Fund flow analysis method and device, electronic equipment and storage medium

Abstract

The invention provides a fund flow analysis method and device, electronic equipment and a storage medium, and the method comprises the steps: obtaining fund flow time series data of a target fund flow index, and carrying out the processing of the fund flow time series data, and obtaining the processed target time series data and feature data; predicting the fund flow based on the target time sequence data through a first time sequence prediction algorithm to obtain first fund flow prediction sequence data; predicting the fund flow based on the target time sequence data and the feature data through a second time sequence prediction algorithm to obtain second fund flow prediction sequence data; fusing the first fund flow prediction sequence data and the second fund flow prediction sequence data to obtain fund flow prediction result data; abnormal root cause analysis is carried out on fund flow prediction result data through a multi-dimensional data index root cause system, fund flow abnormal root cause elements are determined, the prediction precision is improved, and abnormal analysis and positioning are achieved.

Description

Technical Field

[0001] The present application relates to the technical field of time series data processing, and in particular to a method, device, electronic device and storage medium for analyzing capital flows. Background Art

[0002] In reality, banking and financial institutions have a huge customer base and financial business scenarios involve a large amount of capital inflows and outflows every day. Facing a large customer base, the pressure of capital management is very high. For banking and institution managers, it is particularly important to explore the rules of capital flow, accurately predict the inflow and outflow of funds in various scenarios, and identify and analyze abnormal capital flows while ensuring that the risk of capital liquidity is minimized and meeting the daily business operations. Summary of the invention

[0003] The present application aims to solve one of the technical problems in the related art at least to some extent.

[0004] To this end, the first purpose of this application is to propose a cash flow analysis method to improve prediction accuracy and achieve accurate anomaly analysis and positioning.

[0005] The second objective of this application is to provide a fund flow analysis device.

[0006] The third objective of the present application is to provide an electronic device.

[0007] A fourth objective of the present application is to provide a computer-readable storage medium.

[0008] A fifth object of the present application is to provide a computer program product.

[0009] To achieve the above-mentioned purpose, the first embodiment of the present application proposes a method for analyzing capital flow, including:

[0010] Acquire the capital flow time series data of the target capital flow indicator, and process the capital flow time series data to obtain processed target time series data and characteristic data;

[0011] Using a first time series prediction algorithm, predicting the capital flow based on the target time series data, to obtain first capital flow prediction sequence data;

[0012] Using a second time series prediction algorithm, predicting the capital flow based on the target time series data and the characteristic data, to obtain second capital flow prediction series data;

[0013] The first capital flow prediction sequence data and the second capital flow prediction sequence data are integrated to obtain capital flow prediction result data;

[0014] The abnormal root cause analysis of the cash flow forecast result data is performed through a multi-dimensional data indicator root cause system to determine the abnormal root cause elements of the cash flow.

[0015] In some implementations, the processing of the capital flow time series data to obtain characteristic data includes:

[0016] Based on the temporal order of the capital flow time series data, characteristic data of the capital flow time series data is retroactively constructed.

[0017] In some implementations, based on the temporal order of the capital flow time series data, retroactively constructing feature data of the capital flow time series data includes:

[0018] Get trend correlation between multiple money flow indicators;

[0019] Based on the trend correlation between the multiple capital flow indicators, relevant data of the first capital flow indicator that has a trend correlation with the target capital flow indicator is used as feature data of the capital flow time series data.

[0020] In some implementations, the processing of the capital flow time series data to obtain processed target time series data includes:

[0021] Performing change point detection on the capital flow time series data to obtain a target change point of the capital flow time series data;

[0022] In chronological order, the sequence values ​​of N time points before the target change point are updated to a first value, and the sequence values ​​of M time points after the target change point are updated to a second value, to obtain processed target time series data.

[0023] In some implementations, the first time series prediction algorithm uses a decomposable time series model for prediction, and the second time series prediction algorithm uses a complex neural network model for prediction.

[0024] In some implementations, the decomposable time series model is a Prophet model, which uses an additive model to fit and predict the cash flow time series. The additive model divides the cash flow time series into a superposition of trend terms, seasonal terms, holiday terms, and noise terms; the trend term is used to fit the non-periodic trend changes of the cash flow time series, and the seasonal term is used to fit the weekly, monthly, and quarterly cyclical change trends of the cash flow time series, using Fourier series for approximation; the holiday term represents the impact of potential jump points of the fitted cash flow time series on the prediction, and the noise term is used to represent the random fluctuations of the unpredicted cash flow time series.

[0025] In some implementations, the complex neural network model is a NBEATSx model that implements predictions using a basic block and multiple double residual stacks, wherein the basic block is an explanatory trend and seasonality model, and the input of the basic block includes target time series data and feature data; the target time series data and feature data are converted into an implicit state through a fully connected layer, and then the implicit state is converted into an expansion coefficient through two linear layers; and prediction data is obtained based on the target time series data through mapping.

[0026] In some implementations, different double residual stacks respectively use harmonic functions, polynomial trend terms and external covariates to respectively fit information related to cycles, trends and external information in the capital flow time series.

[0027] In some implementations, the fusing of the first capital flow prediction sequence data and the second capital flow prediction sequence data to obtain capital flow prediction result data includes:

[0028] By using the weighted average method, the first fund flow prediction sequence data and the second fund flow prediction sequence data are integrated to obtain fund flow prediction result data. To achieve the above purpose, the second embodiment of the present application proposes a fund flow analysis device, including:

[0029] A data acquisition module is used to acquire the capital flow time series data of the target capital flow indicator, and process the capital flow time series data to obtain the processed target time series data and feature data;

[0030] A first model prediction module is used to predict the capital flow based on the target time series data by using a first time series prediction algorithm to obtain first capital flow prediction sequence data;

[0031] A second model prediction module is used to predict the capital flow based on the target time series data and the feature data by using a second time series prediction algorithm to obtain second capital flow prediction sequence data;

[0032] A data fusion module, used for fusing the first capital flow prediction sequence data and the second capital flow prediction sequence data to obtain capital flow prediction result data;

[0033] The root cause analysis module is used to perform abnormal root cause analysis on the fund flow prediction result data through a multi-dimensional data indicator root cause system to determine the abnormal root cause elements of the fund flow.

[0034] In some implementations, when the data acquisition module processes the capital flow time series data to obtain characteristic data, it is used to:

[0035] Based on the temporal order of the capital flow time series data, characteristic data of the capital flow time series data is retroactively constructed.

[0036] In some implementations, the data acquisition module, when retroactively constructing the feature data of the capital flow time series data based on the temporal order of the capital flow time series data, is used to:

[0037] Get trend correlation between multiple money flow indicators;

[0038] Based on the trend correlation between the multiple capital flow indicators, relevant data of the first capital flow indicator that has a trend correlation with the target capital flow indicator is used as feature data of the capital flow time series data.

[0039] In some implementations, when the data acquisition module processes the capital flow time series data to obtain the processed target time series data, it is also used to:

[0040] Performing change point detection on the capital flow time series data to obtain a target change point of the capital flow time series data;

[0041] In chronological order, the sequence values ​​of N time points before the target change point are updated to a first value, and the sequence values ​​of M time points after the target change point are updated to a second value, to obtain processed target time series data.

[0042] In some implementations, the first time series prediction algorithm uses a decomposable time series model for prediction, and the second time series prediction algorithm uses a complex neural network model for prediction.

[0043] In some implementations, the decomposable time series model is an additive Prophet model, which includes a superposition of trend terms, seasonal terms, holiday terms, and noise terms; wherein the trend term is used to fit the non-periodic trend changes of the capital flow time series, and the seasonal term is used to fit the weekly, monthly, and quarterly cyclical change trends of the capital flow time series, using Fourier series for approximation; the holiday term represents the impact of potential jump points of the fitting capital flow time series on the prediction, and the noise term is used to represent the random fluctuations of the unpredicted capital flow time series.

[0044] In some implementations, the complex neural network model is a NBEATSx model, which includes a plurality of double residual stacks, each of which includes a plurality of basic blocks, and the basic blocks are explanatory trend and seasonal models; the second time series prediction algorithm is used to predict the capital flow based on the target time series data and the feature data to obtain the second capital flow prediction series data; including:

[0045] Inputting the target time series data and the feature data into the basic Block, and transforming the target time series data and the feature data into a hidden state through the fully connected layer of the basic Block;

[0046] An expansion coefficient is generated based on the implicit state, and the expansion coefficient is mapped into time series data to obtain second capital flow prediction series data.

[0047] In some implementations, different double residual stacks respectively use harmonic functions, polynomial trend terms and external covariates to respectively fit information related to cycles, trends and external information in the capital flow time series.

[0048] In some implementations, the data fusion module is specifically used to:

[0049] The first capital flow prediction sequence data and the second capital flow prediction sequence data are integrated by weighted average method to obtain capital flow prediction result data.

[0050] To achieve the above-mentioned purpose, the third aspect embodiment of the present application proposes an electronic device, comprising: a processor, and a memory communicatively connected to the processor; the memory stores computer-executable instructions; the processor executes the computer-executable instructions stored in the memory to implement the method described in the first aspect.

[0051] To achieve the above-mentioned purpose, the fourth aspect of the present application proposes a computer-readable storage medium, in which computer-readable storage medium is stored computer-executable instructions, and when the computer-executable instructions are executed by a processor, they are used to implement the method described in the first aspect.

[0052] To achieve the above-mentioned purpose, the fifth aspect of the present application proposes a computer program product, including a computer program, which implements the method described in the first aspect when executed by a processor.

[0053] The cash flow analysis method, device, electronic device and storage medium provided in this application predict the institutional cash flow by introducing two time series algorithm integrated models, thereby improving the accuracy of model predictions; by introducing the root cause analysis method to analyze the causes of change points or anomalies in the cash flow, effective data support is provided for management decision-making. Before model prediction, the time series data is processed through change point detection and change point processing strategies, so that changes in data structure can be explicitly represented during model prediction, further improving the accuracy of model predictions. Before model prediction, the feature data of the time series data is constructed to improve the accuracy of model predictions.

[0054] Additional aspects and advantages of the present application will be given in part in the description below, and in part will become apparent from the description below, or will be learned through the practice of the present application. BRIEF DESCRIPTION OF THE DRAWINGS

[0055] The above and / or additional aspects and advantages of the present application will become apparent and easily understood from the following description of the embodiments in conjunction with the accompanying drawings, in which:

[0056] Figure 1 A flow chart of a method for analyzing capital flow provided in an embodiment of the present application;

[0057] Figure 2 A schematic diagram of a process for processing capital flow time series data provided in an embodiment of the present application;

[0058] Figure 3 A schematic diagram of a process for constructing characteristic data of capital flow time series data provided by an embodiment of the present application;

[0059] Figure 4 A schematic diagram of a process for detecting and processing change points of capital flow time series data provided by an embodiment of the present application;

[0060] Figure 5 A schematic diagram of a process flow of a second time series prediction algorithm provided in an embodiment of the present application;

[0061] Figure 6 A flow chart of a method for analyzing capital flows provided in another embodiment of the present application;

[0062] Figure 7 A block diagram of a fund flow analysis device provided in an embodiment of the present application;

[0063] Figure 8 A block diagram of an electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0064] Embodiments of the present application are described in detail below, and examples of the embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are intended to be used to explain the present application, and should not be construed as limiting the present application.

[0065] The following describes the cash flow analysis method, device and equipment of the embodiments of the present application with reference to the accompanying drawings.

[0066] Figure 1 A flow chart of a method for analyzing cash flow provided in an embodiment of the present application.

[0067] It should be noted that the execution subject of the cash flow analysis method of the embodiment of the present application is the cash flow analysis device of the embodiment of the present application, and the cash flow analysis device can be configured in an electronic device so that the electronic device can perform the cash flow analysis function.

[0068] like Figure 1 As shown, the fund flow analysis method includes the following steps:

[0069] Step 101, acquiring the capital flow time series data of the target capital flow indicator, and processing the capital flow time series data to obtain the processed target time series data and characteristic data.

[0070] As an implementation method, real-time or batch collection of business data including liability business, asset business, intermediary business and off-balance sheet business within financial institutions is performed to form relevant data indicators, such as deposit inflow and outflow amounts, loan inflow and outflow amounts, etc., to form the fund flow time series data of each fund flow indicator. For example, the fund flow time series data corresponding to the deposit inflow indicator of the head office and the fund flow time series data corresponding to the loan outflow indicator of the branch are obtained, so that the fund flow of the corresponding indicator can be predicted through the time series prediction algorithm in the future.

[0071] After obtaining the capital flow time series data of each capital flow indicator, the capital flow time series data can be processed. For example, the characteristics of the capital flow time series data can be constructed to improve the accuracy of the prediction when the time series prediction algorithm is used for subsequent prediction; in addition, the capital flow time series data can be subjected to change point detection and change point processing.

[0072] In some embodiments, processing the fund flow time series data includes:

[0073] Step 201, based on the time sequence of the capital flow time series data, retroactively construct feature data of the capital flow time series data.

[0074] Illustratively, during data preprocessing, feature data is constructed based on time-series traceability, including but not limited to: whether it is a working day, whether it is a holiday, which day of the holiday it is, the number of days until the next holiday, which day of the week it is (Sunday is 1), which week of the year it is, the season it is, whether there are major deposit and loan policies, etc.

[0075] In some embodiments, based on the temporal order of the capital flow time series data, the characteristic data of the capital flow time series data is retroactively constructed; including:

[0076] Step 301, obtaining trend correlations between multiple fund flow indicators.

[0077] Step 302: Based on the trend correlation between the multiple capital flow indicators, the relevant data of the first capital flow indicator having a trend correlation with the target capital flow indicator is used as the characteristic data of the capital flow time series data.

[0078] Among them, trend correlation can be understood as coordinated changes, with a coordinated trend of change.

[0079] It can be understood that it is noted that the overall trends among the indicators are related. When predicting one of the indicators, other indicators can also be used as features to improve the accuracy of the prediction.

[0080] In some embodiments, a method for detecting and processing change points of capital flow time series data includes:

[0081] Step 401, performing change point detection on the capital flow time series data to obtain the target change point of the capital flow time series data.

[0082] Step 402: Update the sequence values ​​of N time points before the target change point to a first value, and update the sequence values ​​of M time points after the target change point to a second value, in a time sequence order.

[0083] It should be noted that the change points in the time series data are processed so that the changes in the data structure can be explicitly represented during model prediction.

[0084] It should be noted that in the analysis and prediction of time series, accurate detection of structural changes is crucial. Newly emerging distribution patterns often cause historical data to lose their representativeness, which in turn affects the effectiveness of models trained on these data. Structural changes and outliers are two different concepts. Outliers refer to observations that deviate significantly from the normal behavior defined by the basic structure or distribution of the time series. Change points refer to changes in the overall structure of the time series, resulting in changes in distribution characteristics. Change point detection (CPD) is called change point detection. Its basic definition is that in a sequence or process, when a statistical characteristic (distribution type, distribution parameter) changes at a certain point in time due to systematic factors rather than accidental factors, the time point is called a change point. Change point identification in time series is of great significance for economic analysis and modeling. The existence of change points will have a significant impact on the parameter estimation and prediction of the model. Therefore, accurately identifying change points in time series is crucial to understanding and predicting the changing trends of economic phenomena. The cumulative sum mean change point detection method provides an effective tool to help researchers and decision makers better understand the structural changes of data and make more accurate predictions and decisions. However, they account for a relatively small proportion in the data set, which causes these outliers to seriously affect the accuracy of the model and are almost impossible to be predicted by the trained model (screening out outliers is one approach, but it may destroy the consistency of the distribution of the training set and the test set, resulting in a decrease in generalization ability).

[0085] The following is an example of how to implement change point detection.

[0086] Cumulative Sum Mean Changepoint Detection: CUSUM (cumulative sum) is a widely used method for detecting mean changes in time series. The principle is to calculate the cumulative sum of deviations from the expected value. If the mean of the series remains stable, the cumulative sum will show random fluctuations around zero. A sustained positive (or negative) change in the cumulative sum indicates that the mean has moved upward (or downward). Changepoint detection refers to the point where a major structural break or transition has occurred in the time series. These changes may be caused by external factors such as data generation, technology, or consumer behavior. Detecting these change points is very important because it helps us understand and quantify the changes.

[0087] Time series change point processing strategy: introduce dummy variables, assign 0 to the n time points before the change point and assign 1 to the m time points after the change point, so that structural changes can be explicitly represented in subsequent prediction models.

[0088] Therefore, processing the capital flow time series data may include acquiring feature data and detecting and processing change points. Acquiring feature data can improve the accuracy of model predictions. Detecting and processing change points can enable the model in subsequent steps to extract the characteristics of changes in data structure and improve prediction accuracy.

[0089] Step 102: predicting the capital flow based on the target time series data by using a first time series prediction algorithm to obtain first capital flow prediction sequence data.

[0090] In some embodiments, the first time series prediction algorithm is a decomposable time series model. Optionally, the decomposable time series model is an additive Prophet model, which includes a superposition of trend terms, seasonal terms, holiday terms, and noise terms; wherein the trend term is used to fit the non-periodic trend changes of the capital flow time series, the seasonal term is used to fit the weekly, monthly, and seasonal cyclical change trends of the capital flow time series, and Fourier series is used for approximation; the holiday term represents the impact of the potential jump point of the fitting capital flow time series on the prediction, and the noise term is used to represent the random fluctuations of the unpredicted capital flow time series.

[0091] Specifically, the Prophet time series algorithm is used to regard the capital flow time series as a function of t, and the capital flow time series is decomposed into trend terms, seasonal terms, holiday terms and noise terms. There are two modes: additive model and multiplicative model. The core formula of the additive model is as follows:

[0092] y(t)=g(t)+s(t)+h(t)+∈(t)

[0093] Among them, g(t) represents the trend term, s(t) represents the season term, h(t) represents the holiday term (or generally refers to external variables), and ∈(t) represents the noise term. The trend term is used to fit the non-periodic trend changes of the time series, such as: upward and downward trends. According to the trend mode, it can be divided into linear trend and nonlinear trend. The formula of linear trend is:

[0094] g(t)=k(t)+m

[0095] The formula for the nonlinear trend is:

[0096]

[0097] Among them, C represents the upper capacity, that is, the upper limit that g(t) can reach; k represents the growth rate; m represents the offset parameter, the turning point where the slope of the trend term changes. By adjusting m, the curve can be shifted left and right. In the specific implementation, C and k are both functions of time t, and the change of k is discontinuous. In order to make g(t) continuous, a series of complex transformations will be introduced. The seasonal term is used to fit the periodic change trend of weeks, months, seasons, etc., and the Fourier series is used to approximate:

[0098]

[0099] remember:

[0100]

[0101] Among them, s(t) = X(t)β, β is what needs to be learned, [a1, b1, ..., a n , b n ] T Subordinate to the normal distribution N(0,σ 2 ). P represents the period (365.25 for years and 7 for weeks), and N represents the number of approximations used (10 for years and 3 for weeks).

[0102] The holiday item is used to indicate the impact of potential jump points on the forecast, such as holidays, emergencies, etc. Since each holiday has a different impact on the time series, different holidays can be regarded as independent models. Different holidays are set with different window values, indicating that the holiday will affect the time series for a period of time before and after.

[0103] Assume that there are L different holidays, Di represents the date set of the holiday window, and the holiday item can be expressed as:

[0104] Z(t)=[1(t∈D1,…,1(t∈DL)]

[0105] h(t)=Z(t)κ

[0106] Among them, κ also obeys the normal distribution N(0,ν 2 ), the larger the ν value is, the greater the impact of holidays on the model is; the smaller the value is, the smaller the effect of holidays on the model is.

[0107] Finally, the noise term is used to represent unpredicted random fluctuations.

[0108] According to the above introduction, the Prophet model is actually training: k, m in the trend term; β in the seasonal term; κ in the holiday term; and the error term ∈(t). The Prophet model is suitable for predicting time series data that meets the following conditions:

[0109] Training data: Have at least one complete cycle of data to allow the model to fully learn the rules;

[0110] Data trends: Data has certain normal periodic effects, such as weekend effect, seasonal effect, etc.

[0111] Change situation: clarify the time point and window period when the change may occur, such as: Double Eleven, Spring Festival, etc.;

[0112] Missing value situation: The missing values ​​and outliers of historical data remain within a reasonable range.

[0113] Step 103: predict the capital flow based on the target time series data by using a second time series prediction algorithm to obtain second capital flow prediction series data.

[0114] In some embodiments, the second time series prediction algorithm is a complex neural network model, and optionally, the complex neural network model is a NBeatsx deep learning algorithm. N-BEATS is a deep learning model designed specifically for time series prediction, emphasizing interpretability. Specifically: 1) In addition to time series, it allows the input of external covariates to extract useful information and time-dependent information in the data. 2) Interpretable time series signal decomposition, flexible fitting of multiple seasonality in time series through nonlinear transformation, and interaction of factors such as holidays with external covariates.

[0115] In some embodiments, the NBEATSx model includes a plurality of double residual stacks, each of which includes a plurality of basic blocks, and the basic blocks are explanatory trend and seasonal models.

[0116] On the basis of the previous embodiment, the second time series prediction algorithm is used to predict the capital flow based on the target time series data and the feature data to obtain the second capital flow prediction series data; including:

[0117] Step 501, input the target time series data and the feature data into a basic Block, and transform the target time series data and the feature data into an implicit state through a fully connected layer of the basic Block;

[0118] Step 502, generate an expansion coefficient based on the implicit state, and map the expansion coefficient to time series data; to obtain the second fund flow forecast sequence data. Specifically, the NBEATSx model is predicted by the basic Block and the double residual Stack. The basic function can be a general structure or an explanatory trend and seasonal model. In addition to the historical sequence, the input of the block also includes external covariates:

[0119]

[0120] Through the fully connected layer, both are transformed into hidden states h s，b , and then two linear layers are used to transform the implicit state into an expansion coefficient At this time, the expansion coefficient and The information in external covariates and historical series is preserved.

[0121]

[0122] After obtaining the expansion coefficient, the next step is to map and obtain the prediction of past and future values. and In order to improve the interpretability of the model, different stacks can use harmonic functions, polynomial trend terms and external covariates to fit the information related to the cycle, trend and external information in the sequence, as follows:

[0123]

[0124] Step 104, integrating the first capital flow prediction sequence data and the second capital flow prediction sequence data to obtain capital flow prediction result data.

[0125] In some embodiments, the first capital flow prediction sequence data and the second capital flow prediction sequence data are integrated to obtain the capital flow prediction result data; including: integrating the first capital flow prediction sequence data and the second capital flow prediction sequence data through a weighted average method to obtain the capital flow prediction result data.

[0126] An integrated model of two highly interpretable algorithms, the Prophet time series algorithm and the NBeatsx deep learning algorithm, is introduced to predict institutional capital flows, thereby improving the accuracy of model predictions.

[0127] Step 105, performing abnormal root cause analysis on the cash flow prediction result data through a multi-dimensional data indicator root cause system to determine the abnormal cash flow root cause elements.

[0128] It can be understood that when the overall values ​​of the inflow and outflow indicators of the capital flow are abnormal, locating the root cause of the abnormality is a key step to resolve the abnormality. However, this step is often challenging, especially when the root cause is a combination of multiple dimensional attribute values ​​(such as transaction channels, product types, etc.).

[0129] This embodiment adopts the Adtributor root cause analysis method to perform abnormal root cause analysis. The core process of Adtributor root cause analysis includes the following steps:

[0130] Step 1: Calculate the difference (surprise) between the actual value and predicted value of a factor in a certain dimension in the total revenue:

[0131] A: Calculate p ij ,q ij Among them, p ij It represents the ratio of the predicted income corresponding to factor j under dimension i to the total predicted income. The specific formula is:

[0132] p ij (m) = Fij (m) / F(m)

[0133] q ij (m) = A ij (m) / A(m)

[0134] B: Calculate p ij ,q ij The difference is calculated using JS divergence. The greater the direct difference, the greater the JS divergence value.

[0135] JS divergence: D JS (P, Q) = 0.5 (i∑p i logp i +q i 2p i +i∑q i logp i +q i 2q i )

[0136] Step 2: Calculate the ratio of the fluctuation of a factor to the total fluctuation under a certain dimension (Explanatory power), where the fluctuation represents the difference between the true value and the abnormal value. The specific formula is as follows:

[0137] EP ij =(A ij (m)-F ij (m)) / (A(m)-F(m))

[0138] Step 3: Sort and get the attribution results.

[0139] The dimensions are sorted by calculating the surprise of the dimension (the sum of the surprises of all elements in the dimension) to determine the dimension (such as the province) where the root cause is located. The explanatory power of each element is calculated within the dimension, and when the sum of the explanatory power of the elements exceeds the threshold, these elements are considered to be the root cause.

[0140] This embodiment uses a root cause analysis method to determine the root cause of the abnormality of the capital flow time series data, so as to handle the abnormal situation accordingly.

[0141] It should be noted that the analysis of capital flow in this application is for bank financial institutions such as branches, sub-branches, and head offices, and is not an analysis of abnormal capital flow of individual or corporate customers. The above analysis provides macro data support for decision-making for management at different levels of banking institutions.

[0142] The cash flow analysis method of the embodiment of the present application predicts the institutional cash flow by introducing two time series algorithm integrated models, thereby improving the accuracy of model predictions; by introducing the root cause analysis method to analyze the causes of change points or anomalies in the cash flow, effective data support is provided for management decision-making. Before model prediction, the time series data is processed through change point detection and change point processing strategies, so that the changes in data structure can be explicitly represented during model prediction, further improving the accuracy of the prediction.

[0143] Figure 6 The following is a flow chart of another method for analyzing capital flow provided in the embodiment of the present application. Figure 2 As shown, the capital flow analysis method may include the following steps:

[0144] Step 601, obtaining the capital flow time series data of the target capital flow indicator.

[0145] Step 602: based on the time sequence of the capital flow time series data, retroactively construct characteristic data of the capital flow time series data.

[0146] Step 603, performing change point detection and change point processing on the capital flow time series data to obtain processed target time series data.

[0147] It should be noted that the implementation of the above steps may specifically refer to step 101 in the above embodiment, the principles are the same and will not be repeated here.

[0148] Step 604, predicting the capital flow based on the target time series data by using the Prophet time series algorithm to obtain first capital flow prediction series data.

[0149] It should be noted that the implementation method of this step may specifically refer to step 102 in the above embodiment, the principle is the same and will not be repeated here.

[0150] Step 605, using the NBeatsx deep learning algorithm, predict the capital flow based on the target time series data and the feature data to obtain second capital flow prediction sequence data.

[0151] It should be noted that the implementation method of this step may specifically refer to step 103 in the above embodiment, the principle is the same and will not be repeated here.

[0152] Step 606: The first capital flow prediction sequence data and the second capital flow prediction sequence data are integrated to obtain capital flow prediction result data.

[0153] It should be noted that the implementation method of this step may specifically refer to step 104 in the above embodiment, the principle is the same and will not be repeated here.

[0154] Step 607, performing abnormal root cause analysis on the cash flow forecast result data through the multi-dimensional data indicator root cause system to determine the abnormal cash flow root cause elements.

[0155] It should be noted that the implementation method of this step may specifically refer to step 105 in the above embodiment, the principle is the same and will not be repeated here.

[0156] In order to implement the above embodiment, the present application also proposes a fund flow analysis device. Figure 7 This is a schematic diagram of the structure of a fund flow analysis device provided in an embodiment of the present application. Figure 7 As shown, the fund flow analysis device may include: a data acquisition module 701 , a first model prediction module 702 , a second model prediction module 703 , a data fusion module 704 and a root cause analysis module 705 .

[0157] The data acquisition module 301 is used to acquire the capital flow time series data of the target capital flow indicator, and process the capital flow time series data to obtain the processed target time series data and feature data;

[0158] The first model prediction module 302 is used to predict the capital flow based on the target time series data by using the first time series prediction algorithm to obtain the first capital flow prediction sequence data;

[0159] The second model prediction module 703 is used to predict the capital flow based on the target time series data by using a second time series prediction algorithm to obtain second capital flow prediction series data;

[0160] A data fusion module 704 is used to fuse the first capital flow prediction sequence data and the second capital flow prediction sequence data to obtain capital flow prediction result data;

[0161] The root cause analysis module 705 is used to perform abnormal root cause analysis on the capital flow prediction result data through a multi-dimensional data indicator root cause system to determine the abnormal root cause elements of the capital flow.

[0162] Furthermore, in a possible implementation of the embodiment of the present application, the data acquisition module 301 processes the capital flow time series data to obtain characteristic data, which is used to:

[0163] Based on the temporal order of the capital flow time series data, the characteristic data of the capital flow time series data is retroactively constructed.

[0164] Furthermore, in a possible implementation of the embodiment of the present application, the data acquisition module 301, when retroactively constructing feature data of the capital flow time series data based on the temporal order of the capital flow time series data, is used to:

[0165] Get trend correlation between multiple money flow indicators;

[0166] Based on the trend correlation between multiple capital flow indicators, relevant data of the first capital flow indicator that has a trend correlation with the target capital flow indicator is used as feature data of the capital flow time series data.

[0167] Furthermore, in a possible implementation of the embodiment of the present application, when the data acquisition module 301 processes the capital flow time series data to obtain the processed target time series data, it is also used to:

[0168] Perform change point detection on the capital flow time series data to obtain the target change point of the capital flow time series data;

[0169] In chronological order, the sequence values ​​of N time points before the target change point are updated to the first value, and the sequence values ​​of M time points after the target change point are updated to the second value, so as to obtain the processed target time series data.

[0170] Furthermore, in a possible implementation of an embodiment of the present application, the first time series prediction algorithm uses a decomposable time series model for prediction, and the second time series prediction algorithm uses a complex neural network model for prediction.

[0171] Furthermore, in a possible implementation of an embodiment of the present application, the decomposable time series model is an additive Prophet model, which includes a superposition of trend terms, seasonal terms, holiday terms and noise terms; wherein the trend term is used to fit the non-periodic trend changes of the capital flow time series, the seasonal term is used to fit the weekly, monthly and quarterly cyclical change trends of the capital flow time series, and Fourier series is used for approximation; the holiday term represents the impact of the potential jump points of the fitting capital flow time series on the prediction, and the noise term is used to represent the random fluctuations of the unpredicted capital flow time series.

[0172] Further, in a possible implementation of the embodiment of the present application, the NBEATSx model includes a plurality of double residual stacks, each of which includes a plurality of basic blocks, and the basic blocks are explanatory trend and seasonal models; the second model prediction module 703 is specifically used for:

[0173] Input the target time series data and the feature data into the basic Block, and transform the target time series data and the feature data into a hidden state through the fully connected layer of the basic Block;

[0174] An expansion coefficient is generated based on the implicit state, and the expansion coefficient is mapped to time series data to obtain the second capital flow prediction series data.

[0175] Furthermore, in a possible implementation of an embodiment of the present application, different double residual stacks respectively use harmonic functions, polynomial trend terms and external covariates to respectively fit information related to cycles, trends and external information in the capital flow time series.

[0176] Furthermore, in a possible implementation of the embodiment of the present application, the data fusion module 704 is specifically configured to:

[0177] The first capital flow prediction sequence data and the second capital flow prediction sequence data are integrated through the weighted average method to obtain the capital flow prediction result data.

[0178] It should be noted that the above explanations of the embodiment of the cash flow analysis method are also applicable to the cash flow analysis device of this embodiment, and will not be repeated here.

[0179] Based on the above embodiment, the embodiment of the present application also provides a possible implementation method of a capital flow analysis device. On the basis of the previous embodiment, the capital flow analysis device also includes:

[0180] The fund flow analysis device of the embodiment of the present application predicts the institutional fund flow by introducing two time series algorithm integrated models, thereby improving the accuracy of model prediction; by introducing the root cause analysis method to analyze the causes of change points or anomalies in the fund flow, it provides effective data support for management decision-making. Before model prediction, the time series data is processed through change point detection and change point processing strategies, so that the changes in data structure can be explicitly represented during model prediction, further improving the accuracy of the prediction.

[0181] In order to implement the above embodiment, the present application also proposes an electronic device. Figure 8 , Figure 8 Schematic diagram of the structure of the electronic device provided in the embodiment of the present application. Figure 8 As shown, the electronic device 800 includes: a processor 801, and a memory 802 communicatively connected to the processor 401; the memory 802 stores computer-executable instructions; the processor 801 executes the computer-executable instructions stored in the memory to implement the method provided in the aforementioned embodiment.

[0182] In order to implement the above embodiments, the present application also proposes a computer-readable storage medium, in which computer-executable instructions are stored. When the computer-executable instructions are executed by a processor, they are used to implement the methods provided by the above embodiments.

[0183] In order to implement the above embodiments, the present application also proposes a computer program product, including a computer program, which implements the methods provided by the above embodiments when executed by a processor.

[0184] The collection, storage, use, processing, transmission, provision and disclosure of user personal information involved in this application are in compliance with relevant laws and regulations and do not violate public order and good morals.

[0185] It should be noted that personal information from users should be collected for legitimate and reasonable purposes and should not be shared or sold outside of these legitimate uses. In addition, such collection / sharing should be carried out after receiving the user's informed consent, including but not limited to notifying the user to read the user agreement / user notice and sign the agreement / authorization including authorization of relevant user information before the user uses the function. In addition, any necessary steps should be taken to protect and safeguard access to such personal information data and ensure that others who have access to personal information data comply with its privacy policy and procedures.

[0186] The present application is expected to provide an implementation scheme for users to selectively block the use or access of personal information data. That is, the present disclosure is expected to provide hardware and / or software to prevent or block access to such personal information data. Once the personal information data is no longer needed, the risk can be minimized by limiting data collection and deleting the data. In addition, when applicable, such personal information is de-identified to protect the privacy of the user.

[0187] In the description of the aforementioned embodiments, the description with reference to the terms "one embodiment", "some embodiments", "example", "specific example", or "some examples" etc. means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present application. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described may be combined in any one or more embodiments or examples in a suitable manner. In addition, those skilled in the art may combine and combine the different embodiments or examples described in this specification and the features of the different embodiments or examples, without contradiction.

[0188] In addition, the terms "first" and "second" are used for descriptive purposes only and should not be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features. Therefore, the features defined as "first" and "second" may explicitly or implicitly include at least one of the features. In the description of this application, the meaning of "plurality" is at least two, such as two, three, etc., unless otherwise clearly and specifically defined.

[0189] Any process or method description in a flowchart or otherwise described herein may be understood to represent a module, fragment or portion of code comprising one or more executable instructions for implementing the steps of a custom logical function or process, and the scope of the preferred embodiments of the present application includes alternative implementations in which functions may not be performed in the order shown or discussed, including performing functions in a substantially simultaneous manner or in the reverse order depending on the functions involved, which should be understood by technicians in the technical field to which the embodiments of the present application belong.

[0190] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as an ordered list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by an instruction execution system, device or apparatus (such as a computer-based system, a system including a processor, or other system that can fetch instructions from an instruction execution system, device or apparatus and execute the instructions), or in combination with these instruction execution systems, devices or apparatuses. For the purpose of this specification, "computer-readable medium" can be any device that can contain, store, communicate, propagate or transmit a program for use by an instruction execution system, device or apparatus, or in combination with these instruction execution systems, devices or apparatuses. More specific examples of computer-readable media (a non-exhaustive list) include the following: an electrical connection with one or more wires (electronic device), a portable computer disk box (magnetic device), a random access memory (RAM), a read-only memory (ROM), an erasable and programmable read-only memory (EPROM or flash memory), a fiber optic device, and a portable compact disk read-only memory (CDROM). In addition, the computer-readable medium may even be paper or other suitable medium on which the program is printed, since the program may be obtained electronically, for example, by optically scanning the paper or other medium and then editing, interpreting or processing in other suitable ways if necessary, and then stored in a computer memory.

[0191] It should be understood that the various parts of the present application can be implemented by hardware, software, firmware or a combination thereof. In the above-mentioned embodiments, multiple steps or methods can be implemented by software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented by hardware, as in another embodiment, it can be implemented by any one of the following technologies known in the art or their combination: a discrete logic circuit having a logic gate circuit for implementing a logic function for a data signal, a dedicated integrated circuit having a suitable combination of logic gate circuits, a programmable gate array (PGA), a field programmable gate array (FPGA), etc.

[0192] A person skilled in the art may understand that all or part of the steps in the method for implementing the above-mentioned embodiment may be completed by instructing related hardware through a program, and the program may be stored in a computer-readable storage medium, which, when executed, includes one or a combination of the steps of the method embodiment.

[0193] In addition, each functional unit in each embodiment of the present application may be integrated into a processing module, or each unit may exist physically separately, or two or more units may be integrated into one module. The above-mentioned integrated module may be implemented in the form of hardware or in the form of a software functional module. If the integrated module is implemented in the form of a software functional module and sold or used as an independent product, it may also be stored in a computer-readable storage medium.

[0194] The storage medium mentioned above may be a read-only memory, a magnetic disk or an optical disk, etc. Although the embodiments of the present application have been shown and described above, it can be understood that the above embodiments are exemplary and cannot be understood as limiting the present application. A person of ordinary skill in the art may change, modify, replace and modify the above embodiments within the scope of the present application.

Claims

1. A method for analyzing capital flow, characterized in that: The following steps are involved: Acquire the capital flow time series data of the target capital flow indicator, and process the capital flow time series data to obtain processed target time series data and characteristic data; Using a first time series prediction algorithm, predicting the capital flow based on the target time series data, to obtain first capital flow prediction sequence data; Using a second time series prediction algorithm, predicting the capital flow based on the target time series data and the characteristic data, to obtain second capital flow prediction sequence data; The first capital flow prediction sequence data and the second capital flow prediction sequence data are integrated to obtain capital flow prediction result data; The abnormal root cause analysis of the cash flow forecast result data is performed through a multi-dimensional data indicator root cause system to determine the abnormal root cause elements of the cash flow.

2. The method according to claim 1, characterized in that: The processing of the capital flow time series data to obtain characteristic data includes: Based on the temporal order of the capital flow time series data, characteristic data of the capital flow time series data is retroactively constructed.

3. The method according to claim 2, characterized in that The step of retrospectively constructing characteristic data of the capital flow time series data based on the time sequence of the capital flow time series data comprises: Get trend correlation between multiple money flow indicators; Based on the trend correlation between the multiple capital flow indicators, relevant data of the first capital flow indicator that has a trend correlation with the target capital flow indicator is used as feature data of the capital flow time series data.

4. The method according to claim 1 or 2, characterized in that: The processing of the capital flow time series data to obtain processed target time series data includes: Performing change point detection on the capital flow time series data to obtain a target change point of the capital flow time series data; In chronological order, the sequence values ​​of N time points before the target change point are updated to a first value, and the sequence values ​​of M time points after the target change point are updated to a second value, to obtain processed target time series data.

5. The method according to claim 1, characterized in that The first time series prediction algorithm uses a decomposable time series model for prediction, and the second time series prediction algorithm uses a complex neural network model for prediction.

6. The method according to claim 1, characterized in that The step of fusing the first capital flow prediction sequence data and the second capital flow prediction sequence data to obtain capital flow prediction result data comprises: The first capital flow prediction sequence data and the second capital flow prediction sequence data are integrated by weighted average method to obtain capital flow prediction result data.

7. The method according to claim 5, characterized in that The decomposable time series model is an additive Prophet model, which includes the superposition of trend terms, seasonal terms, holiday terms and noise terms; wherein the trend term is used to fit the non-periodic trend changes of the capital flow time series, and the seasonal term is used to fit the weekly, monthly and quarterly cyclical change trends of the capital flow time series, using Fourier series for approximation; the holiday term represents the impact of the potential jump points of the fitting capital flow time series on the prediction, and the noise term is used to represent the random fluctuations of the unpredicted capital flow time series.

8. The method according to claim 5, characterized in that The complex neural network model is a NBEATSx model, which includes a plurality of double residual stacks, each of which includes a plurality of basic blocks, and the basic blocks are explanatory trend and seasonal models; the second time series prediction algorithm is used to predict the capital flow based on the target time series data and the feature data to obtain the second capital flow prediction series data; including: Inputting the target time series data and the feature data into the basic Block, and transforming the target time series data and the feature data into a hidden state through the fully connected layer of the basic Block; An expansion coefficient is generated based on the implicit state, and the expansion coefficient is mapped into time series data to obtain second capital flow prediction series data.

9. The method according to claim 8, characterized in that Different double residual stacks respectively use harmonic functions, polynomial trend terms and external covariates to fit the information related to cycles, trends and external information in the fund flow time series.

10. A computer program product, characterized in that The invention comprises a computer program, which implements the method according to any one of claims 1 to 9 when being executed by a processor.

Images