Risk early warning method and device, electronic equipment and storage medium

Abstract

The embodiment of the invention provides a risk early warning method and device, electronic equipment and a storage medium, and belongs to the technical field of financial science and technology and artificial intelligence. The method comprises the following steps: screening target risk index data from product risk index data according to product risk value data; and performing risk early warning on the target risk index data by using the risk early warning model to obtain a product value risk early warning level. According to the embodiment of the invention, the target risk index data is screened from the product risk index data according to the product risk value data, the product risk index data irrelevant to the product risk value data can be removed, and the accuracy of subsequent risk early warning can be improved; and secondly, product value risk early warning is carried out on the target risk index data by using the risk early warning model, risk early warning can be carried out for a nonlinear relationship between complex time sequences in the risk data, and the accuracy of risk early warning is improved.

Description

Technical Field

[0001] The present application relates to the fields of financial technology and artificial intelligence technology, and in particular to a risk warning method and device, electronic equipment and storage medium. Background Art

[0002] At present, risk warning is usually carried out on labeled risk data samples through neural networks (such as convolutional neural networks (CNN)). For example, in the risk warning process of financial products, the fluctuation of bond interest rates will affect the market value of bonds. For companies holding a large amount of bond investments, it is necessary to evaluate the investment risk of bond investment caused by changes in bond interest rates, so that companies can adjust their investment strategies based on investment risks. However, this method only analyzes the linear relationship between risk data samples during risk warning, and it is difficult to analyze the nonlinear relationship of complex time series in risk data samples, resulting in a low accuracy rate of risk warning. Therefore, how to improve the accuracy of risk warning has become an urgent problem to be solved. Summary of the invention

[0003] The main purpose of the embodiments of the present application is to propose a risk warning method and device, an electronic device and a storage medium, aiming to improve the accuracy of risk warning.

[0004] To achieve the above objectives, a first aspect of an embodiment of the present application proposes a risk warning method, the method comprising:

[0005] Acquire initial risk data; wherein the initial risk data is time-series-based data, and the initial risk data includes product risk indicator data and product risk value data;

[0006] Filtering target risk indicator data from the product risk indicator data according to the product risk value data; wherein the target risk indicator data is data that generates risk for the product risk value data;

[0007] Obtaining a pre-trained risk warning model; wherein the risk warning model includes a state detection gate, a forget gate, and an activation gate;

[0008] Using the state detection gate to perform state detection on the target risk indicator data to obtain an indicator risk state; the indicator risk state is used to indicate the risk level of the target risk indicator data;

[0009] Using the forget gate to perform abnormal index filtering on the target risk index data to obtain normal risk index data;

[0010] The activation gate is used to activate the indicator risk status and the normal data of the risk indicator to obtain the product value risk warning level of the product risk value sample.

[0011] In some embodiments, the activation gate is used to activate the indicator risk status and the normal data of the risk indicator to obtain the product value risk warning level of the product risk value sample, including:

[0012] The risk indicator risk status is updated according to the normal data of the risk indicator to obtain an updated risk indicator status;

[0013] Obtaining an output weight matrix and an output bias term of the activation gate;

[0014] The update indicator risk status and the risk indicator normal data are activated according to a preset activation function, the output weight matrix and the output bias term to obtain the product value risk warning level of the product risk value sample.

[0015] In some embodiments, the step of filtering out target risk indicator data from the product risk indicator data according to the product risk value data includes:

[0016] Obtaining the correlation between the product risk indicator data at a preset moment and the product risk value data at the preset moment, and obtaining the product risk data correlation relationship; the product risk data correlation relationship is used to indicate that: the product risk indicator data generates risk to the product risk value data, the product risk value data generates risk to the product risk indicator data, and the product risk indicator data and the product risk value data generate risk to each other;

[0017] The target risk indicator data is filtered out from the product risk indicator data according to the product risk data association relationship.

[0018] In some embodiments, the acquiring the association relationship between the product risk indicator data at a preset time and the product risk value data at the preset time to obtain the product risk data association relationship includes:

[0019] Obtaining the time lag parameter of the product risk indicator data at the preset time;

[0020] Acquire a first lag item of the product risk indicator data at the preset moment, and a first lag parameter of the first lag item;

[0021] Acquire a second lag item of the product risk value data at the preset moment, and a second lag parameter of the second lag item;

[0022] Constructing a first time series model for the time lag parameter, the first lag term, the first lag parameter, the second lag term and the second lag parameter to obtain a first time series model;

[0023] Constructing a second time series model for the time lag parameter, the first lag term, the first lag parameter, the second lag term and the second lag parameter to obtain a second time series model;

[0024] The association relationship between the product risk indicator data at the preset moment and the product risk value data at the preset moment is obtained according to the first time series model and the second time series model to obtain a product risk data association relationship.

[0025] In some embodiments, after obtaining the association relationship between the product risk indicator data at the preset time and the product risk value data at the preset time according to the first time series model and the second time series model, and obtaining the product risk data association relationship, the method includes:

[0026] Obtaining first residual data of the first time series model, and obtaining second residual data of the second time series model;

[0027] For the first time series model, when the second lag parameter is not zero and the first residual data changes, the product risk data association relationship is a unidirectional risk association relationship;

[0028] For the second time series model, when the first lag parameter is not zero and the second residual data changes, the product risk data association relationship is a reverse risk association relationship;

[0029] For the first time series model and the second time series model, when the first lag parameter and the second lag parameter are not zero and the first residual data and the second residual data change, the product risk data association relationship is a bidirectional risk association relationship.

[0030] In some embodiments, the acquiring the association relationship between the product risk indicator data at a preset time and the product risk value data at the preset time to obtain the product risk data association relationship further includes:

[0031] Obtaining the current time window, and performing time series fluctuation amplitude detection on the product risk value data according to the current time window to obtain the current window time series fluctuation limit;

[0032] According to the current time window, the correlation relationship between the product risk indicator data at the preset time and the product risk value data at the preset time is obtained within the time series fluctuation limit of the current window to obtain the current window correlation relationship;

[0033] Obtaining an adjacent forward time window of the current time window, and performing a time series fluctuation amplitude detection on the product risk value data according to the adjacent forward time window to obtain a forward window time series fluctuation limit;

[0034] According to the adjacent forward time windows, the correlation relationship between the product risk indicator data at the preset time and the product risk value data at the preset time is obtained within the forward window time series fluctuation limit to obtain a forward window correlation relationship;

[0035] Obtaining an adjacent backward time window of the current time window, and performing a time series fluctuation amplitude detection on the product risk value data according to the adjacent backward time window to obtain a backward window time series fluctuation limit;

[0036] According to the adjacent backward time windows, the correlation relationship between the product risk indicator data at the preset time and the product risk value data at the preset time is obtained within the backward window time series fluctuation limit to obtain a backward window correlation relationship;

[0037] The current window association relationship, the forward window association relationship and the backward window association relationship are merged to obtain the product risk data association relationship.

[0038] In some embodiments, after obtaining the initial risk data, the method further includes:

[0039] Normalizing the initial risk data to obtain normalized risk data;

[0040] Performing a time trend test on the normalized risk data to obtain stable time trend risk data;

[0041] Performing a unit root test on the normalized risk data to obtain stable unit root risk data;

[0042] The stationary time trend risk data and the stationary unit root risk data are fused to obtain fused test risk data, and the fused test risk data is used as the initial risk data.

[0043] To achieve the above-mentioned purpose, a second aspect of an embodiment of the present application provides a risk warning device, the device comprising:

[0044] A risk data acquisition module, used to acquire initial risk data; wherein the initial risk data is time-series-based data, and the initial risk data includes product risk indicator data and product risk value data;

[0045] An indicator data screening module, used for screening target risk indicator data from the product risk indicator data according to the product risk value data; wherein the target risk indicator data is data that generates risk to the product risk value data;

[0046] The early warning model acquisition module is used to acquire a pre-trained risk early warning model; wherein the risk early warning model includes a state detection gate, a forget gate and an activation gate;

[0047] A state detection module, used to perform state detection on the target risk indicator data using the state detection gate to obtain an indicator risk state;

[0048] An abnormal indicator filtering module is used to use the forget gate to perform abnormal indicator filtering on the target risk indicator data to obtain normal risk indicator data;

[0049] The risk warning module is used to use the activation gate to activate the indicator risk status and the normal data of the risk indicator to obtain the product value risk warning level of the product risk value sample.

[0050] To achieve the above objectives, a third aspect of an embodiment of the present application proposes an electronic device, which includes a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, the method described in the first aspect is implemented.

[0051] To achieve the above-mentioned purpose, the fourth aspect of an embodiment of the present application proposes a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the method described in the first aspect is implemented.

[0052] The risk warning method and device, electronic device and storage medium proposed in the present application obtain initial risk data; wherein the initial risk data is time-series data, and the initial risk data includes product risk indicator data and product risk value data; target risk indicator data is screened out from the product risk indicator data according to the product risk value data; wherein the target risk indicator data is data that generates risks for the product risk value data; a pre-trained risk warning model is obtained; wherein the risk warning model includes a state detection gate, a forget gate and an activation gate; the state detection gate is used to perform state detection on the target risk indicator data to obtain the indicator risk state; the indicator risk state is used to indicate the risk degree of the target risk indicator data; the forget gate is used to perform abnormal indicator filtering processing on the target risk indicator data to obtain normal risk indicator data; the activation gate is used to activate the indicator risk state and the normal risk indicator data to obtain the product value risk warning level of the product risk value sample. This application firstly filters out target risk indicator data from product risk indicator data according to product risk value data, and can remove product risk indicator data irrelevant to product risk value data, which helps to improve the accuracy of subsequent risk warnings; secondly, the state detection gate of the risk warning model is used to perform state detection on the target risk indicator data, which can preliminarily identify the risk level of the risk indicator, and use the forgetting gate to filter out abnormal indicators on the target risk indicator data, so as to further improve the data quality of the risk indicator; finally, the activation gate is used to activate the indicator risk state and the normal data of the risk indicator, which can provide risk warnings for the nonlinear relationship between complex time series in the risk data, thereby improving the accuracy of risk warnings. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] Figure 1 is a flow chart of the risk warning method provided in an embodiment of the present application;

[0054] Figure 2 is another flow chart of the risk warning method provided by an embodiment of the present application;

[0055] Figure 3 yes Figure 1 Flow chart of step S102 in FIG.

[0056] Figure 4 yes Figure 3 Flow chart of step S301 in FIG.

[0057] Figure 5 yes Figure 4 Flow chart of step S406 in FIG.

[0058] Figure 6 yes Figure 3 Flow chart of step S301 in FIG.

[0059] Figure 7 yes Figure 1 Flow chart of step S106 in FIG.

[0060] Figure 8 is a schematic diagram of the structure of a risk warning device provided in an embodiment of the present application;

[0061] Fig. 9 It is a schematic diagram of the hardware structure of the electronic device provided in the embodiment of the present application. DETAILED DESCRIPTION

[0062] In order to make the purpose, technical solution and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0063] It should be noted that, although the functional modules are divided in the device schematic diagram and the logical order is shown in the flowchart, in some cases, the steps shown or described may be performed in a different order than the module division in the device or the order in the flowchart. The terms "first", "second", etc. in the specification, claims and the above drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence.

[0064] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as those commonly understood by those skilled in the art to which this application belongs. The terms used herein are only for the purpose of describing the embodiments of this application and are not intended to limit this application.

[0065] First, some nouns involved in this application are analyzed:

[0066] Artificial intelligence (AI) is a new technical science that studies and develops theories, methods, technologies and application systems for simulating, extending and expanding human intelligence. AI is a branch of computer science that attempts to understand the essence of intelligence and produce a new type of intelligent machine that can respond in a similar way to human intelligence. Research in this field includes robotics, speech recognition, image recognition, natural language processing and expert systems. AI can simulate the information process of human consciousness and thinking. AI is also a theory, method, technology and application system that uses digital computers or machines controlled by digital computers to simulate, extend and expand human intelligence, perceive the environment, acquire knowledge and use knowledge to obtain the best results.

[0067] Based on this, the embodiments of the present application provide a risk warning method and device, an electronic device and a storage medium, aiming to improve the accuracy of risk warning.

[0068] The risk warning method and device, electronic device and storage medium provided in the embodiments of the present application are specifically illustrated through the following embodiments. First, the risk warning method in the embodiments of the present application is described.

[0069] The embodiments of the present application can acquire and process relevant data based on artificial intelligence technology. Among them, artificial intelligence (AI) is the theory, method, technology and application system that uses digital computers or machines controlled by digital computers to simulate, extend and expand human intelligence, perceive the environment, acquire knowledge and use knowledge to obtain the best results.

[0070] AI basic technologies generally include sensors, dedicated AI chips, cloud computing, distributed storage, big data processing technology, operation / interaction systems, mechatronics, etc. AI software technologies mainly include computer vision technology, robotics technology, biometrics technology, speech processing technology, natural language processing technology, and machine learning / deep learning.

[0071] The risk warning method provided in the embodiment of the present application relates to the fields of financial technology and artificial intelligence technology. The risk warning method provided in the embodiment of the present application can be applied to the terminal, can also be applied to the server side, and can also be software running in the terminal or the server side. In some embodiments, the terminal can be a smart phone, a tablet computer, a laptop computer, a desktop computer, etc.; the server side can be configured as an independent physical server, or it can be configured as a server cluster or distributed system composed of multiple physical servers, and can also be configured as a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, CDN, and big data and artificial intelligence platforms; the software can be an application that implements the risk warning method, etc., but is not limited to the above forms.

[0072] The present application can be used in many general or special computer system environments or configurations. For example: personal computers, server computers, handheld or portable devices, tablet devices, multiprocessor systems, microprocessor-based systems, set-top boxes, programmable consumer electronics, network PCs, minicomputers, mainframe computers, distributed computing environments including any of the above systems or devices, etc. The present application can be described in the general context of computer-executable instructions executed by a computer, such as program modules. Generally, program modules include routines, programs, objects, components, data structures, etc. that perform specific tasks or implement specific abstract data types. The present application can also be practiced in distributed computing environments, in which tasks are performed by remote processing devices connected through a communication network. In a distributed computing environment, program modules can be located in local and remote computer storage media including storage devices.

[0073] Figure 1 is an optional flow chart of the risk warning method provided in the embodiment of the present application. Figure 1 The method may include but is not limited to steps S101 to S106.

[0074] Step S101, obtaining initial risk data; wherein the initial risk data is time-series-based data, and the initial risk data includes product risk indicator data and product risk value data.

[0075] Step S102, filtering out target risk indicator data from the product risk indicator data according to the product risk value data; wherein the target risk indicator data is data that generates risks to the product risk value data.

[0076] Step S103, obtaining a pre-trained risk warning model; wherein the risk warning model includes a state detection gate, a forget gate and an activation gate.

[0077] Step S104, using a state detection gate to perform state detection on the target risk indicator data to obtain an indicator risk state; the indicator risk state is used to indicate the risk level of the target risk indicator data.

[0078] Step S105, using a forget gate to perform abnormal index filtering processing on the target risk index data to obtain normal risk index data.

[0079] Step S106, using the activation gate to activate the indicator risk status and the normal data of the risk indicator to obtain the product value risk warning level of the product risk value sample.

[0080] Steps S101 to S106 shown in the embodiment of the present application are performed by obtaining initial risk data; wherein the initial risk data is time-series data, and the initial risk data includes product risk indicator data and product risk value data; target risk indicator data is screened out from the product risk indicator data according to the product risk value data; wherein the target risk indicator data is data that generates risks for the product risk value data; a pre-trained risk warning model is obtained; wherein the risk warning model includes a state detection gate, a forget gate and an activation gate; the state detection gate is used to perform state detection on the target risk indicator data to obtain an indicator risk state; the indicator risk state is used to indicate the risk level of the target risk indicator data; the forget gate is used to perform abnormal indicator filtering processing on the target risk indicator data to obtain normal risk indicator data; the activation gate is used to activate the indicator risk state and the normal risk indicator data to obtain the product value risk warning level of the product risk value sample. This application firstly filters out target risk indicator data from product risk indicator data according to product risk value data, and can remove product risk indicator data irrelevant to product risk value data, which helps to improve the accuracy of subsequent risk warnings; secondly, the state detection gate of the risk warning model is used to perform state detection on the target risk indicator data, which can preliminarily identify the risk level of the risk indicator, and use the forgetting gate to filter out abnormal indicators on the target risk indicator data, so as to further improve the data quality of the risk indicator; finally, the activation gate is used to activate the indicator risk state and the normal data of the risk indicator, which can provide risk warnings for the nonlinear relationship between complex time series in the risk data, thereby improving the accuracy of risk warnings.

[0081] In step S101 of some embodiments, specifically, the initial risk data is time-series-based data, and the initial risk data includes product risk indicator data and product risk value data.

[0082] Specifically, product risk indicator data refers to quantitative indicators that change over time and are used to measure and evaluate the risks that a product may face during its operation; product risk value data refers to product value data that changes over time and is used to measure and evaluate the product value when a product faces risks.

[0083] For example, in the insurance field, product risk indicator data may include user credit scores, historical claims records, historical claims amounts, insurance premiums, foreign exchange rates, certificate of deposit rates, bill interest rates, GDP growth rates, inflation rates, fixed asset investment growth rates, non-performing loan rates, asset returns, average stock market price-earnings ratios, stock prices, total foreign debts and short-term foreign debts and other product indicators of insurance products within three years.

[0084] For example, in the insurance field, product risk value data may include insurance claim costs, bond prices, bond yields, and insurance yields of insurance products within three years.

[0085] Specifically, product risk indicator data and product risk value data can be obtained from the enterprise backend database.

[0086] See also Figure 2 In some embodiments, after step S101, the risk warning method includes but is not limited to steps S201 to S204:

[0087] Step S201, normalizing the initial risk data to obtain normalized risk data.

[0088] Step S202, performing a time trend test on the normalized risk data to obtain stable time trend risk data.

[0089] Step S203, performing a unit root test on the normalized risk data to obtain stationary unit root risk data.

[0090] Step S204, the stationary time trend risk data and the stationary unit root risk data are fused to obtain fused test risk data, and the fused test risk data is used as the initial risk data.

[0091] In step S201 of some embodiments, normalizing the risk data refers to converting the initial risk data into standard data in the interval [0, 1].

[0092] In step S202 of some embodiments, the time trend risk data refers to the stationarity data obtained after performing a KPSS (Kwiatkowski-Phillips-Schmidt-Shin Test) stationarity test on the time trend of the normalized risk data.

[0093] Specifically, in the KPSS test, there are null hypothesis and alternative hypothesis. The null hypothesis is that the time series trend of the normalized risk data is stationary, which means that after removing a certain time trend, the remaining residual time series is stationary; the alternative hypothesis is that the time series trend of the normalized risk data is non-stationary, which means that the fluctuation of the time series increases with time, making the time series not trend stationary.

[0094] Specifically, the KPSS statistic can be calculated by statistical software or programming tools, and the statistic is actually based on the square of the cumulative sum of the residual time series.

[0095] Furthermore, the KPSS statistic is compared with the preset KPSS critical value. If the KPSS statistic is greater than the KPSS critical value, the null hypothesis is rejected and the trend of the time series is considered to be non-stationary. If the KPSS statistic is less than or equal to the KPSS critical value, the null hypothesis cannot be rejected, that is, the statistical trend of the time series is considered to be stationary. The non-stationary data is removed to obtain the final stationary time trend risk data.

[0096] In step S203 of some embodiments, the stationary unit root risk data refers to data obtained by performing an ADF (Augmented Dickey-Fuller Test) stationarity test on the unit root of the normalized risk data.

[0097] Specifically, in the ADF test, there are the null hypothesis of unit root and the alternative hypothesis of unit root. The null hypothesis of unit root is that the time series of normalized risk data has a unit root, that is, the trend of the time series is non-stationary; the alternative hypothesis of unit root is that the time series of normalized risk data does not have a unit root, that is, the trend of the time series is stationary.

[0098] Specifically, the ADF statistic can be calculated by using statistical software or programming tools (such as R language, EViews, etc.).

[0099] Furthermore, the ADF statistic is compared with the preset ADF critical value. If the ADF statistic is less than the critical value, the null hypothesis is rejected and the trend of the time series is considered to be stationary. If the ADF statistic is greater than or equal to the ADF critical value, the null hypothesis is accepted, that is, the statistical trend of the time series is considered to be non-stationary. The non-stationary data is removed to obtain the final stationary unit root risk data.

[0100] In step S204 of some embodiments, specifically, if the ADF test rejects the null hypothesis of the unit root and the KPSS test cannot reject the null hypothesis, it is considered that the trend of the time series is statistically stationary, because the ADF test shows that the time series does not have a unit root, and the KPSS test does not detect obvious non-stationarity.

[0101] Furthermore, if the ADF test accepts the null hypothesis of unit root and the KPSS test rejects the null hypothesis, it indicates that the trend of the time series is considered to be statistically non-stationary.

[0102] In this embodiment, by fusing the steady time trend risk data and the steady unit root risk data, since the steady time trend risk data mainly focuses on the time trend of the normalized risk data, while the steady unit root risk data focuses on the test of the unit root, the two test methods are combined to verify each other, thereby improving the accuracy and reliability of the test results.

[0103] See also Figure 3 In some embodiments, step S102 may include but is not limited to steps S301 to S302:

[0104] Step S301, obtain the correlation between the product risk indicator data at a preset moment and the product risk value data at a preset moment, and obtain the product risk data correlation relationship; the product risk data correlation relationship is used to indicate that: the product risk indicator data generates risk to the product risk value data, the product risk value data generates risk to the product risk indicator data, and the product risk indicator data and the product risk value data generate risk to each other.

[0105] Step S302: Filter out target risk indicator data from product risk indicator data according to the product risk data association relationship.

[0106] In step S301 of some embodiments, specifically, the preset time may be any time within the past three years.

[0107] Specifically, the product risk data association relationship can be used to indicate that: product risk indicator data generates risk to product risk value data, product risk value data generates risk to product risk indicator data, and product risk indicator data and product risk value data generate risk to each other.

[0108] For example, in the insurance field, the product risk data association relationship can generate risks for insurance claim costs through indicators such as user credit scores, historical claims records, historical claims amounts, and insurance premiums; it can also generate risks for insurance claim amounts, bond prices, bond yields, and insurance yields for insurance premiums; it can also generate risks between insurance claim amounts and insurance premiums.

[0109] See also Figure 4 In some embodiments, step S301 may include but is not limited to steps S401 to S406:

[0110] Step S401, obtaining the time lag parameter of the product risk indicator data at a preset time.

[0111] Step S402, obtaining a first lag item of product risk indicator data at a preset time, and a first lag parameter of the first lag item;

[0112] Step S403, obtaining a second lag item of the product risk value data at a preset time, and a second lag parameter of the second lag item.

[0113] Step S404 , constructing a first time series model for the time lag parameter, the first lag term, the first lag parameter, the second lag term and the second lag parameter to obtain a first time series model.

[0114] Step S405 , constructing a second time series model based on the time lag parameter, the first lag term, the first lag parameter, the second lag term and the second lag parameter to obtain a second time series model.

[0115] Step S406, obtaining the correlation between the product risk indicator data at a preset time and the product risk value data at a preset time according to the first time series model and the second time series model, and obtaining the product risk data correlation relationship.

[0116] In step S401 of some embodiments, the time lag parameter refers to the time lag step that affects the current product value.

[0117] For example, in the insurance field, when you need to predict the amount of health insurance claims for a company in the coming year, you can consider the health insurance claim records of the past three years (i.e., the time lag step).

[0118] In step S402 of some embodiments, specifically, the first lag item refers to the lagged product risk indicator data, and whether a risk is generated for the product risk value data at the current moment can be determined in combination with the first lag parameter.

[0119] For example, in the insurance field, the first lag item could be the amount of health insurance claims paid over the past three years.

[0120] Specifically, the first lag parameter refers to the degree of influence of the lag item of the product risk indicator data on the product risk value data at the current moment.

[0121] For example, in the insurance field, the first lag parameter can be the degree to which the amount of health insurance payments in the past three years affects the current insurance claim cost.

[0122] In step S403 of some embodiments, specifically, the second lag item refers to the lagged product risk value data, and the second lag parameter can be combined to determine whether a risk is generated for the product risk value data at the current moment or for the product risk indicator data at the current moment.

[0123] For example, in the insurance field, the second lag term could be the insurance claims amount over the past three years.

[0124] Specifically, the second lag parameter refers to the lag term of the product risk value data to the product risk value data at the current moment.

[0125] For example, in the insurance field, the second lag parameter may be the degree of influence of the insurance claim amount in the past three years on the insurance premium rate at the current moment.

[0126] In step S404 of some embodiments, the first timing model may be represented by the following formula:

[0127]

[0128] Among them, Y t represents the first time series model at time t; P represents the time lag parameter; j represents the lag time; a j represents the second lag parameter at the jth lag time; y t-j represents the second lag term at the tjth lag time; b j represents the first lag parameter at the jth lag time; x t-j represents the first lag term at the tjth lag time.

[0129] Specifically, the first time series model can allow external variables (i.e., product risk indicator data at historical moments) to be combined with product risk value data at historical moments to conduct risk assessment on product risk value data at the current moment, which helps to improve the accuracy of subsequent risk warnings.

[0130] In step S405 of some embodiments, the second timing model may be represented by the following formula:

[0131]

[0132] In, X t represents the second time series model at time t; P represents the time lag parameter; j represents the lag time; a j represents the second lag parameter at the jth lag time; y t-j represents the second lag term at the tjth lag time; b j represents the first lag parameter at the jth lag time; x t-j represents the first lag term at the tjth lag time.

[0133] Specifically, the second time series model can allow external variables (i.e., product risk value data at historical moments) to be combined with product risk index data at historical moments to conduct risk assessment on product risk index data at the current moment, which helps to improve the accuracy of subsequent risk warnings.

[0134] See also Figure 5 In some embodiments, step S406 may include but is not limited to steps S501 to S504:

[0135] Step S501, obtaining first residual data of a first time series model, and obtaining second residual data of a second time series model.

[0136] Step S502: for the first time series model, when the second lag parameter is not zero and the first residual data changes, the product risk data association relationship is a unidirectional risk association relationship.

[0137] Step S503: for the second time series model, when the first lag parameter is not zero and the second residual data changes, the product risk data association relationship is a reverse risk association relationship.

[0138] Step S504: for the first time series model and the second time series model, when the first lag parameter and the second lag parameter are not zero and the first residual data and the second residual data change, the product risk data association relationship is a bidirectional risk association relationship.

[0139] In step S501 of some embodiments, specifically, the first residual data refers to the difference between the predicted value and the true value of the first time series model.

[0140] For example, in the insurance field, the true value may be the insurance claim amount at the current moment, and the predicted value may be the insurance claim amount at the current moment predicted by the first time series model.

[0141] Specifically, the second residual data refers to the difference between the predicted value and the true value of the second time series model.

[0142] For example, in the insurance field, the true value may be the insurance premium rate at the current moment, and the predicted value may be the insurance premium rate at the current moment predicted by the second time series model.

[0143] In step S502 of some embodiments, specifically, the unidirectional risk association relationship indicates that the product risk indicator data generates risk to the product risk value data.

[0144] For example, in the insurance field, product risk indicator data in the past three years (such as user credit scores, historical claims records, historical claims amounts, insurance premiums, etc.) have a risk impact on the insurance claim amount of the insurance product at the current moment.

[0145] Specifically, since the second lag parameter corresponds to the second lag item of the product risk indicator data, when the second lag parameter is not zero, it means that the second lag item is also not zero.

[0146] Specifically, the first residual data is added to the first time series model. If the first residual data decreases or increases due to the addition of the second lag term to the first time series model, it can be determined that the second lag term has a risk impact on the product risk value data.

[0147] For example, in the insurance field, the second lag parameter b2 at the second lag time is not zero, indicating that x t-2 The corresponding historical compensation amount is not zero, and x is excluded t-2 The corresponding historical compensation amount will cause the first residual data to decrease, which means that the historical compensation amount has a risk impact on the insurance claim amount of the insurance product at the current moment.

[0148] In this embodiment, for the first time series model, when the second lag parameter is not zero and the first residual data changes, the product risk data association relationship is a unidirectional risk association relationship, which can clarify the path and direction of product risk transmission, help to more clearly understand the spread of risks between different products, and facilitate the subsequent improvement of the accuracy of risk warnings.

[0149] In step S503 of some embodiments, specifically, the reverse risk association relationship indicates that the product risk value data generates risk to the product risk indicator data.

[0150] For example, in the insurance field, product risk value data over the past three years (such as insurance claims, bond prices, bond yields, and insurance yields, etc.) have a risk impact on the insurance premiums of insurance products at the current moment.

[0151] Specifically, since the first lag parameter corresponds to the first lag item of the product risk value data, when the first lag parameter is not zero, it means that the first lag item is also not zero.

[0152] Specifically, the second residual data is added to the second time series model. If the second residual data decreases or increases due to the addition of the first lag term to the second time series model, it can be determined that the first lag term has a risk impact on the product risk value data.

[0153] For example, in the insurance field, the first lag parameter a3 at the third lag time is not zero, indicating that Y t-3 The corresponding insurance claim amount is not zero, and Y is eliminated t-3 The corresponding insurance claim amount will cause the second residual data to decrease, which means that the insurance claim amount has a risk impact on the insurance premium rate of the insurance product at the current moment.

[0154] In this embodiment, for the second time series model, when the first lag parameter is not zero and the second residual data changes, the product risk data association relationship is a reverse risk association relationship. When a change in the risk indicator of a time series model is detected, risk assessment and early warning can be quickly performed on another time series model, thereby improving the efficiency of risk early warning. Product investment or configuration can also be made in two areas where risks change in opposite directions to balance the overall product risk exposure and ensure product value.

[0155] In step S504 of some embodiments, specifically, the bidirectional risk association relationship indicates that the product risk value data and the product risk indicator data generate risks mutually.

[0156] For example, in the insurance field, product risk value data (such as insurance claim amounts, etc.) in the past three years have a risk impact on the insurance premiums of insurance products at the current moment, and product risk indicator data (such as insurance premiums) in the past three years have an impact on the insurance claim amounts of insurance products at the current moment.

[0157] Specifically, the associated second lag term with a unidirectional wind direction correlation in the first time series model is extracted, and the associated first lag term with a reverse correlation in the second time series model is extracted, and the overlapping part of the associated first lag term and the associated second lag term is extracted, so as to determine the bidirectional risk correlation.

[0158] For example, in the insurance field, Y t-3 The corresponding insurance claim amount has a risk impact on the insurance premium rate of the insurance product at the current moment, X t-3 The corresponding insurance premium rate affects the insurance claim amount of the insurance product at the current moment, indicating that X t-3 The corresponding insurance premium rate and Y t-3 There is a two-way risk correlation relationship between the corresponding insurance claim amounts.

[0159] In this embodiment, for the first time series model and the second time series model, when the first lag parameter and the second lag parameter are not zero and the first residual data and the second residual data change, the product risk data association relationship is a two-way risk association relationship. Since the lag parameter exists and is not zero, it means that the risk impact has a certain time delay. By detecting this two-way risk association relationship, product risk assessment and early warning can be carried out in advance when the insurance risk signal first appears, thereby improving the efficiency of risk warning.

[0160] See also Figure 6 In some embodiments, step S301 may also include but is not limited to steps S601 to S606:

[0161] Step S601, obtain the current time window, and perform time series fluctuation amplitude detection on the product risk value data according to the current time window to obtain the time series fluctuation limit of the current window.

[0162] Step S602, according to the current time window, obtain the correlation between the product risk indicator data at the preset time and the product risk value data at the preset time within the current window time series fluctuation limit to obtain the current window correlation.

[0163] Step S603, obtaining the adjacent forward time window of the current time window, and performing time series fluctuation amplitude detection on the product risk value data according to the adjacent forward time window to obtain the forward window time series fluctuation limit.

[0164] Step S604, according to adjacent forward time windows, obtain the correlation between the product risk indicator data at a preset time and the product risk value data at a preset time within the forward window time series fluctuation limit to obtain the forward window correlation.

[0165] Step S605, obtaining the adjacent backward time window of the current time window, and performing time series fluctuation amplitude detection on the product risk value data according to the adjacent backward time window to obtain the backward window time series fluctuation limit.

[0166] Step S606, according to adjacent backward time windows, obtain the correlation between the product risk indicator data at a preset time and the product risk value data at a preset time within the backward window time series fluctuation limit to obtain the backward window correlation.

[0167] Step S607, the current window association relationship, the forward window association relationship and the backward window association relationship are merged to obtain the product risk data association relationship.

[0168] In step S601 of some embodiments, specifically, the current time window is used to focus on a specific time period of the product risk value data.

[0169] Specifically, the current window time series fluctuation limit includes the upper fluctuation limit and the lower fluctuation limit. The upper fluctuation limit can be used to represent the average fluctuation range of adjacent intervals between elements of the product risk value data time series; the lower fluctuation limit can be used to describe the fluctuation range between two time points between elements of the product risk value data time series.

[0170] Specifically, the upper boundary of fluctuation can be expressed by the following formula:

[0171]

[0172] Among them, D Wt represents the upper boundary of the fluctuation of the current window time series fluctuation limit of the current time window at time t; Wt represents the current time window at time t; m represents the size of the current time window; k represents the iteration variable with an iteration step of 1, which is used to traverse the adjacent elements of the time series of the product risk value data of the current time window.

[0173] Specifically, the fluctuation lower boundary can be expressed by the following formula:

[0174]

[0175] Among them, D′ Wtrepresents the lower boundary of the fluctuation of the current window time series fluctuation limit of the current time window at time t; Wt represents the current time window at time t; m represents the size of the current time window; k represents the iteration variable with an iteration step of 2, which is used to traverse the elements of the time series of the product risk value data of the current time window.

[0176] Specifically, the upper boundary of fluctuation refers to gradually differentiating the indicator data of two adjacent risk products within the current time window until the pairwise difference of the indicator data of the risk products in the current time window is completed, and all differential values ​​are divided by the current time window size m to obtain the average differential value, which is the upper boundary of the current window time series fluctuation limit.

[0177] Specifically, the lower boundary of fluctuation refers to gradually differentiating the two risk product indicator data separated by two time points in the current time window until the pairwise difference of the risk product indicator data in the current time window is completed, and all differential values ​​are divided by the current time window size 2m to obtain the average differential value, which is the lower boundary of the current window time series fluctuation limit.

[0178] In this embodiment, the time series fluctuation amplitude of the product risk value data is detected according to the current time window, and the time series fluctuation limit of the current window is obtained. It is possible to focus on a specific time period in the product risk value data according to the window step and window size of the current time window, and comprehensively cover the entire time series through window sliding. It is possible to observe and analyze the changing trend of the product value risk in more detail, and accurately measure the time series fluctuation amplitude of the product risk value data to adapt to different time scales and data characteristics, thereby improving the accuracy of subsequent risk warnings.

[0179] In step S602 of some embodiments, specifically, by obtaining the correlation between the product risk indicator data at a preset moment and the product risk value data at a preset moment within the current window time series fluctuation limit, the method for obtaining the current window correlation relationship is consistent with the method for obtaining the correlation between the product risk indicator data at a preset moment and the product risk value data at a preset moment to obtain the product risk data correlation relationship, and will not be repeated here.

[0180] Specifically, by combining the fluctuation range of product risk value data within the specified current window time series fluctuation limit, more accurate product association relationships can be further obtained for the focused time period.

[0181] In step S603 of some embodiments, specifically, the adjacent forward time window is a time window added based on the current time window, and is used to further explore the volatility of the time series.

[0182] Specifically, the forward window time series fluctuation limit includes the forward fluctuation upper boundary and the forward fluctuation lower boundary. Among them, the forward fluctuation upper boundary can be used to describe the average fluctuation amplitude of the adjacent intervals between elements of the product risk value data in the forward window time period focused by the adjacent forward time window; the forward fluctuation lower boundary can be used to describe the fluctuation amplitude of the interval between two time points between elements of the product risk value data in the forward window time period focused by the adjacent forward time window.

[0183] Specifically, the method for obtaining the correlation between the product risk index data at a preset moment and the product risk value data at a preset moment within the forward window time series fluctuation limit is consistent with the method for obtaining the correlation between the product risk index data at a preset moment and the product risk value data at a preset moment within the window time series fluctuation limit, and will not be repeated here.

[0184] In this embodiment, by increasing the volatility of the forward time window exploration sequence, when significant fluctuations in the risk product value data are found, the range of window detection can be adjusted, which helps to timely update risk warning parameters, further capture potential risk signals, and facilitate improving the accuracy of subsequent risk warnings.

[0185] In step S604 of some embodiments, specifically, by obtaining the correlation between the product risk indicator data at a preset moment and the product risk value data at a preset moment within the forward window time series fluctuation limit, the method for obtaining the forward window correlation relationship is consistent with the method for obtaining the correlation between the product risk indicator data at a preset moment and the product risk value data at a preset moment to obtain the product risk data correlation relationship, and will not be repeated here.

[0186] In step S605 of some embodiments, specifically, the adjacent backward time window is a time window added based on the current time window, and is used to further explore the backward volatility of the time series.

[0187] Specifically, the backward window time series fluctuation limit includes the backward fluctuation upper boundary and the backward fluctuation lower boundary. Among them, the backward fluctuation upper boundary can be used to describe the average fluctuation amplitude of the adjacent intervals between elements of the product risk value data in the backward window time period focused by the adjacent backward time windows; the backward fluctuation lower boundary can be used to describe the fluctuation amplitude of the product risk value data between two time points in the backward window time period focused by the adjacent backward time windows.

[0188] Specifically, the method for obtaining the correlation between the product risk index data at a preset moment and the product risk value data at a preset moment within the backward window time series fluctuation limit is consistent with the method for obtaining the correlation between the product risk index data at a preset moment and the product risk value data at a preset moment within the window time series fluctuation limit, and will not be repeated here.

[0189] In this embodiment, by increasing the backward time window to explore the volatility of the sequence, when significant fluctuations in the risk product value data are found, potential risk signals can be further captured, thereby facilitating improving the accuracy of subsequent risk warnings.

[0190] In step S606 of some embodiments, specifically, by obtaining the correlation relationship between the product risk indicator data at a preset moment and the product risk value data at a preset moment within the backward window time series fluctuation limit, the method for obtaining the backward window correlation relationship is consistent with the method for obtaining the correlation relationship between the product risk indicator data at a preset moment and the product risk value data at a preset moment to obtain the product risk data correlation relationship, and will not be repeated here.

[0191] In step S607 of some embodiments, specifically, the forward fluctuation amplitude of the risk product value data described by the forward window time series fluctuation limit is compared with the current fluctuation amplitude of the risk product value data described by the current window time series fluctuation limit. If the forward fluctuation amplitude is greater than the current fluctuation amplitude, it means that the correlation relationship of the forward adjacent time window has been transformed compared with the correlation relationship of the current time window, so it is necessary to further integrate the correlation relationships of different windows to obtain the final product risk data correlation relationship.

[0192] Specifically, different weights can be assigned to different windows based on the significance of the product risk correlation relationship in different windows, and weighted fusion can be performed based on different weights. If the weighted fusion result is greater than 0, it means that the product risk indicator data has a positive impact on the product risk value data; if the weighted fusion result is equal to 0, it means that the product risk indicator data has no impact on the product risk value data; if the weighted fusion result is less than 0, it means that the product risk indicator data has a negative impact on the product risk value data.

[0193] For example, in the insurance field, insurance companies want to evaluate the impact of stock market fluctuations on insurance claims. The current window correlation shows that stock market fluctuations have a positive impact on claim costs, with a weight of 0.5 and an impact coefficient of 0.3. The forward window correlation shows that stock market fluctuations have no significant impact on claim costs, with a weight of 0.3 and an impact coefficient of 0.1. The backward window correlation shows that stock market fluctuations have a negative impact on claim costs, with a weight of 0.2 and an impact coefficient of -0.2.

[0194] Furthermore, different windows are standardized to determine that the current standard influence coefficient of the current window is 0.3 / (0.3-(-0.2)=0.6, the forward standard influence coefficient of the forward window is 0.1 / (0.3-(-0.2)=0.2, and the backward standard influence coefficient of the backward window is -0.2 / (0.3-(-0.2)=-0.4, and weighted fusion is performed based on the standard influence coefficient, that is, (0.5*0.6)+(0.3*0.2)+(0.2*(-0.4))=0.3+0.06-0.08=0.28. Since the result of weighted fusion is 0.28, which is greater than 0, it means that stock market fluctuations have a positive impact on the insurance claim amount.

[0195] In this embodiment, by fusing the current window association, the forward window association and the backward window association, the information of different time windows can be comprehensively considered, the detection deviation in the relationship conversion area is reduced, and the accuracy of obtaining product risk association is significantly improved, which facilitates the subsequent improvement of the accuracy of risk warning.

[0196] In step S202 of some embodiments, the target risk indicator data is data that generates risks for the product risk value data.

[0197] Specifically, based on the risk product association relationship, product risk indicator data that poses risks to product risk value data within a historical time period can be determined.

[0198] For example, in the insurance field, for vehicle insurance claims, since the historical claims amount has a risk impact on the insurance claims amount of the current insurance product, the historical claims amount is used as the target risk indicator data; and the vehicle type does not have a significant risk impact on the vehicle insurance claim amount, so the vehicle type is removed to achieve the screening of product risk indicator data.

[0199] In this embodiment, target risk indicator data is screened out from product risk indicator data based on the correlation relationship between product risk data, and product risk indicator data that is not related to product risk value data can be removed, thereby achieving more accurate identification and management of product value risks, which helps to improve the accuracy of subsequent risk warnings.

[0200] In step S103 of some embodiments, the pre-trained risk warning model may be a LSTM (Long Short-term Memory network) model.

[0201] Specifically, the risk warning model includes a state detection gate, a forget gate and an activation gate. The state detection gate is used to determine the risk level of the data sequence of the target risk indicator data within a period of time, and can preliminarily identify the risk level of the risk indicator; the forget gate is used to filter and process abnormal indicator data, thereby improving the data quality of the risk indicator; the activation gate is used to identify the product value risk warning level of the product risk value sample, and can provide risk warning for the nonlinear relationship between complex time series in the risk data, thereby improving the accuracy of risk warning.

[0202] Specifically, when training the risk warning model, first obtain the standard product value risk warning level of the product risk value sample, and use the state detection gate to perform state detection on the target risk indicator data to obtain the indicator risk state, then use the forgetting gate to filter out abnormal indicators on the target risk indicator data to obtain normal risk indicator data, and finally use the activation gate to activate the indicator risk state and the normal risk indicator data to obtain the predicted product value risk warning level of the product risk value sample, and use the loss function (such as the sigmoid function) to calculate the loss value of the standard product value risk warning level and the predicted product value risk warning level, and adjust the parameters of the risk warning model according to the loss value until the loss value meets the preset threshold, thereby obtaining the trained risk warning model.

[0203] In step S104 of some embodiments, the indicator risk status is used to represent the risk level of the target risk indicator data.

[0204] For example, the target risk indicator data is the historical claims data of insurance product A. The state detection gate identifies that the frequency of claims in the historical claims data in the past three years has been increasing year by year, indicating that the risk level of insurance product A is relatively high, that is, the indicator risk of insurance product A is in a high-risk state.

[0205] Specifically, the target risk indicator data can be tested for status using the following formula:

[0206]

[0207] Among them, i t represents the indicator risk status at time t, represents the sigmoid activation function, w i represents the input weight matrix of the input gate, h t-1 represents the indicator risk status of the target risk indicator data output at the input gate t-1, x t represents the target risk indicator data at time t, b i Represents the input bias term of the input gate.

[0208] In this embodiment, by utilizing the state detection gate to perform state detection on the target risk indicator data, the new information that should be added to the current unit state can be controlled through the input gate, thereby achieving selective absorption of new target risk indicator data, helping to preliminarily identify the risk level of the risk indicator and facilitating subsequent improvement in the accuracy of risk warnings.

[0209] In step S105 of some embodiments, the normal risk indicator data refers to the target risk indicator data after filtering the abnormal indicators.

[0210] For example, the target risk indicator data is the historical claims data of insurance product A. The forget gate identifies that the historical claims data has an abnormally high claim record in the past three years, and this abnormally high historical claim record was caused by a natural disaster. In this case, the historical claims data is not specific and representative for the assessment of future product risk value, and the abnormally high historical claim record is forgotten.

[0211] Specifically, the target risk indicator data can be filtered for abnormal indicators using the following formula:

[0212]

[0213] Among them, f t represents the normal data of risk indicators at time t, represents the sigmoid activation function, w f represents the forget weight matrix of the forget gate, represents the normal risk index data output by the target risk index data at the forget gate t-1, x t represents the target risk indicator data at time t, b f Represents the forget bias term of the forget gate.

[0214] Furthermore, the value range of the sigmoid function is between [0,1]. When the input value is greater than 0.5, it means that the target risk indicator data of the previous moment is completely retained; when the input value is less than 0.5, it means that the target risk indicator data of the previous moment is not retained at all.

[0215] In this embodiment, by using the forgetting gate to filter out abnormal indicators of the target risk indicator data, the target risk indicator data can be filtered out of abnormal indicators, which helps to eliminate outdated or no longer relevant risk factors, further improve the data quality of risk indicators, and reduce noise and interference information in the data, ensuring that the model can retain long-term data that is critical to risk warning, further improving the accuracy of risk warning.

[0216] In step S106 of some embodiments, the product value risk warning level of the product risk value sample includes a safety level, a low risk level, a medium risk level, and a high risk level.

[0217] For example, in the insurance field, if the growth of insurance claims output by the model is within the normal fluctuation range and there is no significant upward trend, it can be judged as a safe level; if the growth of insurance claims is slight but shows a certain upward trend, it can be judged as a low-risk level; if the growth of insurance claims reaches a medium level, or there is an adverse change in insurance premiums, it can be judged as a medium-risk level; if the growth of insurance claims is significant and similar to historical high-risk periods, it can be judged as a high-risk level.

[0218] See also Figure 7 In some embodiments, step S106 may include but is not limited to steps S701 to S703:

[0219] Step S701, updating the indicator risk state according to the normal data of the risk indicator to obtain the updated indicator risk state.

[0220] Step S702, obtaining the output weight matrix and output bias term of the activation gate.

[0221] Step S703, activating the update indicator risk status and the risk indicator normal data according to the preset activation function, output weight matrix and output bias term to obtain the product value risk warning level of the product risk value sample.

[0222] In step S701 of some embodiments, updating the indicator risk state is obtained by updating the indicator risk state based on the normal data of the risk indicator of the forgetting gate.

[0223] Specifically, the updated indicator risk status can be obtained by the following formula:

[0224]

[0225] Among them, c t represents the updated indicator risk status at time t, i t represents the risk status of the indicator at time t, h t-1 represents the indicator risk state of the target risk indicator data output at the input gate t-1, f t represents the normal data of risk indicators at time t, The target risk indicator data is the normal risk indicator data output by the forget gate at time t-1.

[0226] In this embodiment, by updating the indicator risk status according to the normal data of the risk indicator to obtain the updated indicator risk status, it is possible to selectively focus on the indicator data that may have a risk impact, which helps to subsequently improve the accuracy of risk warnings.

[0227] In step S702 of some embodiments, specifically, the output weight matrix refers to the weight matrix of the activation gate, and the output weight matrix is ​​composed of two matrices. Among them, the first matrix Q1 corresponds to the product value risk warning level output at time t-1, and the dimension of Q1 is the product of the vector dimension dc of the unit state and the vector dimension dh of the product value risk warning level; the second matrix Q2 corresponds to the input target risk indicator data, and the dimension of Q2 is the product of the vector dimension dc of the unit state and the vector dimension dx of the target risk indicator data.

[0228] Specifically, the output bias term is the baseline or starting point of the model, which is used to adjust the predicted value of the model to ensure that when the independent variable takes the value of zero, the predicted value is consistent with the actual baseline of the target variable.

[0229] In step S703 of some embodiments, the product value risk warning level of the product risk value sample can be obtained by the following formula:

[0230]

[0231] Among them, h t It represents the product value risk warning level of the product risk value sample at time t. represents the sigmoid activation function, w o represents the output weight matrix of the activation gate, h″ t-1 Indicates the product value risk warning level of the product risk value sample output at the activation gate t-1 time when the update indicator risk status and the normal data of the risk indicator are updated. t represents the target risk indicator data at time t, b o represents the output bias term of the activation gate, c t Represents the updated indicator risk state at time t. The tan function is used to introduce a linear bias to control the dynamic range of the unit state in the model.

[0232] For example, if the detection status gate outputs that the growth of the insurance claim amount of health insurance is within the normal fluctuation range (such as -5% to +5%), and the forgetting gate outputs that the trend of historical claim data is stable without a significant upward or downward trend, the activation gate outputs a safety level; if the detection status gate outputs that the insurance claim amount of health insurance has increased slightly (such as +5% to +10%), and the detection status gate outputs that the trend of historical claim data is stable and shows a slight increase, but has not reached a significant level, the activation gate outputs a low risk level; if the detection status gate outputs that the insurance claim amount of health insurance has increased to a medium level (such as +10% to +20%), and the forgetting gate outputs that the trend of historical claim data has shown an increase, or the current growth is similar to a certain period in history but has not reached a high risk level, the activation gate outputs a medium risk level; if the detection status gate outputs that the insurance claim amount has increased significantly (such as above +20%), which is similar to a historical high-risk period, and the forgetting gate outputs that the trend of historical claim data has shown a significant increase, or the current growth is consistent with the growth of a historical high-risk period, the activation gate outputs a high risk level.

[0233] In this embodiment, by activating the update indicator risk status and the risk indicator normal data according to the preset activation function, output weight matrix and output bias term, the intensity and importance of the output information can be controlled by the sigmoid function of the activation gate to ensure that the output risk warning signal is neither too sensitive nor too slow, so as to achieve the best warning effect, and adjust its own output according to the feedback results of the detection status gate and the forgetting gate, so as to carry out risk warning for the nonlinear relationship between the complex time series in the risk data, thereby improving the accuracy of risk warning.

[0234] The embodiment of the present application obtains initial risk data; wherein the initial risk data is time-series data, and the initial risk data includes product risk indicator data and product risk value data; target risk indicator data is screened out from the product risk indicator data according to the product risk value data; wherein the target risk indicator data is data that generates risks for the product risk value data; a pre-trained risk warning model is obtained; wherein the risk warning model includes a state detection gate, a forget gate and an activation gate; the state detection gate is used to perform state detection on the target risk indicator data to obtain an indicator risk state; the indicator risk state is used to indicate the risk degree of the target risk indicator data; the forget gate is used to perform abnormal indicator filtering processing on the target risk indicator data to obtain normal risk indicator data; the activation gate is used to activate the indicator risk state and the normal risk indicator data to obtain the product value risk warning level of the product risk value sample. This application firstly filters out target risk indicator data from product risk indicator data according to product risk value data, and can remove product risk indicator data irrelevant to product risk value data, which helps to improve the accuracy of subsequent risk warnings; secondly, the state detection gate of the risk warning model is used to perform state detection on the target risk indicator data, which can preliminarily identify the risk level of the risk indicator, and use the forgetting gate to filter out abnormal indicators on the target risk indicator data, so as to further improve the data quality of the risk indicator; finally, the activation gate is used to activate the indicator risk state and the normal data of the risk indicator, which can provide risk warnings for the nonlinear relationship between complex time series in the risk data, thereby improving the accuracy of risk warnings.

[0235] See also Figure 8 The embodiment of the present application also provides a risk warning device, which can implement the above-mentioned risk warning method, and the device includes:

[0236] A risk data acquisition module is used to acquire initial risk data; wherein the initial risk data is time-series-based data, and the initial risk data includes product risk indicator data and product risk value data;

[0237] The indicator data screening module is used to screen out target risk indicator data from the product risk indicator data according to the product risk value data; wherein the target risk indicator data is data that generates risks to the product risk value data;

[0238] The warning model acquisition module is used to obtain a pre-trained risk warning model; wherein the risk warning model includes a state detection gate, a forget gate and an activation gate;

[0239] A state detection module is used to perform state detection on target risk indicator data using a state detection gate to obtain the indicator risk state;

[0240] The abnormal indicator filtering module is used to filter the target risk indicator data using the forget gate to obtain normal risk indicator data;

[0241] The risk warning module is used to activate the indicator risk status and the normal data of the risk indicator by using the activation gate to obtain the product value risk warning level of the product risk value sample.

[0242] The specific implementation of the risk warning device is basically the same as the specific implementation of the above-mentioned risk warning method, and will not be repeated here.

[0243] The embodiment of the present application also provides an electronic device, the electronic device includes a memory and a processor, the memory stores a computer program, and the processor implements the above risk warning method when executing the computer program. The electronic device can be any smart terminal including a tablet computer, a car computer, etc.

[0244] See also Fig. 9 , Fig. 9 The hardware structure of an electronic device of another embodiment is illustrated, and the electronic device includes:

[0245] The processor 901 may be implemented by a general-purpose CPU (Central Processing Unit), a microprocessor, an application-specific integrated circuit (ASIC), or one or more integrated circuits, and is used to execute relevant programs to implement the technical solutions provided in the embodiments of the present application;

[0246] The memory 902 can be implemented in the form of a read-only memory (ROM), a static storage device, a dynamic storage device, or a random access memory (RAM). The memory 902 can store an operating system and other application programs. When the technical solution provided in the embodiment of this specification is implemented by software or firmware, the relevant program code is stored in the memory 902, and the processor 901 calls and executes the risk warning method of the embodiment of this application;

[0247] Input / output interface 903, used to implement information input and output;

[0248] Communication interface 904, used to realize communication interaction between the device and other devices, which can be realized by wired mode (such as USB, network cable, etc.) or wireless mode (such as mobile network, WIFI, Bluetooth, etc.);

[0249] A bus 905 that transmits information between various components of the device (e.g., the processor 901, the memory 902, the input / output interface 903, and the communication interface 904);

[0250] The processor 901 , the memory 902 , the input / output interface 903 and the communication interface 904 are connected to each other in communication within the device via a bus 905 .

[0251] An embodiment of the present application also provides a computer-readable storage medium, which stores a computer program, and when the computer program is executed by a processor, the above-mentioned risk warning method is implemented.

[0252] The memory, as a non-transient computer-readable storage medium, can be used to store non-transient software programs and non-transient computer executable programs. In addition, the memory may include a high-speed random access memory, and may also include a non-transient memory, such as at least one disk storage device, a flash memory device, or other non-transient solid-state storage device. In some embodiments, the memory may optionally include a memory remotely disposed relative to the processor, and these remote memories may be connected to the processor via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0253] The risk warning method, risk warning device, electronic device and storage medium provided in the embodiments of the present application obtain initial risk data; wherein the initial risk data is time-series data, and the initial risk data includes product risk indicator data and product risk value data; target risk indicator data is screened out from the product risk indicator data according to the product risk value data; wherein the target risk indicator data is data that generates risks for the product risk value data; a pre-trained risk warning model is obtained; wherein the risk warning model includes a state detection gate, a forget gate and an activation gate; the state detection gate is used to perform state detection on the target risk indicator data to obtain an indicator risk state; the indicator risk state is used to indicate the risk degree of the target risk indicator data; the forget gate is used to perform abnormal indicator filtering processing on the target risk indicator data to obtain normal risk indicator data; the activation gate is used to activate the indicator risk state and the normal risk indicator data to obtain the product value risk warning level of the product risk value sample. This application firstly filters out target risk indicator data from product risk indicator data according to product risk value data, and can remove product risk indicator data irrelevant to product risk value data, which helps to improve the accuracy of subsequent risk warnings; secondly, the state detection gate of the risk warning model is used to perform state detection on the target risk indicator data, which can preliminarily identify the risk level of the risk indicator, and use the forgetting gate to filter out abnormal indicators on the target risk indicator data, so as to further improve the data quality of the risk indicator; finally, the activation gate is used to activate the indicator risk state and the normal data of the risk indicator, which can provide risk warnings for the nonlinear relationship between complex time series in the risk data, thereby improving the accuracy of risk warnings.

[0254] The embodiments described in the embodiments of the present application are intended to more clearly illustrate the technical solutions of the embodiments of the present application and do not constitute a limitation on the technical solutions provided in the embodiments of the present application. Those skilled in the art will appreciate that with the evolution of technology and the emergence of new application scenarios, the technical solutions provided in the embodiments of the present application are also applicable to similar technical problems.

[0255] Those skilled in the art will appreciate that the technical solutions shown in the figures do not constitute a limitation on the embodiments of the present application, and may include more or fewer steps than shown in the figures, or a combination of certain steps, or different steps.

Claims

1. A risk early warning method, characterized in that: The method comprises: Acquire initial risk data; wherein the initial risk data is time-series-based data, and the initial risk data includes product risk indicator data and product risk value data; Filtering target risk indicator data from the product risk indicator data according to the product risk value data; wherein the target risk indicator data is data that generates risk for the product risk value data; Obtaining a pre-trained risk warning model; wherein the risk warning model includes a state detection gate, a forget gate, and an activation gate; Using the state detection gate to perform state detection on the target risk indicator data to obtain an indicator risk state; the indicator risk state is used to indicate the risk level of the target risk indicator data; Using the forget gate to perform abnormal index filtering on the target risk index data to obtain normal risk index data; The activation gate is used to activate the indicator risk status and the normal data of the risk indicator to obtain the product value risk warning level of the product risk value sample.

2. The method according to claim 1, characterized in that The activation gate is used to activate the indicator risk status and the normal data of the risk indicator to obtain the product value risk warning level of the product risk value sample, including: The risk indicator risk status is updated according to the normal data of the risk indicator to obtain an updated risk indicator status; Obtaining an output weight matrix and an output bias term of the activation gate; The update indicator risk status and the risk indicator normal data are activated according to a preset activation function, the output weight matrix and the output bias term to obtain the product value risk warning level of the product risk value sample.

3. The method according to claim 1, characterized in that The step of screening out target risk indicator data from the product risk indicator data according to the product risk value data includes: Obtaining the correlation between the product risk indicator data at a preset moment and the product risk value data at the preset moment, and obtaining the product risk data correlation relationship; the product risk data correlation relationship is used to indicate that: the product risk indicator data generates risk to the product risk value data, the product risk value data generates risk to the product risk indicator data, and the product risk indicator data and the product risk value data generate risk to each other; The target risk indicator data is filtered out from the product risk indicator data according to the product risk data association relationship.

4. The method according to claim 3, characterized in that: The acquiring the association relationship between the product risk indicator data at the preset time and the product risk value data at the preset time to obtain the product risk data association relationship includes: Obtaining the time lag parameter of the product risk indicator data at the preset time; Acquire a first lag item of the product risk indicator data at the preset moment, and a first lag parameter of the first lag item; Acquire a second lag item of the product risk value data at the preset moment, and a second lag parameter of the second lag item; Constructing a first time series model for the time lag parameter, the first lag term, the first lag parameter, the second lag term and the second lag parameter to obtain a first time series model; Constructing a second time series model for the time lag parameter, the first lag term, the first lag parameter, the second lag term and the second lag parameter to obtain a second time series model; The association relationship between the product risk indicator data at the preset moment and the product risk value data at the preset moment is obtained according to the first time series model and the second time series model to obtain a product risk data association relationship.

5. The method according to claim 4, characterized in that After obtaining the correlation relationship between the product risk indicator data at the preset time and the product risk value data at the preset time according to the first time series model and the second time series model, and obtaining the product risk data correlation relationship, the method includes: Obtaining first residual data of the first time series model, and obtaining second residual data of the second time series model; For the first time series model, when the second lag parameter is not zero and the first residual data changes, the product risk data association relationship is a unidirectional risk association relationship; For the second time series model, when the first lag parameter is not zero and the second residual data changes, the product risk data association relationship is a reverse risk association relationship; For the first time series model and the second time series model, when the first lag parameter and the second lag parameter are not zero and the first residual data and the second residual data change, the product risk data association relationship is a bidirectional risk association relationship.

6. The method according to claim 3, characterized in that: The step of acquiring the correlation between the product risk indicator data at a preset time and the product risk value data at the preset time to obtain the product risk data correlation relationship further includes: Obtaining the current time window, and performing time series fluctuation amplitude detection on the product risk value data according to the current time window to obtain the current window time series fluctuation limit; According to the current time window, the correlation relationship between the product risk indicator data at the preset time and the product risk value data at the preset time is obtained within the time series fluctuation limit of the current window to obtain the current window correlation relationship; Obtaining an adjacent forward time window of the current time window, and performing a time series fluctuation amplitude detection on the product risk value data according to the adjacent forward time window to obtain a forward window time series fluctuation limit; According to the adjacent forward time windows, the correlation relationship between the product risk indicator data at the preset time and the product risk value data at the preset time is obtained within the forward window time series fluctuation limit to obtain a forward window correlation relationship; Obtaining an adjacent backward time window of the current time window, and performing a time series fluctuation amplitude detection on the product risk value data according to the adjacent backward time window to obtain a backward window time series fluctuation limit; According to the adjacent backward time windows, the correlation relationship between the product risk indicator data at the preset time and the product risk value data at the preset time is obtained within the backward window time series fluctuation limit to obtain a backward window correlation relationship; The current window association relationship, the forward window association relationship and the backward window association relationship are merged to obtain the product risk data association relationship.

7. The method according to any one of claims 1 to 4, characterized in that: After obtaining the initial risk data, the method further includes: Normalizing the initial risk data to obtain normalized risk data; Performing a time trend test on the normalized risk data to obtain stable time trend risk data; Performing a unit root test on the normalized risk data to obtain stable unit root risk data; The stationary time trend risk data and the stationary unit root risk data are fused to obtain fused test risk data, and the fused test risk data is used as the initial risk data.

8. A risk warning device, characterized in that: The device comprises: A risk data acquisition module, used to acquire initial risk data; wherein the initial risk data is time-series-based data, and the initial risk data includes product risk indicator data and product risk value data; An indicator data screening module, used for screening target risk indicator data from the product risk indicator data according to the product risk value data; wherein the target risk indicator data is data that generates risk to the product risk value data; The early warning model acquisition module is used to acquire a pre-trained risk early warning model; wherein the risk early warning model includes a state detection gate, a forget gate and an activation gate; A state detection module, used to perform state detection on the target risk indicator data using the state detection gate to obtain an indicator risk state; An abnormal indicator filtering module is used to use the forget gate to perform abnormal indicator filtering on the target risk indicator data to obtain normal risk indicator data; The risk warning module is used to use the activation gate to activate the indicator risk status and the normal data of the risk indicator to obtain the product value risk warning level of the product risk value sample.

9. An electronic device, characterized in that: The electronic device includes a memory and a processor, the memory stores a computer program, and the processor implements the risk warning method according to any one of claims 1 to 7 when executing the computer program.

10. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the risk warning method according to any one of claims 1 to 7 is implemented.

Images