more memories other than the ROM 902 and the RAM 903. The processor 901 may also perform various operations of the method flow according to the embodiment of the present application by executing the programs stored in one or more memories.

[0135] According to an embodiment of the present application, electronic device 900 may further include an input / output (I / O) interface 905, which is also connected to bus 904. Electronic device 900 may also include one or more of the following components connected to I / O interface 905: an input section 906 including a keyboard, mouse, etc.; an output section 907 including devices such as a cathode ray tube (CRT), liquid crystal display (LCD), and speakers; a storage section 908 including a hard disk; and a communication section 909 including a network interface card such as a LAN card or modem. Communication section 909 performs communication processing via a network such as the Internet. A drive 910 is also connected to I / O interface 905 as needed. Removable media 911, such as a magnetic disk, optical disk, magneto-optical disk, semiconductor memory, etc., is installed in drive 910 as needed, so that computer programs read from the removable media can be installed into storage section 908 as needed.

[0136] This application also provides a computer-readable storage medium, which may be included in the device / apparatus / system described in the above embodiments, or may exist independently and not be incorporated into the device / apparatus / system. The computer-readable storage medium carries one or more programs, and when the one or more programs are executed, the method according to the embodiments of this application is implemented.

[0137] According to an embodiment of the present application, a computer-readable storage medium may be a non-volatile computer-readable storage medium, and may include, for example, but not limited to: a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination thereof. In the present application, a computer-readable storage medium may be any tangible medium that contains or stores a program that can be used by or in conjunction with an instruction execution system, apparatus, or device. For example, according to an embodiment of the present application, a computer-readable storage medium may include the ROM 902 and / or RAM 903 described above and / or one or more memories other than ROM 902 and RAM 903.

[0138] The embodiments of the present application also include a computer program product, which includes a computer program containing program code for executing the method shown in the flowchart. When the computer program product is executed in a computer system, the program code is used to enable the computer system to implement the time series prediction method for business data provided in the embodiments of the present application.

[0139] The computer program executes the above functions defined in the system / device of the embodiment of the present application when the processor 901 executes the computer program. According to the embodiment of the present application, the system, device, module, unit, etc. described above can be implemented by a computer program module.

[0140] In one embodiment, the computer program may be stored on a tangible storage medium such as an optical storage device or a magnetic storage device. In another embodiment, the computer program may be transmitted and distributed in the form of a signal on a network medium, downloaded and installed via the communication portion 909, and / or installed from a removable medium 911. The program code contained in the computer program may be transmitted using any appropriate network medium, including but not limited to wireless, wired, or any suitable combination thereof.

[0141] In such an embodiment, the computer program can be downloaded and installed from a network via the communication section 909, and / or installed from a removable medium 911. When the computer program is executed by the processor 901, the above-mentioned functions defined in the system of the embodiment of the present application are performed. According to the embodiment of the present application, the systems, devices, means, modules, units, etc. described above can be implemented by computer program modules.

[0142] According to an embodiment of the present application, the program code for executing the computer program provided by the embodiment of the present application can be written in any combination of one or more programming languages. Specifically, these computer programs can be implemented using high-level procedural and / or object-oriented programming languages, and / or assembly / machine languages. Programming languages ​​include, but are not limited to, languages ​​such as Java, C++, Python, "C" or similar programming languages. The program code can be executed entirely on the user computing device, partially on the user device, partially on a remote computing device, or entirely on a remote computing device or server. In the case of a remote computing device, the remote computing device can be connected to the user computing device through any type of network, including a local area network (LAN) or a wide area network (WAN), or can be connected to an external computing device (for example, using an Internet service provider to connect via the Internet).

[0143] The flowcharts and block diagrams in the accompanying drawings illustrate the possible implementation architecture, functions and operations of the systems, methods and computer program products according to various embodiments of the present application. In this regard, each box in the flowchart or block diagram can represent a module, program segment, or a part of code, and the above-mentioned module, program segment, or a part of code contains one or more executable instructions for realizing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the box can also occur in an order different from that marked in the accompanying drawings. For example, two boxes represented in succession can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each box in the block diagram or flowchart, and the combination of the boxes in the block diagram or flowchart, can be implemented with a dedicated hardware-based system that performs the specified function or operation, or can be implemented with a combination of dedicated hardware and computer instructions.

[0144] Those skilled in the art will appreciate that the features described in the various embodiments of this application may be combined and / or coupled in various ways, even if such combinations or couplings are not explicitly described in this application. In particular, the features described in the various embodiments of this application may be combined and / or coupled in various ways without departing from the spirit and teachings of this application. All such combinations and / or couplings fall within the scope of this application.

Claims

1. A time series prediction method for business data, characterized in that: The method comprises: Performing variational modal decomposition on the business time series data to obtain a modal feature sequence corresponding to the business time series data; the business time series data includes business data before the date to be tested; Performing feature extraction on the modal feature sequence corresponding to the business time series data to obtain principal component data corresponding to the business time series data; Constructing feature data to be measured based on the business index and the principal component data corresponding to the business time series data; and Based on the feature data to be tested, a pre-trained long short-term memory network model is used to predict the business data results of the date to be tested.

2. The method according to claim 1, characterized in that The constructing of the feature data to be measured based on the business index and the principal component data corresponding to the business time series data includes: Using the business index and the principal component data corresponding to the business time series data as a feature data stream, and based on the prediction time step, processing the feature data stream using a sliding window to construct the feature data to be measured; The size of the sliding window is determined based on the prediction time step.

3. The method according to claim 1, characterized in that The method of predicting the business data results of the date to be tested based on the feature data to be tested and using a pre-trained long short-term memory network model includes: Normalizing the characteristic data to be measured; and The standardized feature data to be tested is input into the pre-trained long short-term memory network model for prediction, and the business data results of the date to be tested are output.

4. The method according to claim 3, characterized in that The standardizing the characteristic data to be measured includes: Calculate the maximum value and the minimum value corresponding to each column of features according to the feature data to be measured; and Based on the maximum value and the minimum value, each characteristic data value in the characteristic data to be measured is mapped to a preset value range using linear transformation.

5. The method according to claim 1, wherein The extracting features of the modal feature sequence corresponding to the business time series data to obtain principal component data corresponding to the business time series data includes: Normalizing the modal feature sequence corresponding to the business time series data by means; Calculate the covariance matrix based on the modal feature sequence after mean normalization; Calculating the cumulative variance contribution rate according to the eigenvalues ​​of the covariance matrix; and The principal component data corresponding to the business time series data is determined according to the cumulative variance contribution rate and the modal feature sequence after mean normalization.

6. The method according to claim 2, characterized in that The long short-term memory network model includes a long short-term memory network unit based on the prediction time step; The long short-term memory network unit includes a forget gate, an input gate and an output gate; The forget gate is used to receive the historical hidden state and control the retention ratio of the current input data; The input gate is used to filter valid information of the current input data; and The output gate is used to generate the current hidden state.

7. A time series prediction device for business data, characterized in that: The device comprises: A modal decomposition module is used to perform variational modal decomposition on the business time series data to obtain a modal feature sequence corresponding to the business time series data; the business time series data includes business data before the date to be tested; A feature extraction module is used to extract features from the modal feature sequence corresponding to the business time series data to obtain principal component data corresponding to the business time series data; A sample construction module, configured to construct feature data to be tested based on the business index and the principal component data corresponding to the business time series data; and The prediction module is used to predict the business data results of the date to be tested based on the feature data to be tested and using a pre-trained long short-term memory network model.

8. An electronic device comprising: one or more processors; a memory for storing one or more computer programs, It is characterized in that the one or more processors execute the one or more computer programs to implement the steps of the method according to any one of claims 1 to 6.

9. A computer-readable storage medium having a computer program or instruction stored thereon, characterized in that: When the computer program or instruction is executed by a processor, the steps of the method according to any one of claims 1 to 6 are implemented.

10. A computer program product comprising a computer program or instructions, characterized in that When the computer program or instruction is executed by a processor, the steps of the method according to any one of claims 1 to 6 are implemented.

Images